spx_usrreq.c revision 76166
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: head/sys/netipx/spx_usrreq.c 76166 2001-05-01 08:13:21Z markm $
37 */
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/lock.h>
42#include <sys/malloc.h>
43#include <sys/mbuf.h>
44#include <sys/mutex.h>
45#include <sys/proc.h>
46#include <sys/protosw.h>
47#include <sys/socket.h>
48#include <sys/socketvar.h>
49
50#include <net/route.h>
51#include <netinet/tcp_fsm.h>
52
53#include <netipx/ipx.h>
54#include <netipx/ipx_pcb.h>
55#include <netipx/ipx_var.h>
56#include <netipx/spx.h>
57#include <netipx/spx_timer.h>
58#include <netipx/spx_var.h>
59#include <netipx/spx_debug.h>
60
61/*
62 * SPX protocol implementation.
63 */
64static u_short 	spx_iss;
65static u_short	spx_newchecks[50];
66static int	spx_hardnosed;
67static int	spx_use_delack = 0;
68static int	traceallspxs = 0;
69static struct	spx 	spx_savesi;
70static struct	spx_istat spx_istat;
71
72/* Following was struct spxstat spxstat; */
73#ifndef spxstat
74#define spxstat spx_istat.newstats
75#endif
76
77static int spx_backoff[SPX_MAXRXTSHIFT+1] =
78    { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
79
80static	struct spxpcb *spx_close(struct spxpcb *cb);
81static	struct spxpcb *spx_disconnect(struct spxpcb *cb);
82static	struct spxpcb *spx_drop(struct spxpcb *cb, int errno);
83static	int spx_output(struct spxpcb *cb, struct mbuf *m0);
84static	int spx_reass(struct spxpcb *cb, struct spx *si);
85static	void spx_setpersist(struct spxpcb *cb);
86static	void spx_template(struct spxpcb *cb);
87static	struct spxpcb *spx_timers(struct spxpcb *cb, int timer);
88static	struct spxpcb *spx_usrclosed(struct spxpcb *cb);
89
90static	int spx_usr_abort(struct socket *so);
91static	int spx_accept(struct socket *so, struct sockaddr **nam);
92static	int spx_attach(struct socket *so, int proto, struct proc *p);
93static	int spx_bind(struct socket *so, struct sockaddr *nam, struct proc *p);
94static	int spx_connect(struct socket *so, struct sockaddr *nam,
95			struct proc *p);
96static	int spx_detach(struct socket *so);
97static	int spx_usr_disconnect(struct socket *so);
98static	int spx_listen(struct socket *so, struct proc *p);
99static	int spx_rcvd(struct socket *so, int flags);
100static	int spx_rcvoob(struct socket *so, struct mbuf *m, int flags);
101static	int spx_send(struct socket *so, int flags, struct mbuf *m,
102		     struct sockaddr *addr, struct mbuf *control,
103		     struct proc *p);
104static	int spx_shutdown(struct socket *so);
105static	int spx_sp_attach(struct socket *so, int proto, struct proc *p);
106
107struct	pr_usrreqs spx_usrreqs = {
108	spx_usr_abort, spx_accept, spx_attach, spx_bind,
109	spx_connect, pru_connect2_notsupp, ipx_control, spx_detach,
110	spx_usr_disconnect, spx_listen, ipx_peeraddr, spx_rcvd,
111	spx_rcvoob, spx_send, pru_sense_null, spx_shutdown,
112	ipx_sockaddr, sosend, soreceive, sopoll
113};
114
115struct	pr_usrreqs spx_usrreq_sps = {
116	spx_usr_abort, spx_accept, spx_sp_attach, spx_bind,
117	spx_connect, pru_connect2_notsupp, ipx_control, spx_detach,
118	spx_usr_disconnect, spx_listen, ipx_peeraddr, spx_rcvd,
119	spx_rcvoob, spx_send, pru_sense_null, spx_shutdown,
120	ipx_sockaddr, sosend, soreceive, sopoll
121};
122
123void
124spx_init()
125{
126
127	spx_iss = 1; /* WRONG !! should fish it out of TODR */
128}
129
130void
131spx_input(m, ipxp)
132	register struct mbuf *m;
133	register struct ipxpcb *ipxp;
134{
135	register struct spxpcb *cb;
136	register struct spx *si = mtod(m, struct spx *);
137	register struct socket *so;
138	int dropsocket = 0;
139	short ostate = 0;
140
141	spxstat.spxs_rcvtotal++;
142	if (ipxp == NULL) {
143		panic("No ipxpcb in spx_input\n");
144		return;
145	}
146
147	cb = ipxtospxpcb(ipxp);
148	if (cb == NULL)
149		goto bad;
150
151	if (m->m_len < sizeof(*si)) {
152		if ((m = m_pullup(m, sizeof(*si))) == NULL) {
153			spxstat.spxs_rcvshort++;
154			return;
155		}
156		si = mtod(m, struct spx *);
157	}
158	si->si_seq = ntohs(si->si_seq);
159	si->si_ack = ntohs(si->si_ack);
160	si->si_alo = ntohs(si->si_alo);
161
162	so = ipxp->ipxp_socket;
163
164	if (so->so_options & SO_DEBUG || traceallspxs) {
165		ostate = cb->s_state;
166		spx_savesi = *si;
167	}
168	if (so->so_options & SO_ACCEPTCONN) {
169		struct spxpcb *ocb = cb;
170
171		so = sonewconn(so, 0);
172		if (so == NULL) {
173			goto drop;
174		}
175		/*
176		 * This is ugly, but ....
177		 *
178		 * Mark socket as temporary until we're
179		 * committed to keeping it.  The code at
180		 * ``drop'' and ``dropwithreset'' check the
181		 * flag dropsocket to see if the temporary
182		 * socket created here should be discarded.
183		 * We mark the socket as discardable until
184		 * we're committed to it below in TCPS_LISTEN.
185		 */
186		dropsocket++;
187		ipxp = (struct ipxpcb *)so->so_pcb;
188		ipxp->ipxp_laddr = si->si_dna;
189		cb = ipxtospxpcb(ipxp);
190		cb->s_mtu = ocb->s_mtu;		/* preserve sockopts */
191		cb->s_flags = ocb->s_flags;	/* preserve sockopts */
192		cb->s_flags2 = ocb->s_flags2;	/* preserve sockopts */
193		cb->s_state = TCPS_LISTEN;
194	}
195
196	/*
197	 * Packet received on connection.
198	 * reset idle time and keep-alive timer;
199	 */
200	cb->s_idle = 0;
201	cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
202
203	switch (cb->s_state) {
204
205	case TCPS_LISTEN:{
206		struct sockaddr_ipx *sipx, ssipx;
207		struct ipx_addr laddr;
208
209		/*
210		 * If somebody here was carying on a conversation
211		 * and went away, and his pen pal thinks he can
212		 * still talk, we get the misdirected packet.
213		 */
214		if (spx_hardnosed && (si->si_did != 0 || si->si_seq != 0)) {
215			spx_istat.gonawy++;
216			goto dropwithreset;
217		}
218		sipx = &ssipx;
219		bzero(sipx, sizeof *sipx);
220		sipx->sipx_len = sizeof(*sipx);
221		sipx->sipx_family = AF_IPX;
222		sipx->sipx_addr = si->si_sna;
223		laddr = ipxp->ipxp_laddr;
224		if (ipx_nullhost(laddr))
225			ipxp->ipxp_laddr = si->si_dna;
226		if (ipx_pcbconnect(ipxp, (struct sockaddr *)sipx, &proc0)) {
227			ipxp->ipxp_laddr = laddr;
228			spx_istat.noconn++;
229			goto drop;
230		}
231		spx_template(cb);
232		dropsocket = 0;		/* committed to socket */
233		cb->s_did = si->si_sid;
234		cb->s_rack = si->si_ack;
235		cb->s_ralo = si->si_alo;
236#define THREEWAYSHAKE
237#ifdef THREEWAYSHAKE
238		cb->s_state = TCPS_SYN_RECEIVED;
239		cb->s_force = 1 + SPXT_KEEP;
240		spxstat.spxs_accepts++;
241		cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
242		}
243		break;
244	/*
245	 * This state means that we have heard a response
246	 * to our acceptance of their connection
247	 * It is probably logically unnecessary in this
248	 * implementation.
