2 * Copyright (c) 1984, 1985, 1986, 1987, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)spp_usrreq.c 8.1 (Berkeley) 6/10/93
34 * $FreeBSD: src/sys/netns/spp_usrreq.c,v 1.11 1999/08/28 00:49:53 peter Exp $
35 * $DragonFly: src/sys/netproto/ns/spp_usrreq.c,v 1.2 2003/06/17 04:28:53 dillon Exp $
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/malloc.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/errno.h>
48 #include <net/route.h>
49 #include <netinet/tcp_fsm.h>
52 #include <netns/ns_pcb.h>
53 #include <netns/idp.h>
54 #include <netns/idp_var.h>
55 #include <netns/ns_error.h>
57 #include <netns/spidp.h>
58 #include <netns/spp_timer.h>
59 #include <netns/spp_var.h>
60 #include <netns/spp_debug.h>
63 * SP protocol implementation.
68 spp_iss = 1; /* WRONG !! should fish it out of TODR */
70 struct spidp spp_savesi;
72 extern int sppconsdebug;
74 int spp_use_delack = 0;
75 u_short spp_newchecks[50];
79 register struct mbuf *m;
80 register struct nspcb *nsp;
82 register struct sppcb *cb;
83 register struct spidp *si = mtod(m, struct spidp *);
84 register struct socket *so;
89 sppstat.spps_rcvtotal++;
91 panic("No nspcb in spp_input");
96 if (cb == 0) goto bad;
98 if (m->m_len < sizeof(*si)) {
99 if ((m = m_pullup(m, sizeof(*si))) == 0) {
100 sppstat.spps_rcvshort++;
103 si = mtod(m, struct spidp *);
105 si->si_seq = ntohs(si->si_seq);
106 si->si_ack = ntohs(si->si_ack);
107 si->si_alo = ntohs(si->si_alo);
109 so = nsp->nsp_socket;
110 if (so->so_options & SO_DEBUG || traceallspps) {
111 ostate = cb->s_state;
114 if (so->so_options & SO_ACCEPTCONN) {
115 struct sppcb *ocb = cb;
117 so = sonewconn(so, 0);
122 * This is ugly, but ....
124 * Mark socket as temporary until we're
125 * committed to keeping it. The code at
126 * ``drop'' and ``dropwithreset'' check the
127 * flag dropsocket to see if the temporary
128 * socket created here should be discarded.
129 * We mark the socket as discardable until
130 * we're committed to it below in TCPS_LISTEN.
133 nsp = (struct nspcb *)so->so_pcb;
134 nsp->nsp_laddr = si->si_dna;
136 cb->s_mtu = ocb->s_mtu; /* preserve sockopts */
137 cb->s_flags = ocb->s_flags; /* preserve sockopts */
138 cb->s_flags2 = ocb->s_flags2; /* preserve sockopts */
139 cb->s_state = TCPS_LISTEN;
143 * Packet received on connection.
144 * reset idle time and keep-alive timer;
147 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
149 switch (cb->s_state) {
153 register struct sockaddr_ns *sns;
154 struct ns_addr laddr;
157 * If somebody here was carying on a conversation
158 * and went away, and his pen pal thinks he can
159 * still talk, we get the misdirected packet.
161 if (spp_hardnosed && (si->si_did != 0 || si->si_seq != 0)) {
165 am = m_get(M_DONTWAIT, MT_SONAME);
168 am->m_len = sizeof (struct sockaddr_ns);
169 sns = mtod(am, struct sockaddr_ns *);
170 sns->sns_len = sizeof(*sns);
171 sns->sns_family = AF_NS;
172 sns->sns_addr = si->si_sna;
173 laddr = nsp->nsp_laddr;
174 if (ns_nullhost(laddr))
175 nsp->nsp_laddr = si->si_dna;
176 if (ns_pcbconnect(nsp, am)) {
177 nsp->nsp_laddr = laddr;
184 dropsocket = 0; /* committed to socket */
185 cb->s_did = si->si_sid;
186 cb->s_rack = si->si_ack;
187 cb->s_ralo = si->si_alo;
188 #define THREEWAYSHAKE
190 cb->s_state = TCPS_SYN_RECEIVED;
191 cb->s_force = 1 + SPPT_KEEP;
192 sppstat.spps_accepts++;
193 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
197 * This state means that we have heard a response
198 * to our acceptance of their connection
199 * It is probably logically unnecessary in this
202 case TCPS_SYN_RECEIVED: {
203 if (si->si_did!=cb->s_sid) {
208 nsp->nsp_fport = si->si_sport;
209 cb->s_timer[SPPT_REXMT] = 0;
210 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
212 cb->s_state = TCPS_ESTABLISHED;
213 sppstat.spps_accepts++;
218 * This state means that we have gotten a response
219 * to our attempt to establish a connection.
220 * We fill in the data from the other side,
221 * telling us which port to respond to, instead of the well-
222 * known one we might have sent to in the first place.
223 * We also require that this is a response to our
227 if (si->si_did!=cb->s_sid) {
231 sppstat.spps_connects++;
232 cb->s_did = si->si_sid;
233 cb->s_rack = si->si_ack;
234 cb->s_ralo = si->si_alo;
235 cb->s_dport = nsp->nsp_fport = si->si_sport;
236 cb->s_timer[SPPT_REXMT] = 0;
237 cb->s_flags |= SF_ACKNOW;
239 cb->s_state = TCPS_ESTABLISHED;
240 /* Use roundtrip time of connection request for initial rtt */
242 cb->s_srtt = cb->s_rtt << 3;
243 cb->s_rttvar = cb->s_rtt << 1;
244 SPPT_RANGESET(cb->s_rxtcur,
245 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
246 SPPTV_MIN, SPPTV_REXMTMAX);
250 if (so->so_options & SO_DEBUG || traceallspps)
251 spp_trace(SA_INPUT, (u_char)ostate, cb, &spp_savesi, 0);
253 m->m_len -= sizeof (struct idp);
254 m->m_pkthdr.len -= sizeof (struct idp);
255 m->m_data += sizeof (struct idp);
257 if (spp_reass(cb, si)) {
260 if (cb->s_force || (cb->s_flags & (SF_ACKNOW|SF_WIN|SF_RXT)))
261 (void) spp_output(cb, (struct mbuf *)0);
262 cb->s_flags &= ~(SF_WIN|SF_RXT);
268 si->si_seq = ntohs(si->si_seq);
269 si->si_ack = ntohs(si->si_ack);
270 si->si_alo = ntohs(si->si_alo);
271 ns_error(dtom(si), NS_ERR_NOSOCK, 0);
272 if (cb->s_nspcb->nsp_socket->so_options & SO_DEBUG || traceallspps)
273 spp_trace(SA_DROP, (u_char)ostate, cb, &spp_savesi, 0);
278 if (cb == 0 || cb->s_nspcb->nsp_socket->so_options & SO_DEBUG ||
280 spp_trace(SA_DROP, (u_char)ostate, cb, &spp_savesi, 0);
284 int spprexmtthresh = 3;
287 * This is structurally similar to the tcp reassembly routine
288 * but its function is somewhat different: It merely queues
289 * packets up, and suppresses duplicates.
292 register struct sppcb *cb;
293 register struct spidp *si;
295 register struct spidp_q *q;
296 register struct mbuf *m;
297 register struct socket *so = cb->s_nspcb->nsp_socket;
298 char packetp = cb->s_flags & SF_HI;
305 * Update our news from them.
307 if (si->si_cc & SP_SA)
308 cb->s_flags |= (spp_use_delack ? SF_DELACK : SF_ACKNOW);
309 if (SSEQ_GT(si->si_alo, cb->s_ralo))
310 cb->s_flags |= SF_WIN;
311 if (SSEQ_LEQ(si->si_ack, cb->s_rack)) {
312 if ((si->si_cc & SP_SP) && cb->s_rack != (cb->s_smax + 1)) {
313 sppstat.spps_rcvdupack++;
315 * If this is a completely duplicate ack
316 * and other conditions hold, we assume
317 * a packet has been dropped and retransmit
318 * it exactly as in tcp_input().