249	 */
250	 case TCPS_SYN_RECEIVED: {
251		if (si->si_did != cb->s_sid) {
252			spx_istat.wrncon++;
253			goto drop;
254		}
255#endif
256		ipxp->ipxp_fport =  si->si_sport;
257		cb->s_timer[SPXT_REXMT] = 0;
258		cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
259		soisconnected(so);
260		cb->s_state = TCPS_ESTABLISHED;
261		spxstat.spxs_accepts++;
262		}
263		break;
264
265	/*
266	 * This state means that we have gotten a response
267	 * to our attempt to establish a connection.
268	 * We fill in the data from the other side,
269	 * telling us which port to respond to, instead of the well-
270	 * known one we might have sent to in the first place.
271	 * We also require that this is a response to our
272	 * connection id.
273	 */
274	case TCPS_SYN_SENT:
275		if (si->si_did != cb->s_sid) {
276			spx_istat.notme++;
277			goto drop;
278		}
279		spxstat.spxs_connects++;
280		cb->s_did = si->si_sid;
281		cb->s_rack = si->si_ack;
282		cb->s_ralo = si->si_alo;
283		cb->s_dport = ipxp->ipxp_fport =  si->si_sport;
284		cb->s_timer[SPXT_REXMT] = 0;
285		cb->s_flags |= SF_ACKNOW;
286		soisconnected(so);
287		cb->s_state = TCPS_ESTABLISHED;
288		/* Use roundtrip time of connection request for initial rtt */
289		if (cb->s_rtt) {
290			cb->s_srtt = cb->s_rtt << 3;
291			cb->s_rttvar = cb->s_rtt << 1;
292			SPXT_RANGESET(cb->s_rxtcur,
293			    ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
294			    SPXTV_MIN, SPXTV_REXMTMAX);
295			    cb->s_rtt = 0;
296		}
297	}
298	if (so->so_options & SO_DEBUG || traceallspxs)
299		spx_trace(SA_INPUT, (u_char)ostate, cb, &spx_savesi, 0);
300
301	m->m_len -= sizeof(struct ipx);
302	m->m_pkthdr.len -= sizeof(struct ipx);
303	m->m_data += sizeof(struct ipx);
304
305	if (spx_reass(cb, si)) {
306		m_freem(m);
307	}
308	if (cb->s_force || (cb->s_flags & (SF_ACKNOW|SF_WIN|SF_RXT)))
309		spx_output(cb, (struct mbuf *)NULL);
310	cb->s_flags &= ~(SF_WIN|SF_RXT);
311	return;
312
313dropwithreset:
314	if (dropsocket)
315		soabort(so);
316	si->si_seq = ntohs(si->si_seq);
317	si->si_ack = ntohs(si->si_ack);
318	si->si_alo = ntohs(si->si_alo);
319	m_freem(dtom(si));
320	if (cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG || traceallspxs)
321		spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
322	return;
323
324drop:
325bad:
326	if (cb == 0 || cb->s_ipxpcb->ipxp_socket->so_options & SO_DEBUG ||
327            traceallspxs)
328		spx_trace(SA_DROP, (u_char)ostate, cb, &spx_savesi, 0);
329	m_freem(m);
330}
331
332static int spxrexmtthresh = 3;
333
334/*
335 * This is structurally similar to the tcp reassembly routine
336 * but its function is somewhat different:  It merely queues
337 * packets up, and suppresses duplicates.
338 */
339static int
340spx_reass(cb, si)
341register struct spxpcb *cb;
342register struct spx *si;
343{
344	register struct spx_q *q;
345	register struct mbuf *m;
346	register struct socket *so = cb->s_ipxpcb->ipxp_socket;
347	char packetp = cb->s_flags & SF_HI;
348	int incr;
349	char wakeup = 0;
350
351	if (si == SI(0))
352		goto present;
353	/*
354	 * Update our news from them.
355	 */
356	if (si->si_cc & SPX_SA)
357		cb->s_flags |= (spx_use_delack ? SF_DELACK : SF_ACKNOW);
358	if (SSEQ_GT(si->si_alo, cb->s_ralo))
359		cb->s_flags |= SF_WIN;
360	if (SSEQ_LEQ(si->si_ack, cb->s_rack)) {
361		if ((si->si_cc & SPX_SP) && cb->s_rack != (cb->s_smax + 1)) {
362			spxstat.spxs_rcvdupack++;
363			/*
364			 * If this is a completely duplicate ack
365			 * and other conditions hold, we assume
366			 * a packet has been dropped and retransmit
367			 * it exactly as in tcp_input().
368			 */
369			if (si->si_ack != cb->s_rack ||
370			    si->si_alo != cb->s_ralo)
371				cb->s_dupacks = 0;
372			else if (++cb->s_dupacks == spxrexmtthresh) {
373				u_short onxt = cb->s_snxt;
374				int cwnd = cb->s_cwnd;
375
376				cb->s_snxt = si->si_ack;
377				cb->s_cwnd = CUNIT;
378				cb->s_force = 1 + SPXT_REXMT;
379				spx_output(cb, (struct mbuf *)NULL);
380				cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
381				cb->s_rtt = 0;
382				if (cwnd >= 4 * CUNIT)
383					cb->s_cwnd = cwnd / 2;
384				if (SSEQ_GT(onxt, cb->s_snxt))
385					cb->s_snxt = onxt;
386				return (1);
387			}
388		} else
389			cb->s_dupacks = 0;
390		goto update_window;
391	}
392	cb->s_dupacks = 0;
393	/*
394	 * If our correspondent acknowledges data we haven't sent
395	 * TCP would drop the packet after acking.  We'll be a little
396	 * more permissive
397	 */
398	if (SSEQ_GT(si->si_ack, (cb->s_smax + 1))) {
399		spxstat.spxs_rcvacktoomuch++;
400		si->si_ack = cb->s_smax + 1;
401	}
402	spxstat.spxs_rcvackpack++;
403	/*
404	 * If transmit timer is running and timed sequence
405	 * number was acked, update smoothed round trip time.
406	 * See discussion of algorithm in tcp_input.c
407	 */
408	if (cb->s_rtt && SSEQ_GT(si->si_ack, cb->s_rtseq)) {
409		spxstat.spxs_rttupdated++;
410		if (cb->s_srtt != 0) {
411			register short delta;
412			delta = cb->s_rtt - (cb->s_srtt >> 3);
413			if ((cb->s_srtt += delta) <= 0)
414				cb->s_srtt = 1;
415			if (delta < 0)
416				delta = -delta;
417			delta -= (cb->s_rttvar >> 2);
418			if ((cb->s_rttvar += delta) <= 0)
419				cb->s_rttvar = 1;
420		} else {
421			/*
422			 * No rtt measurement yet
423			 */
424			cb->s_srtt = cb->s_rtt << 3;
425			cb->s_rttvar = cb->s_rtt << 1;
426		}
427		cb->s_rtt = 0;
428		cb->s_rxtshift = 0;
429		SPXT_RANGESET(cb->s_rxtcur,
430			((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
431			SPXTV_MIN, SPXTV_REXMTMAX);
432	}
433	/*
434	 * If all outstanding data is acked, stop retransmit
435	 * timer and remember to restart (more output or persist).
436	 * If there is more data to be acked, restart retransmit
437	 * timer, using current (possibly backed-off) value;
438	 */
439	if (si->si_ack == cb->s_smax + 1) {
440		cb->s_timer[SPXT_REXMT] = 0;
441		cb->s_flags |= SF_RXT;
442	} else if (cb->s_timer[SPXT_PERSIST] == 0)
443		cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
444	/*
445	 * When new data is acked, open the congestion window.
446	 * If the window gives us less than ssthresh packets
447	 * in flight, open exponentially (maxseg at a time).
448	 * Otherwise open linearly (maxseg^2 / cwnd at a time).
449	 */
450	incr = CUNIT;
451	if (cb->s_cwnd > cb->s_ssthresh)
452		incr = max(incr * incr / cb->s_cwnd, 1);
453	cb->s_cwnd = min(cb->s_cwnd + incr, cb->s_cwmx);
454	/*
455	 * Trim Acked data from output queue.