320 if (si->si_ack != cb->s_rack ||
321 si->si_alo != cb->s_ralo)
323 else if (++cb->s_dupacks == spprexmtthresh) {
324 u_short onxt = cb->s_snxt;
325 int cwnd = cb->s_cwnd;
327 cb->s_snxt = si->si_ack;
329 cb->s_force = 1 + SPPT_REXMT;
330 (void) spp_output(cb, (struct mbuf *)0);
331 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
333 if (cwnd >= 4 * CUNIT)
334 cb->s_cwnd = cwnd / 2;
335 if (SSEQ_GT(onxt, cb->s_snxt))
345 * If our correspondent acknowledges data we haven't sent
346 * TCP would drop the packet after acking. We'll be a little
349 if (SSEQ_GT(si->si_ack, (cb->s_smax + 1))) {
350 sppstat.spps_rcvacktoomuch++;
351 si->si_ack = cb->s_smax + 1;
353 sppstat.spps_rcvackpack++;
355 * If transmit timer is running and timed sequence
356 * number was acked, update smoothed round trip time.
357 * See discussion of algorithm in tcp_input.c
359 if (cb->s_rtt && SSEQ_GT(si->si_ack, cb->s_rtseq)) {
360 sppstat.spps_rttupdated++;
361 if (cb->s_srtt != 0) {
362 register short delta;
363 delta = cb->s_rtt - (cb->s_srtt >> 3);
364 if ((cb->s_srtt += delta) <= 0)
368 delta -= (cb->s_rttvar >> 2);
369 if ((cb->s_rttvar += delta) <= 0)
373 * No rtt measurement yet
375 cb->s_srtt = cb->s_rtt << 3;
376 cb->s_rttvar = cb->s_rtt << 1;
380 SPPT_RANGESET(cb->s_rxtcur,
381 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
382 SPPTV_MIN, SPPTV_REXMTMAX);
385 * If all outstanding data is acked, stop retransmit
386 * timer and remember to restart (more output or persist).
387 * If there is more data to be acked, restart retransmit
388 * timer, using current (possibly backed-off) value;
390 if (si->si_ack == cb->s_smax + 1) {
391 cb->s_timer[SPPT_REXMT] = 0;
392 cb->s_flags |= SF_RXT;
393 } else if (cb->s_timer[SPPT_PERSIST] == 0)
394 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
396 * When new data is acked, open the congestion window.
397 * If the window gives us less than ssthresh packets
398 * in flight, open exponentially (maxseg at a time).
399 * Otherwise open linearly (maxseg^2 / cwnd at a time).
402 if (cb->s_cwnd > cb->s_ssthresh)
403 incr = max(incr * incr / cb->s_cwnd, 1);
404 cb->s_cwnd = min(cb->s_cwnd + incr, cb->s_cwmx);
406 * Trim Acked data from output queue.