456	 */
457	while ((m = so->so_snd.sb_mb) != NULL) {
458		if (SSEQ_LT((mtod(m, struct spx *))->si_seq, si->si_ack))
459			sbdroprecord(&so->so_snd);
460		else
461			break;
462	}
463	sowwakeup(so);
464	cb->s_rack = si->si_ack;
465update_window:
466	if (SSEQ_LT(cb->s_snxt, cb->s_rack))
467		cb->s_snxt = cb->s_rack;
468	if (SSEQ_LT(cb->s_swl1, si->si_seq) || ((cb->s_swl1 == si->si_seq &&
469	    (SSEQ_LT(cb->s_swl2, si->si_ack))) ||
470	     (cb->s_swl2 == si->si_ack && SSEQ_LT(cb->s_ralo, si->si_alo)))) {
471		/* keep track of pure window updates */
472		if ((si->si_cc & SPX_SP) && cb->s_swl2 == si->si_ack
473		    && SSEQ_LT(cb->s_ralo, si->si_alo)) {
474			spxstat.spxs_rcvwinupd++;
475			spxstat.spxs_rcvdupack--;
476		}
477		cb->s_ralo = si->si_alo;
478		cb->s_swl1 = si->si_seq;
479		cb->s_swl2 = si->si_ack;
480		cb->s_swnd = (1 + si->si_alo - si->si_ack);
481		if (cb->s_swnd > cb->s_smxw)
482			cb->s_smxw = cb->s_swnd;
483		cb->s_flags |= SF_WIN;
484	}
485	/*
486	 * If this packet number is higher than that which
487	 * we have allocated refuse it, unless urgent
488	 */
489	if (SSEQ_GT(si->si_seq, cb->s_alo)) {
490		if (si->si_cc & SPX_SP) {
491			spxstat.spxs_rcvwinprobe++;
492			return (1);
493		} else
494			spxstat.spxs_rcvpackafterwin++;
495		if (si->si_cc & SPX_OB) {
496			if (SSEQ_GT(si->si_seq, cb->s_alo + 60)) {
497				m_freem(dtom(si));
498				return (0);
499			} /* else queue this packet; */
500		} else {
501			/*register struct socket *so = cb->s_ipxpcb->ipxp_socket;
502			if (so->so_state && SS_NOFDREF) {
503				spx_close(cb);
504			} else
505				       would crash system*/
506			spx_istat.notyet++;
507			m_freem(dtom(si));
508			return (0);
509		}
510	}
511	/*
512	 * If this is a system packet, we don't need to
513	 * queue it up, and won't update acknowledge #
514	 */
515	if (si->si_cc & SPX_SP) {
516		return (1);
517	}
518	/*
519	 * We have already seen this packet, so drop.
520	 */
521	if (SSEQ_LT(si->si_seq, cb->s_ack)) {
522		spx_istat.bdreas++;
523		spxstat.spxs_rcvduppack++;
524		if (si->si_seq == cb->s_ack - 1)
525			spx_istat.lstdup++;
526		return (1);
527	}
528	/*
529	 * Loop through all packets queued up to insert in
530	 * appropriate sequence.
531	 */
532	for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
533		if (si->si_seq == SI(q)->si_seq) {
534			spxstat.spxs_rcvduppack++;
535			return (1);
536		}
537		if (SSEQ_LT(si->si_seq, SI(q)->si_seq)) {
538			spxstat.spxs_rcvoopack++;
539			break;
540		}
541	}
542	insque(si, q->si_prev);
543	/*
544	 * If this packet is urgent, inform process
545	 */
546	if (si->si_cc & SPX_OB) {
547		cb->s_iobc = ((char *)si)[1 + sizeof(*si)];
548		sohasoutofband(so);
549		cb->s_oobflags |= SF_IOOB;
550	}
551present:
552#define SPINC sizeof(struct spxhdr)
553	/*
554	 * Loop through all packets queued up to update acknowledge
555	 * number, and present all acknowledged data to user;
556	 * If in packet interface mode, show packet headers.
557	 */
558	for (q = cb->s_q.si_next; q != &cb->s_q; q = q->si_next) {
559		  if (SI(q)->si_seq == cb->s_ack) {
560			cb->s_ack++;
561			m = dtom(q);
562			if (SI(q)->si_cc & SPX_OB) {
563				cb->s_oobflags &= ~SF_IOOB;
564				if (so->so_rcv.sb_cc)
565					so->so_oobmark = so->so_rcv.sb_cc;
566				else
567					so->so_state |= SS_RCVATMARK;
568			}
569			q = q->si_prev;
570			remque(q->si_next);
571			wakeup = 1;
572			spxstat.spxs_rcvpack++;
573#ifdef SF_NEWCALL
574			if (cb->s_flags2 & SF_NEWCALL) {
575				struct spxhdr *sp = mtod(m, struct spxhdr *);
576				u_char dt = sp->spx_dt;
577				spx_newchecks[4]++;
578				if (dt != cb->s_rhdr.spx_dt) {
579					struct mbuf *mm =
580					   m_getclr(M_DONTWAIT, MT_CONTROL);
581					spx_newchecks[0]++;
582					if (mm != NULL) {
583						u_short *s =
584							mtod(mm, u_short *);
585						cb->s_rhdr.spx_dt = dt;
586						mm->m_len = 5; /*XXX*/
587						s[0] = 5;
588						s[1] = 1;
589						*(u_char *)(&s[2]) = dt;
590						sbappend(&so->so_rcv, mm);
591					}
592				}
593				if (sp->spx_cc & SPX_OB) {
594					MCHTYPE(m, MT_OOBDATA);
595					spx_newchecks[1]++;
596					so->so_oobmark = 0;
597					so->so_state &= ~SS_RCVATMARK;
598				}
599				if (packetp == 0) {
600					m->m_data += SPINC;
601					m->m_len -= SPINC;
602					m->m_pkthdr.len -= SPINC;
603				}
604				if ((sp->spx_cc & SPX_EM) || packetp) {
605					sbappendrecord(&so->so_rcv, m);
606					spx_newchecks[9]++;
607				} else
608					sbappend(&so->so_rcv, m);
609			} else
610#endif
611			if (packetp) {
612				sbappendrecord(&so->so_rcv, m);
613			} else {
614				cb->s_rhdr = *mtod(m, struct spxhdr *);
615				m->m_data += SPINC;
616				m->m_len -= SPINC;
617				m->m_pkthdr.len -= SPINC;
618				sbappend(&so->so_rcv, m);
619			}
620		  } else
621			break;
622	}
623	if (wakeup)
624		sorwakeup(so);
625	return (0);
626}
627
628void
629spx_ctlinput(cmd, arg_as_sa, dummy)
630	int cmd;
631	struct sockaddr *arg_as_sa;	/* XXX should be swapped with dummy */
632	void *dummy;
633{
634	caddr_t arg = (/* XXX */ caddr_t)arg_as_sa;
635	struct ipx_addr *na;
636	struct sockaddr_ipx *sipx;
637
638	if (cmd < 0 || cmd > PRC_NCMDS)
639		return;
640
641	switch (cmd) {
642
643	case PRC_ROUTEDEAD:
644		return;
645
646	case PRC_IFDOWN:
647	case PRC_HOSTDEAD:
648	case PRC_HOSTUNREACH:
649		sipx = (struct sockaddr_ipx *)arg;
650		if (sipx->sipx_family != AF_IPX)
651			return;
652		na = &sipx->sipx_addr;
653		break;
654
655	default:
656		break;
657	}
658}
659
660#ifdef notdef
661int
662spx_fixmtu(ipxp)
663register struct ipxpcb *ipxp;
664{
665	register struct spxpcb *cb = (struct spxpcb *)(ipxp->ipxp_pcb);
666	register struct mbuf *m;
667	register struct spx *si;
668	struct ipx_errp *ep;
669	struct sockbuf *sb;
670	int badseq, len;
671	struct mbuf *firstbad, *m0;
672
673	if (cb != NULL) {
674		/*
675		 * The notification that we have sent
676		 * too much is bad news -- we will
677		 * have to go through queued up so far
678		 * splitting ones which are too big and
679		 * reassigning sequence numbers and checksums.
680		 * we should then retransmit all packets from
681		 * one above the offending packet to the last one
682		 * we had sent (or our allocation)
683		 * then the offending one so that the any queued
684		 * data at our destination will be discarded.
685		 */
686		 ep = (struct ipx_errp *)ipxp->ipxp_notify_param;
687		 sb = &ipxp->ipxp_socket->so_snd;
688		 cb->s_mtu = ep->ipx_err_param;
689		 badseq = SI(&ep->ipx_err_ipx)->si_seq;
690		 for (m = sb->sb_mb; m != NULL; m = m->m_act) {
691			si = mtod(m, struct spx *);
692			if (si->si_seq == badseq)
693				break;
694		 }
695		 if (m == NULL)
696			return;
697		 firstbad = m;
698		 /*for (;;) {*/
699			/* calculate length */
700			for (m0 = m, len = 0; m != NULL; m = m->m_next)
701				len += m->m_len;
702			if (len > cb->s_mtu) {
703			}
704		/* FINISH THIS
705		} */
706	}
707}
708#endif
709
710static int
711spx_output(cb, m0)
712	register struct spxpcb *cb;
713	struct mbuf *m0;
714{
715	struct socket *so = cb->s_ipxpcb->ipxp_socket;
716	register struct mbuf *m;
717	register struct spx *si = (struct spx *)NULL;
718	register struct sockbuf *sb = &so->so_snd;
719	int len = 0, win, rcv_win;
720	short span, off, recordp = 0;
721	u_short alo;
722	int error = 0, sendalot;
723#ifdef notdef
724	int idle;
725#endif
726	struct mbuf *mprev;
727
728	if (m0 != NULL) {
729		int mtu = cb->s_mtu;
730		int datalen;
731		/*
732		 * Make sure that packet isn't too big.
733		 */
734		for (m = m0; m != NULL; m = m->m_next) {
735			mprev = m;
736			len += m->m_len;
737			if (m->m_flags & M_EOR)
738				recordp = 1;
739		}
740		datalen = (cb->s_flags & SF_HO) ?
741				len - sizeof(struct spxhdr) : len;
742		if (datalen > mtu) {
743			if (cb->s_flags & SF_PI) {
744				m_freem(m0);
745				return (EMSGSIZE);
746			} else {
747				int oldEM = cb->s_cc & SPX_EM;
748
749				cb->s_cc &= ~SPX_EM;
750				while (len > mtu) {
751					/*
752					 * Here we are only being called
753					 * from usrreq(), so it is OK to
754					 * block.
755					 */
756					m = m_copym(m0, 0, mtu, M_TRYWAIT);
757					if (cb->s_flags & SF_NEWCALL) {
758					    struct mbuf *mm = m;
759					    spx_newchecks[7]++;
760					    while (mm != NULL) {
761						mm->m_flags &= ~M_EOR;
762						mm = mm->m_next;
763					    }
764					}
765					error = spx_output(cb, m);
766					if (error) {
767						cb->s_cc |= oldEM;
768						m_freem(m0);
769						return (error);
770					}
771					m_adj(m0, mtu);
772					len -= mtu;
773				}
774				cb->s_cc |= oldEM;
775			}
776		}
777		/*
778		 * Force length even, by adding a "garbage byte" if
779		 * necessary.
780		 */
781		if (len & 1) {
782			m = mprev;
783			if (M_TRAILINGSPACE(m) >= 1)
784				m->m_len++;
785			else {
786				struct mbuf *m1 = m_get(M_DONTWAIT, MT_DATA);
787
788				if (m1 == NULL) {
789					m_freem(m0);
790					return (ENOBUFS);
791				}
792				m1->m_len = 1;
793				*(mtod(m1, u_char *)) = 0;
794				m->m_next = m1;
795			}
796		}
797		m = m_gethdr(M_DONTWAIT, MT_HEADER);
798		if (m == NULL) {
799			m_freem(m0);
800			return (ENOBUFS);
801		}
802		/*
803		 * Fill in mbuf with extended SP header
804		 * and addresses and length put into network format.
805		 */
806		MH_ALIGN(m, sizeof(struct spx));
807		m->m_len = sizeof(struct spx);
808		m->m_next = m0;
809		si = mtod(m, struct spx *);
810		si->si_i = *cb->s_ipx;
811		si->si_s = cb->s_shdr;
812		if ((cb->s_flags & SF_PI) && (cb->s_flags & SF_HO)) {
813			register struct spxhdr *sh;
814			if (m0->m_len < sizeof(*sh)) {
815				if((m0 = m_pullup(m0, sizeof(*sh))) == NULL) {
816					m_free(m);
817					m_freem(m0);
818					return (EINVAL);
819				}
820				m->m_next = m0;
821			}
822			sh = mtod(m0, struct spxhdr *);
823			si->si_dt = sh->spx_dt;
824			si->si_cc |= sh->spx_cc & SPX_EM;
825			m0->m_len -= sizeof(*sh);
826			m0->m_data += sizeof(*sh);
827			len -= sizeof(*sh);
828		}
829		len += sizeof(*si);
830		if ((cb->s_flags2 & SF_NEWCALL) && recordp) {
831			si->si_cc |= SPX_EM;
832			spx_newchecks[8]++;
833		}
834		if (cb->s_oobflags & SF_SOOB) {
835			/*
836			 * Per jqj@cornell:
837			 * make sure OB packets convey exactly 1 byte.
838			 * If the packet is 1 byte or larger, we
839			 * have already guaranted there to be at least
840			 * one garbage byte for the checksum, and
841			 * extra bytes shouldn't hurt!
842			 */
843			if (len > sizeof(*si)) {
844				si->si_cc |= SPX_OB;
845				len = (1 + sizeof(*si));
846			}
847		}
848		si->si_len = htons((u_short)len);
849		m->m_pkthdr.len = ((len - 1) | 1) + 1;
850		/*
851		 * queue stuff up for output
852		 */
853		sbappendrecord(sb, m);
854		cb->s_seq++;
855	}
856#ifdef notdef
857	idle = (cb->s_smax == (cb->s_rack - 1));
858#endif
859again:
860	sendalot = 0;
861	off = cb->s_snxt - cb->s_rack;
862	win = min(cb->s_swnd, (cb->s_cwnd / CUNIT));
863
864	/*
865	 * If in persist timeout with window of 0, send a probe.
866	 * Otherwise, if window is small but nonzero
867	 * and timer expired, send what we can and go into
868	 * transmit state.
869	 */
870	if (cb->s_force == 1 + SPXT_PERSIST) {
871		if (win != 0) {
872			cb->s_timer[SPXT_PERSIST] = 0;
873			cb->s_rxtshift = 0;
874		}
875	}
876	span = cb->s_seq - cb->s_rack;
877	len = min(span, win) - off;
878
879	if (len < 0) {
880		/*
881		 * Window shrank after we went into it.
882		 * If window shrank to 0, cancel pending
883		 * restransmission and pull s_snxt back
884		 * to (closed) window.  We will enter persist
885		 * state below.  If the widndow didn't close completely,
886		 * just wait for an ACK.
887		 */
888		len = 0;
889		if (win == 0) {
890			cb->s_timer[SPXT_REXMT] = 0;
891			cb->s_snxt = cb->s_rack;
892		}
893	}
894	if (len > 1)
895		sendalot = 1;
896	rcv_win = sbspace(&so->so_rcv);
897
898	/*
899	 * Send if we owe peer an ACK.
900	 */
901	if (cb->s_oobflags & SF_SOOB) {
902		/*
903		 * must transmit this out of band packet
904		 */
905		cb->s_oobflags &= ~ SF_SOOB;
906		sendalot = 1;
907		spxstat.spxs_sndurg++;
908		goto found;
909	}
910	if (cb->s_flags & SF_ACKNOW)
911		goto send;
912	if (cb->s_state < TCPS_ESTABLISHED)
913		goto send;
914	/*
915	 * Silly window can't happen in spx.
916	 * Code from tcp deleted.
917	 */
918	if (len)
919		goto send;
920	/*
921	 * Compare available window to amount of window
922	 * known to peer (as advertised window less
923	 * next expected input.)  If the difference is at least two
924	 * packets or at least 35% of the mximum possible window,
925	 * then want to send a window update to peer.
926	 */
927	if (rcv_win > 0) {
928		u_short delta =  1 + cb->s_alo - cb->s_ack;
929		int adv = rcv_win - (delta * cb->s_mtu);
930
931		if ((so->so_rcv.sb_cc == 0 && adv >= (2 * cb->s_mtu)) ||
932		    (100 * adv / so->so_rcv.sb_hiwat >= 35)) {
933			spxstat.spxs_sndwinup++;
934			cb->s_flags |= SF_ACKNOW;
935			goto send;
936		}
937
938	}
939	/*
940	 * Many comments from tcp_output.c are appropriate here
941	 * including . . .