408 while ((m = so->so_snd.sb_mb) != NULL) {
409 if (SSEQ_LT((mtod(m, struct spidp *))->si_seq, si->si_ack))
410 sbdroprecord(&so->so_snd);
415 cb->s_rack = si->si_ack;
417 if (SSEQ_LT(cb->s_snxt, cb->s_rack))
418 cb->s_snxt = cb->s_rack;
419 if (SSEQ_LT(cb->s_swl1, si->si_seq) || cb->s_swl1 == si->si_seq &&
420 (SSEQ_LT(cb->s_swl2, si->si_ack) ||
421 cb->s_swl2 == si->si_ack && SSEQ_LT(cb->s_ralo, si->si_alo))) {
422 /* keep track of pure window updates */
423 if ((si->si_cc & SP_SP) && cb->s_swl2 == si->si_ack
424 && SSEQ_LT(cb->s_ralo, si->si_alo)) {
425 sppstat.spps_rcvwinupd++;
426 sppstat.spps_rcvdupack--;
428 cb->s_ralo = si->si_alo;
429 cb->s_swl1 = si->si_seq;
430 cb->s_swl2 = si->si_ack;
431 cb->s_swnd = (1 + si->si_alo - si->si_ack);
432 if (cb->s_swnd > cb->s_smxw)
433 cb->s_smxw = cb->s_swnd;
434 cb->s_flags |= SF_WIN;
437 * If this packet number is higher than that which
438 * we have allocated refuse it, unless urgent
440 if (SSEQ_GT(si->si_seq, cb->s_alo)) {
441 if (si->si_cc & SP_SP) {
442 sppstat.spps_rcvwinprobe++;
445 sppstat.spps_rcvpackafterwin++;
446 if (si->si_cc & SP_OB) {
447 if (SSEQ_GT(si->si_seq, cb->s_alo + 60)) {
448 ns_error(dtom(si), NS_ERR_FULLUP, 0);
450 } /* else queue this packet; */
452 /*register struct socket *so = cb->s_nspcb->nsp_socket;
453 if (so->so_state && SS_NOFDREF) {
454 ns_error(dtom(si), NS_ERR_NOSOCK, 0);
459 ns_error(dtom(si), NS_ERR_FULLUP, 0);
464 * If this is a system packet, we don't need to
465 * queue it up, and won't update acknowledge #
467 if (si->si_cc & SP_SP) {
471 * We have already seen this packet, so drop.
473 if (SSEQ_LT(si->si_seq, cb->s_ack)) {
475 sppstat.spps_rcvduppack++;
476 if (si->si_seq == cb->s_ack - 1)
481 * Loop through all packets queued up to insert in
482 * appropriate sequence.
484 for (q = cb->s_q.si_next; q!=&cb->s_q; q = q->si_next) {
485 if (si->si_seq == SI(q)->si_seq) {
486 sppstat.spps_rcvduppack++;
489 if (SSEQ_LT(si->si_seq, SI(q)->si_seq)) {
490 sppstat.spps_rcvoopack++;
494 insque(si, q->si_prev);
496 * If this packet is urgent, inform process
498 if (si->si_cc & SP_OB) {
499 cb->s_iobc = ((char *)si)[1 + sizeof(*si)];
501 cb->s_oobflags |= SF_IOOB;
504 #define SPINC sizeof(struct sphdr)
506 * Loop through all packets queued up to update acknowledge
507 * number, and present all acknowledged data to user;
508 * If in packet interface mode, show packet headers.
510 for (q = cb->s_q.si_next; q!=&cb->s_q; q = q->si_next) {
511 if (SI(q)->si_seq == cb->s_ack) {
514 if (SI(q)->si_cc & SP_OB) {
515 cb->s_oobflags &= ~SF_IOOB;
516 if (so->so_rcv.sb_cc)
517 so->so_oobmark = so->so_rcv.sb_cc;
519 so->so_state |= SS_RCVATMARK;
524 sppstat.spps_rcvpack++;
526 if (cb->s_flags2 & SF_NEWCALL) {
527 struct sphdr *sp = mtod(m, struct sphdr *);
528 u_char dt = sp->sp_dt;
530 if (dt != cb->s_rhdr.sp_dt) {
532 m_getclr(M_DONTWAIT, MT_CONTROL);
537 cb->s_rhdr.sp_dt = dt;
538 mm->m_len = 5; /*XXX*/
541 *(u_char *)(&s[2]) = dt;
542 sbappend(&so->so_rcv, mm);
545 if (sp->sp_cc & SP_OB) {
546 MCHTYPE(m, MT_OOBDATA);
549 so->so_state &= ~SS_RCVATMARK;