942	 * If send window is too small, there is data to transmit, and no
943	 * retransmit or persist is pending, then go to persist state.
944	 * If nothing happens soon, send when timer expires:
945	 * if window is nonzero, transmit what we can,
946	 * otherwise send a probe.
947	 */
948	if (so->so_snd.sb_cc && cb->s_timer[SPXT_REXMT] == 0 &&
949		cb->s_timer[SPXT_PERSIST] == 0) {
950			cb->s_rxtshift = 0;
951			spx_setpersist(cb);
952	}
953	/*
954	 * No reason to send a packet, just return.
955	 */
956	cb->s_outx = 1;
957	return (0);
958
959send:
960	/*
961	 * Find requested packet.
962	 */
963	si = 0;
964	if (len > 0) {
965		cb->s_want = cb->s_snxt;
966		for (m = sb->sb_mb; m != NULL; m = m->m_act) {
967			si = mtod(m, struct spx *);
968			if (SSEQ_LEQ(cb->s_snxt, si->si_seq))
969				break;
970		}
971	found:
972		if (si != NULL) {
973			if (si->si_seq == cb->s_snxt)
974					cb->s_snxt++;
975				else
976					spxstat.spxs_sndvoid++, si = 0;
977		}
978	}
979	/*
980	 * update window
981	 */
982	if (rcv_win < 0)
983		rcv_win = 0;
984	alo = cb->s_ack - 1 + (rcv_win / ((short)cb->s_mtu));
985	if (SSEQ_LT(alo, cb->s_alo))
986		alo = cb->s_alo;
987
988	if (si != NULL) {
989		/*
990		 * must make a copy of this packet for
991		 * ipx_output to monkey with
992		 */
993		m = m_copy(dtom(si), 0, (int)M_COPYALL);
994		if (m == NULL) {
995			return (ENOBUFS);
996		}
997		si = mtod(m, struct spx *);
998		if (SSEQ_LT(si->si_seq, cb->s_smax))
999			spxstat.spxs_sndrexmitpack++;
1000		else
1001			spxstat.spxs_sndpack++;
1002	} else if (cb->s_force || cb->s_flags & SF_ACKNOW) {
1003		/*
1004		 * Must send an acknowledgement or a probe
1005		 */
1006		if (cb->s_force)
1007			spxstat.spxs_sndprobe++;
1008		if (cb->s_flags & SF_ACKNOW)
1009			spxstat.spxs_sndacks++;
1010		m = m_gethdr(M_DONTWAIT, MT_HEADER);
1011		if (m == NULL)
1012			return (ENOBUFS);
1013		/*
1014		 * Fill in mbuf with extended SP header
1015		 * and addresses and length put into network format.
1016		 */
1017		MH_ALIGN(m, sizeof(struct spx));
1018		m->m_len = sizeof(*si);
1019		m->m_pkthdr.len = sizeof(*si);
1020		si = mtod(m, struct spx *);
1021		si->si_i = *cb->s_ipx;
1022		si->si_s = cb->s_shdr;
1023		si->si_seq = cb->s_smax + 1;
1024		si->si_len = htons(sizeof(*si));
1025		si->si_cc |= SPX_SP;
1026	} else {
1027		cb->s_outx = 3;
1028		if (so->so_options & SO_DEBUG || traceallspxs)
1029			spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1030		return (0);
1031	}
1032	/*
1033	 * Stuff checksum and output datagram.
1034	 */
1035	if ((si->si_cc & SPX_SP) == 0) {
1036		if (cb->s_force != (1 + SPXT_PERSIST) ||
1037		    cb->s_timer[SPXT_PERSIST] == 0) {
1038			/*
1039			 * If this is a new packet and we are not currently
1040			 * timing anything, time this one.
1041			 */
1042			if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1043				cb->s_smax = si->si_seq;
1044				if (cb->s_rtt == 0) {
1045					spxstat.spxs_segstimed++;
1046					cb->s_rtseq = si->si_seq;
1047					cb->s_rtt = 1;
1048				}
1049			}
1050			/*
1051			 * Set rexmt timer if not currently set,
1052			 * Initial value for retransmit timer is smoothed
1053			 * round-trip time + 2 * round-trip time variance.
1054			 * Initialize shift counter which is used for backoff
1055			 * of retransmit time.
1056			 */
1057			if (cb->s_timer[SPXT_REXMT] == 0 &&
1058			    cb->s_snxt != cb->s_rack) {
1059				cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1060				if (cb->s_timer[SPXT_PERSIST]) {
1061					cb->s_timer[SPXT_PERSIST] = 0;
1062					cb->s_rxtshift = 0;
1063				}
1064			}
1065		} else if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1066			cb->s_smax = si->si_seq;
1067		}
1068	} else if (cb->s_state < TCPS_ESTABLISHED) {
1069		if (cb->s_rtt == 0)
1070			cb->s_rtt = 1; /* Time initial handshake */
1071		if (cb->s_timer[SPXT_REXMT] == 0)
1072			cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1073	}
1074	{
1075		/*
1076		 * Do not request acks when we ack their data packets or
1077		 * when we do a gratuitous window update.
1078		 */
1079		if (((si->si_cc & SPX_SP) == 0) || cb->s_force)
1080				si->si_cc |= SPX_SA;
1081		si->si_seq = htons(si->si_seq);
1082		si->si_alo = htons(alo);
1083		si->si_ack = htons(cb->s_ack);
1084
1085		if (ipxcksum) {
1086			si->si_sum = ipx_cksum(m, ntohs(si->si_len));
1087		} else
1088			si->si_sum = 0xffff;
1089
1090		cb->s_outx = 4;
1091		if (so->so_options & SO_DEBUG || traceallspxs)
1092			spx_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1093
1094		if (so->so_options & SO_DONTROUTE)
1095			error = ipx_outputfl(m, (struct route *)NULL, IPX_ROUTETOIF);
1096		else
1097			error = ipx_outputfl(m, &cb->s_ipxpcb->ipxp_route, 0);
1098	}
1099	if (error) {
1100		return (error);
1101	}
1102	spxstat.spxs_sndtotal++;
1103	/*
1104	 * Data sent (as far as we can tell).
1105	 * If this advertises a larger window than any other segment,
1106	 * then remember the size of the advertized window.
1107	 * Any pending ACK has now been sent.
1108	 */
1109	cb->s_force = 0;
1110	cb->s_flags &= ~(SF_ACKNOW|SF_DELACK);
1111	if (SSEQ_GT(alo, cb->s_alo))
1112		cb->s_alo = alo;
1113	if (sendalot)
1114		goto again;
1115	cb->s_outx = 5;
1116	return (0);
1117}
1118
1119static int spx_do_persist_panics = 0;
1120
1121static void
1122spx_setpersist(cb)
1123	register struct spxpcb *cb;
1124{
1125	register int t = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1126
1127	if (cb->s_timer[SPXT_REXMT] && spx_do_persist_panics)
1128		panic("spx_output REXMT");
1129	/*
1130	 * Start/restart persistance timer.
1131	 */
1132	SPXT_RANGESET(cb->s_timer[SPXT_PERSIST],
1133	    t*spx_backoff[cb->s_rxtshift],
1134	    SPXTV_PERSMIN, SPXTV_PERSMAX);
1135	if (cb->s_rxtshift < SPX_MAXRXTSHIFT)
1136		cb->s_rxtshift++;
1137}
1138
1139int
1140spx_ctloutput(so, sopt)
1141	struct socket *so;
1142	struct sockopt *sopt;
1143{
1144	struct ipxpcb *ipxp = sotoipxpcb(so);
1145	register struct spxpcb *cb;
1146	int mask, error;
1147	short soptval;
1148	u_short usoptval;
1149	int optval;
1150
1151	error = 0;
1152
1153	if (sopt->sopt_level != IPXPROTO_SPX) {
1154		/* This will have to be changed when we do more general
1155		   stacking of protocols */
1156		return (ipx_ctloutput(so, sopt));
1157	}
1158	if (ipxp == NULL)
1159		return (EINVAL);
1160	else
1161		cb = ipxtospxpcb(ipxp);
1162
1163	switch (sopt->sopt_dir) {
1164	case SOPT_GET:
1165		switch (sopt->sopt_name) {
1166		case SO_HEADERS_ON_INPUT:
1167			mask = SF_HI;
1168			goto get_flags;
1169
1170		case SO_HEADERS_ON_OUTPUT:
1171			mask = SF_HO;
1172		get_flags:
1173			soptval = cb->s_flags & mask;
1174			error = sooptcopyout(sopt, &soptval, sizeof soptval);
1175			break;
1176
1177		case SO_MTU:
1178			usoptval = cb->s_mtu;
1179			error = sooptcopyout(sopt, &usoptval, sizeof usoptval);
1180			break;
1181
1182		case SO_LAST_HEADER:
1183			error = sooptcopyout(sopt, &cb->s_rhdr,
1184					     sizeof cb->s_rhdr);
1185			break;
1186
1187		case SO_DEFAULT_HEADERS:
1188			error = sooptcopyout(sopt, &cb->s_shdr,
1189					     sizeof cb->s_shdr);
1190			break;
1191
1192		default:
1193			error = ENOPROTOOPT;
1194		}
1195		break;
1196
1197	case SOPT_SET:
1198		switch (sopt->sopt_name) {
1199			/* XXX why are these shorts on get and ints on set?
1200			   that doesn't make any sense... */
1201		case SO_HEADERS_ON_INPUT:
1202			mask = SF_HI;
1203			goto set_head;
1204
1205		case SO_HEADERS_ON_OUTPUT:
1206			mask = SF_HO;
1207		set_head:
1208			error = sooptcopyin(sopt, &optval, sizeof optval,
1209					    sizeof optval);
1210			if (error)
1211				break;
1212
1213			if (cb->s_flags & SF_PI) {
1214				if (optval)
1215					cb->s_flags |= mask;
1216				else
1217					cb->s_flags &= ~mask;
1218			} else error = EINVAL;
1219			break;
1220
1221		case SO_MTU:
1222			error = sooptcopyin(sopt, &usoptval, sizeof usoptval,
1223					    sizeof usoptval);
1224			if (error)
1225				break;
1226			cb->s_mtu = usoptval;
1227			break;
1228
1229#ifdef SF_NEWCALL
1230		case SO_NEWCALL:
1231			error = sooptcopyin(sopt, &optval, sizeof optval,
1232					    sizeof optval);
1233			if (error)
1234				break;
1235			if (optval) {
1236				cb->s_flags2 |= SF_NEWCALL;
1237				spx_newchecks[5]++;
1238			} else {
1239				cb->s_flags2 &= ~SF_NEWCALL;
1240				spx_newchecks[6]++;
1241			}
1242			break;
1243#endif
1244
1245		case SO_DEFAULT_HEADERS:
1246			{
1247				struct spxhdr sp;
1248
1249				error = sooptcopyin(sopt, &sp, sizeof sp,
1250						    sizeof sp);
1251				if (error)
1252					break;
1253				cb->s_dt = sp.spx_dt;
1254				cb->s_cc = sp.spx_cc & SPX_EM;
1255			}
1256			break;
1257
1258		default:
1259			error = ENOPROTOOPT;
1260		}
1261		break;
1262	}
1263	return (error);
1264}
1265
1266static int
1267spx_usr_abort(so)
1268	struct socket *so;
1269{
1270	int s;
1271	struct ipxpcb *ipxp;
1272	struct spxpcb *cb;
1273
1274	ipxp = sotoipxpcb(so);
1275	cb = ipxtospxpcb(ipxp);
1276
1277	s = splnet();
1278	spx_drop(cb, ECONNABORTED);
1279	splx(s);
1280	return (0);
1281}
1282
1283/*
1284 * Accept a connection.  Essentially all the work is
1285 * done at higher levels; just return the address
1286 * of the peer, storing through addr.
1287 */
1288static int
1289spx_accept(so, nam)
1290	struct socket *so;
1291	struct sockaddr **nam;
1292{
1293	struct ipxpcb *ipxp;
1294	struct sockaddr_ipx *sipx, ssipx;
1295
1296	ipxp = sotoipxpcb(so);
1297	sipx = &ssipx;
1298	bzero(sipx, sizeof *sipx);
1299	sipx->sipx_len = sizeof *sipx;
1300	sipx->sipx_family = AF_IPX;
1301	sipx->sipx_addr = ipxp->ipxp_faddr;
1302	*nam = dup_sockaddr((struct sockaddr *)sipx, 0);
1303	return (0);
1304}
1305
1306static int
1307spx_attach(so, proto, p)
1308	struct socket *so;
1309	int proto;
1310	struct proc *p;
1311{
1312	int error;
1313	int s;
1314	struct ipxpcb *ipxp;
1315	struct spxpcb *cb;
1316	struct mbuf *mm;
1317	struct sockbuf *sb;
1318
1319	ipxp = sotoipxpcb(so);
1320	cb = ipxtospxpcb(ipxp);
1321
1322	if (ipxp != NULL)
1323		return (EISCONN);
1324	s = splnet();
1325	error = ipx_pcballoc(so, &ipxpcb, p);
1326	if (error)
1327		goto spx_attach_end;
1328	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1329		error = soreserve(so, (u_long) 3072, (u_long) 3072);
1330		if (error)
1331			goto spx_attach_end;
1332	}
1333	ipxp = sotoipxpcb(so);
1334
1335	MALLOC(cb, struct spxpcb *, sizeof *cb, M_PCB, M_NOWAIT | M_ZERO);
1336
1337	if (cb == NULL) {
1338		error = ENOBUFS;
1339		goto spx_attach_end;
1340	}
1341	sb = &so->so_snd;
1342
1343	mm = m_getclr(M_DONTWAIT, MT_HEADER);
1344	if (mm == NULL) {
1345		FREE(cb, M_PCB);
1346		error = ENOBUFS;
1347		goto spx_attach_end;
1348	}
1349	cb->s_ipx = mtod(mm, struct ipx *);
1350	cb->s_state = TCPS_LISTEN;
1351	cb->s_smax = -1;
1352	cb->s_swl1 = -1;
1353	cb->s_q.si_next = cb->s_q.si_prev = &cb->s_q;
1354	cb->s_ipxpcb = ipxp;
1355	cb->s_mtu = 576 - sizeof(struct spx);
1356	cb->s_cwnd = sbspace(sb) * CUNIT / cb->s_mtu;
1357	cb->s_ssthresh = cb->s_cwnd;
1358	cb->s_cwmx = sbspace(sb) * CUNIT / (2 * sizeof(struct spx));
1359	/* Above is recomputed when connecting to account
1360	   for changed buffering or mtu's */
1361	cb->s_rtt = SPXTV_SRTTBASE;
1362	cb->s_rttvar = SPXTV_SRTTDFLT << 2;
1363	SPXT_RANGESET(cb->s_rxtcur,
1364	    ((SPXTV_SRTTBASE >> 2) + (SPXTV_SRTTDFLT << 2)) >> 1,
1365	    SPXTV_MIN, SPXTV_REXMTMAX);
1366	ipxp->ipxp_pcb = (caddr_t)cb;
1367spx_attach_end:
1368	splx(s);
1369	return (error);
1370}
1371
1372static int
1373spx_bind(so, nam, p)
1374	struct socket *so;
1375	struct sockaddr *nam;
1376	struct proc *p;
1377{
1378	struct ipxpcb *ipxp;
1379
1380	ipxp = sotoipxpcb(so);
1381
1382	return (ipx_pcbbind(ipxp, nam, p));
1383}
1384
1385/*
1386 * Initiate connection to peer.
1387 * Enter SYN_SENT state, and mark socket as connecting.
1388 * Start keep-alive timer, setup prototype header,
1389 * Send initial system packet requesting connection.
1390 */
1391static int
1392spx_connect(so, nam, p)
1393	struct socket *so;
1394	struct sockaddr *nam;
1395	struct proc *p;
1396{
1397	int error;
1398	int s;
1399	struct ipxpcb *ipxp;
1400	struct spxpcb *cb;
1401
1402	ipxp = sotoipxpcb(so);
1403	cb = ipxtospxpcb(ipxp);
1404
1405	s = splnet();
1406	if (ipxp->ipxp_lport == 0) {
1407		error = ipx_pcbbind(ipxp, (struct sockaddr *)NULL, p);
1408		if (error)
1409			goto spx_connect_end;
1410	}
1411	error = ipx_pcbconnect(ipxp, nam, p);
1412	if (error)
1413		goto spx_connect_end;
1414	soisconnecting(so);
1415	spxstat.spxs_connattempt++;
1416	cb->s_state = TCPS_SYN_SENT;
1417	cb->s_did = 0;
1418	spx_template(cb);
1419	cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
1420	cb->s_force = 1 + SPXTV_KEEP;
1421	/*
1422	 * Other party is required to respond to
1423	 * the port I send from, but he is not
1424	 * required to answer from where I am sending to,
1425	 * so allow wildcarding.
1426	 * original port I am sending to is still saved in
1427	 * cb->s_dport.
1428	 */
1429	ipxp->ipxp_fport = 0;
1430	error = spx_output(cb, (struct mbuf *)NULL);
1431spx_connect_end:
1432	splx(s);
1433	return (error);
1434}
1435
1436static int
1437spx_detach(so)
1438	struct socket *so;
1439{
1440	int s;
1441	struct ipxpcb *ipxp;
1442	struct spxpcb *cb;
1443
1444	ipxp = sotoipxpcb(so);
1445	cb = ipxtospxpcb(ipxp);
1446
1447	if (ipxp == NULL)
1448		return (ENOTCONN);
1449	s = splnet();
1450	if (cb->s_state > TCPS_LISTEN)
1451		spx_disconnect(cb);
1452	else
1453		spx_close(cb);
1454	splx(s);
1455	return (0);
1456}
1457
1458/*
1459 * We may decide later to implement connection closing
1460 * handshaking at the spx level optionally.
1461 * here is the hook to do it:
1462 */
1463static int
1464spx_usr_disconnect(so)
1465	struct socket *so;
1466{
1467	int s;
1468	struct ipxpcb *ipxp;
1469	struct spxpcb *cb;
1470
1471	ipxp = sotoipxpcb(so);
1472	cb = ipxtospxpcb(ipxp);
1473
1474	s = splnet();
1475	spx_disconnect(cb);
1476	splx(s);
1477	return (0);
1478}
1479
1480static int
1481spx_listen(so, p)
1482	struct socket *so;
1483	struct proc *p;
1484{
1485	int error;
1486	struct ipxpcb *ipxp;
1487	struct spxpcb *cb;
1488
1489	error = 0;
1490	ipxp = sotoipxpcb(so);
1491	cb = ipxtospxpcb(ipxp);
1492
1493	if (ipxp->ipxp_lport == 0)
1494		error = ipx_pcbbind(ipxp, (struct sockaddr *)NULL, p);
1495	if (error == 0)
1496		cb->s_state = TCPS_LISTEN;
1497	return (error);
1498}
1499
1500/*
1501 * After a receive, possibly send acknowledgment
1502 * updating allocation.
1503 */
1504static int
1505spx_rcvd(so, flags)
1506	struct socket *so;
1507	int flags;
1508{
1509	int s;
1510	struct ipxpcb *ipxp;
1511	struct spxpcb *cb;
1512
1513	ipxp = sotoipxpcb(so);
1514	cb = ipxtospxpcb(ipxp);
1515
1516	s = splnet();
1517	cb->s_flags |= SF_RVD;
1518	spx_output(cb, (struct mbuf *)NULL);
1519	cb->s_flags &= ~SF_RVD;
1520	splx(s);
1521	return (0);
1522}
1523
1524static int
1525spx_rcvoob(so, m, flags)
1526	struct socket *so;
1527	struct mbuf *m;
1528	int flags;
1529{
1530	struct ipxpcb *ipxp;
1531	struct spxpcb *cb;
1532
1533	ipxp = sotoipxpcb(so);
1534	cb = ipxtospxpcb(ipxp);
1535
1536	if ((cb->s_oobflags & SF_IOOB) || so->so_oobmark ||
1537	    (so->so_state & SS_RCVATMARK)) {
1538		m->m_len = 1;
1539		*mtod(m, caddr_t) = cb->s_iobc;
1540		return (0);
1541	}
1542	return (EINVAL);
1543}
1544
1545static int
1546spx_send(so, flags, m, addr, controlp, p)
1547	struct socket *so;
1548	int flags;
1549	struct mbuf *m;
1550	struct sockaddr *addr;
1551	struct mbuf *controlp;
1552	struct proc *p;
1553{
1554	int error;
1555	int s;
1556	struct ipxpcb *ipxp;
1557	struct spxpcb *cb;
1558
1559	error = 0;
1560	ipxp = sotoipxpcb(so);
1561	cb = ipxtospxpcb(ipxp);
1562
1563	s = splnet();
1564	if (flags & PRUS_OOB) {
1565		if (sbspace(&so->so_snd) < -512) {
1566			error = ENOBUFS;
1567			goto spx_send_end;
1568		}
1569		cb->s_oobflags |= SF_SOOB;
1570	}
1571	if (controlp != NULL) {
1572		u_short *p = mtod(controlp, u_short *);
1573		spx_newchecks[2]++;
1574		if ((p[0] == 5) && (p[1] == 1)) { /* XXXX, for testing */
1575			cb->s_shdr.spx_dt = *(u_char *)(&p[2]);
1576			spx_newchecks[3]++;
1577		}
1578		m_freem(controlp);
1579	}
1580	controlp = NULL;
1581	error = spx_output(cb, m);
1582	m = NULL;
1583spx_send_end:
1584	if (controlp != NULL)
1585		m_freem(controlp);
1586	if (m != NULL)
1587		m_freem(m);
1588	splx(s);
1589	return (error);
1590}
1591
1592static int
1593spx_shutdown(so)
1594	struct socket *so;
1595{
1596	int error;
1597	int s;
1598	struct ipxpcb *ipxp;
1599	struct spxpcb *cb;
1600
1601	error = 0;
1602	ipxp = sotoipxpcb(so);
1603	cb = ipxtospxpcb(ipxp);
1604
1605	s = splnet();
1606	socantsendmore(so);
1607	cb = spx_usrclosed(cb);
1608	if (cb != NULL)
1609		error = spx_output(cb, (struct mbuf *)NULL);
1610	splx(s);
1611	return (error);
1612}
1613
1614static int
1615spx_sp_attach(so, proto, p)
1616	struct socket *so;
1617	int proto;
1618	struct proc *p;
1619{
1620	int error;
1621	struct ipxpcb *ipxp;
1622
1623	error = spx_attach(so, proto, p);
1624	if (error == 0) {
1625		ipxp = sotoipxpcb(so);
1626		((struct spxpcb *)ipxp->ipxp_pcb)->s_flags |=
1627					(SF_HI | SF_HO | SF_PI);
1628	}
1629	return (error);
1630}
1631
1632/*
1633 * Create template to be used to send spx packets on a connection.
1634 * Called after host entry created, fills
1635 * in a skeletal spx header (choosing connection id),
1636 * minimizing the amount of work necessary when the connection is used.
1637 */
1638static void
1639spx_template(cb)
1640	register struct spxpcb *cb;
1641{
1642	register struct ipxpcb *ipxp = cb->s_ipxpcb;
1643	register struct ipx *ipx = cb->s_ipx;
1644	register struct sockbuf *sb = &(ipxp->ipxp_socket->so_snd);
1645
1646	ipx->ipx_pt = IPXPROTO_SPX;
1647	ipx->ipx_sna = ipxp->ipxp_laddr;
1648	ipx->ipx_dna = ipxp->ipxp_faddr;
1649	cb->s_sid = htons(spx_iss);
1650	spx_iss += SPX_ISSINCR/2;
1651	cb->s_alo = 1;
1652	cb->s_cwnd = (sbspace(sb) * CUNIT) / cb->s_mtu;
1653	cb->s_ssthresh = cb->s_cwnd; /* Try to expand fast to full complement
1654					of large packets */
1655	cb->s_cwmx = (sbspace(sb) * CUNIT) / (2 * sizeof(struct spx));
1656	cb->s_cwmx = max(cb->s_cwmx, cb->s_cwnd);
1657		/* But allow for lots of little packets as well */
1658}
1659
1660/*
1661 * Close a SPIP control block:
1662 *	discard spx control block itself
1663 *	discard ipx protocol control block
1664 *	wake up any sleepers
1665 */
1666static struct spxpcb *
1667spx_close(cb)
1668	register struct spxpcb *cb;
1669{
1670	register struct spx_q *s;
1671	struct ipxpcb *ipxp = cb->s_ipxpcb;
1672	struct socket *so = ipxp->ipxp_socket;
1673	register struct mbuf *m;
1674
1675	s = cb->s_q.si_next;
1676	while (s != &(cb->s_q)) {
1677		s = s->si_next;
1678		m = dtom(s->si_prev);
1679		remque(s->si_prev);
1680		m_freem(m);
1681	}
1682	m_free(dtom(cb->s_ipx));
1683	FREE(cb, M_PCB);
1684	ipxp->ipxp_pcb = 0;
1685	soisdisconnected(so);
1686	ipx_pcbdetach(ipxp);
1687	spxstat.spxs_closed++;
1688	return ((struct spxpcb *)NULL);
1689}
1690
1691/*
1692 *	Someday we may do level 3 handshaking
1693 *	to close a connection or send a xerox style error.
1694 *	For now, just close.
1695 */
1696static struct spxpcb *
1697spx_usrclosed(cb)
1698	register struct spxpcb *cb;
1699{
1700	return (spx_close(cb));
1701}
1702
1703static struct spxpcb *
1704spx_disconnect(cb)
1705	register struct spxpcb *cb;
1706{
1707	return (spx_close(cb));
1708}
1709
1710/*
1711 * Drop connection, reporting
1712 * the specified error.
1713 */
1714static struct spxpcb *
1715spx_drop(cb, errno)
1716	register struct spxpcb *cb;
1717	int errno;
1718{
1719	struct socket *so = cb->s_ipxpcb->ipxp_socket;
1720
1721	/*
1722	 * someday, in the xerox world
1723	 * we will generate error protocol packets
1724	 * announcing that the socket has gone away.
1725	 */
1726	if (TCPS_HAVERCVDSYN(cb->s_state)) {
1727		spxstat.spxs_drops++;
1728		cb->s_state = TCPS_CLOSED;
1729		/*tcp_output(cb);*/
1730	} else
1731		spxstat.spxs_conndrops++;
1732	so->so_error = errno;
1733	return (spx_close(cb));
1734}
1735
1736/*
1737 * Fast timeout routine for processing delayed acks
1738 */
1739void
1740spx_fasttimo()
1741{
1742	register struct ipxpcb *ipxp;
1743	register struct spxpcb *cb;
1744	int s = splnet();
1745
1746	ipxp = ipxpcb.ipxp_next;
1747	if (ipxp != NULL)
1748	for (; ipxp != &ipxpcb; ipxp = ipxp->ipxp_next)
1749		if ((cb = (struct spxpcb *)ipxp->ipxp_pcb) != NULL &&
1750		    (cb->s_flags & SF_DELACK)) {
1751			cb->s_flags &= ~SF_DELACK;
1752			cb->s_flags |= SF_ACKNOW;
1753			spxstat.spxs_delack++;
1754			spx_output(cb, (struct mbuf *)NULL);
1755		}
1756	splx(s);
1757}
1758
1759/*
1760 * spx protocol timeout routine called every 500 ms.
1761 * Updates the timers in all active pcb's and
1762 * causes finite state machine actions if timers expire.
1763 */
1764void
1765spx_slowtimo()
1766{
1767	register struct ipxpcb *ip, *ipnxt;
1768	register struct spxpcb *cb;
1769	int s = splnet();
1770	register int i;
1771
1772	/*
1773	 * Search through tcb's and update active timers.
1774	 */
1775	ip = ipxpcb.ipxp_next;
1776	if (ip == NULL) {
1777		splx(s);
1778		return;
1779	}
1780	while (ip != &ipxpcb) {
1781		cb = ipxtospxpcb(ip);
1782		ipnxt = ip->ipxp_next;
1783		if (cb == NULL)
1784			goto tpgone;
1785		for (i = 0; i < SPXT_NTIMERS; i++) {
1786			if (cb->s_timer[i] && --cb->s_timer[i] == 0) {
1787				spx_timers(cb, i);
1788				if (ipnxt->ipxp_prev != ip)
1789					goto tpgone;
1790			}
1791		}
1792		cb->s_idle++;
1793		if (cb->s_rtt)
1794			cb->s_rtt++;
1795tpgone:
1796		ip = ipnxt;
1797	}
1798	spx_iss += SPX_ISSINCR/PR_SLOWHZ;		/* increment iss */
1799	splx(s);
1800}
1801
1802/*
1803 * SPX timer processing.
1804 */
1805static struct spxpcb *
1806spx_timers(cb, timer)
1807	register struct spxpcb *cb;
1808	int timer;
1809{
1810	long rexmt;
1811	int win;
1812
1813	cb->s_force = 1 + timer;
1814	switch (timer) {
1815
1816	/*
1817	 * 2 MSL timeout in shutdown went off.  TCP deletes connection
1818	 * control block.
1819	 */
1820	case SPXT_2MSL:
1821		printf("spx: SPXT_2MSL went off for no reason\n");
1822		cb->s_timer[timer] = 0;
1823		break;
1824
1825	/*
1826	 * Retransmission timer went off.  Message has not
1827	 * been acked within retransmit interval.  Back off
1828	 * to a longer retransmit interval and retransmit one packet.
1829	 */
1830	case SPXT_REXMT:
1831		if (++cb->s_rxtshift > SPX_MAXRXTSHIFT) {
1832			cb->s_rxtshift = SPX_MAXRXTSHIFT;
1833			spxstat.spxs_timeoutdrop++;
1834			cb = spx_drop(cb, ETIMEDOUT);
1835			break;
1836		}
1837		spxstat.spxs_rexmttimeo++;
1838		rexmt = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1839		rexmt *= spx_backoff[cb->s_rxtshift];
1840		SPXT_RANGESET(cb->s_rxtcur, rexmt, SPXTV_MIN, SPXTV_REXMTMAX);
1841		cb->s_timer[SPXT_REXMT] = cb->s_rxtcur;
1842		/*
1843		 * If we have backed off fairly far, our srtt
1844		 * estimate is probably bogus.  Clobber it
1845		 * so we'll take the next rtt measurement as our srtt;
1846		 * move the current srtt into rttvar to keep the current
1847		 * retransmit times until then.
1848		 */
1849		if (cb->s_rxtshift > SPX_MAXRXTSHIFT / 4 ) {
1850			cb->s_rttvar += (cb->s_srtt >> 2);
1851			cb->s_srtt = 0;
1852		}
1853		cb->s_snxt = cb->s_rack;
1854		/*
1855		 * If timing a packet, stop the timer.
1856		 */
1857		cb->s_rtt = 0;
1858		/*
1859		 * See very long discussion in tcp_timer.c about congestion
1860		 * window and sstrhesh
1861		 */
1862		win = min(cb->s_swnd, (cb->s_cwnd/CUNIT)) / 2;
1863		if (win < 2)
1864			win = 2;
1865		cb->s_cwnd = CUNIT;
1866		cb->s_ssthresh = win * CUNIT;
1867		spx_output(cb, (struct mbuf *)NULL);
1868		break;
1869
1870	/*
1871	 * Persistance timer into zero window.
1872	 * Force a probe to be sent.
1873	 */
1874	case SPXT_PERSIST:
1875		spxstat.spxs_persisttimeo++;
1876		spx_setpersist(cb);
1877		spx_output(cb, (struct mbuf *)NULL);
1878		break;
1879
1880	/*
1881	 * Keep-alive timer went off; send something
1882	 * or drop connection if idle for too long.
1883	 */
1884	case SPXT_KEEP:
1885		spxstat.spxs_keeptimeo++;
1886		if (cb->s_state < TCPS_ESTABLISHED)
1887			goto dropit;
1888		if (cb->s_ipxpcb->ipxp_socket->so_options & SO_KEEPALIVE) {
1889		    	if (cb->s_idle >= SPXTV_MAXIDLE)
1890				goto dropit;
1891			spxstat.spxs_keepprobe++;
1892			spx_output(cb, (struct mbuf *)NULL);
1893		} else
1894			cb->s_idle = 0;
1895		cb->s_timer[SPXT_KEEP] = SPXTV_KEEP;
1896		break;
1897	dropit:
1898		spxstat.spxs_keepdrops++;
1899		cb = spx_drop(cb, ETIMEDOUT);
1900		break;
1901	}
1902	return (cb);
1903}
1904