554 m->m_pkthdr.len -= SPINC;
556 if ((sp->sp_cc & SP_EM) || packetp) {
557 sbappendrecord(&so->so_rcv, m);
560 sbappend(&so->so_rcv, m);
564 sbappendrecord(&so->so_rcv, m);
566 cb->s_rhdr = *mtod(m, struct sphdr *);
569 m->m_pkthdr.len -= SPINC;
570 sbappend(&so->so_rcv, m);
575 if (wakeup) sorwakeup(so);
579 spp_ctlinput(cmd, arg)
584 extern u_char nsctlerrmap[];
585 extern spp_abort(), spp_quench();
586 extern struct nspcb *idp_drop();
587 struct ns_errp *errp;
589 struct sockaddr_ns *sns;
592 if (cmd < 0 || cmd > PRC_NCMDS)
594 type = NS_ERR_UNREACH_HOST;
603 case PRC_HOSTUNREACH:
604 sns = (struct sockaddr_ns *)arg;
605 if (sns->sns_family != AF_NS)
611 errp = (struct ns_errp *)arg;
612 na = &errp->ns_err_idp.idp_dna;
613 type = errp->ns_err_num;
614 type = ntohs((u_short)type);
618 case NS_ERR_UNREACH_HOST:
619 ns_pcbnotify(na, (int)nsctlerrmap[cmd], spp_abort, (long) 0);
624 nsp = ns_pcblookup(na, errp->ns_err_idp.idp_sna.x_port,
628 (void) spp_drop((struct sppcb *)nsp->nsp_pcb,
629 (int)nsctlerrmap[cmd]);
631 (void) idp_drop(nsp, (int)nsctlerrmap[cmd]);
636 ns_pcbnotify(na, 0, spp_quench, (long) 0);
640 * When a source quench is received, close congestion window
641 * to one packet. We will gradually open it again as we proceed.
646 struct sppcb *cb = nstosppcb(nsp);
655 register struct nspcb *nsp;
657 register struct sppcb *cb = (struct sppcb *)(nsp->nsp_pcb);
658 register struct mbuf *m;
659 register struct spidp *si;
663 struct mbuf *firstbad, *m0;
667 * The notification that we have sent
668 * too much is bad news -- we will
669 * have to go through queued up so far
670 * splitting ones which are too big and
671 * reassigning sequence numbers and checksums.
672 * we should then retransmit all packets from
673 * one above the offending packet to the last one
674 * we had sent (or our allocation)
675 * then the offending one so that the any queued
676 * data at our destination will be discarded.
678 ep = (struct ns_errp *)nsp->nsp_notify_param;
679 sb = &nsp->nsp_socket->so_snd;
680 cb->s_mtu = ep->ns_err_param;
681 badseq = SI(&ep->ns_err_idp)->si_seq;
682 for (m = sb->sb_mb; m; m = m->m_act) {
683 si = mtod(m, struct spidp *);
684 if (si->si_seq == badseq)
690 /* calculate length */
691 for (m0 = m, len = 0; m ; m = m->m_next)
693 if (len > cb->s_mtu) {
702 register struct sppcb *cb;
705 struct socket *so = cb->s_nspcb->nsp_socket;
706 register struct mbuf *m;
707 register struct spidp *si = (struct spidp *) 0;
708 register struct sockbuf *sb = &so->so_snd;
709 int len = 0, win, rcv_win;
710 short span, off, recordp = 0;
712 int error = 0, sendalot;
723 * Make sure that packet isn't too big.
725 for (m = m0; m ; m = m->m_next) {
728 if (m->m_flags & M_EOR)
731 datalen = (cb->s_flags & SF_HO) ?
732 len - sizeof (struct sphdr) : len;
734 if (cb->s_flags & SF_PI) {
738 int oldEM = cb->s_cc & SP_EM;
743 * Here we are only being called
744 * from usrreq(), so it is OK to
747 m = m_copym(m0, 0, mtu, M_WAIT);
748 if (cb->s_flags & SF_NEWCALL) {
752 mm->m_flags &= ~M_EOR;
756 error = spp_output(cb, m);
769 * Force length even, by adding a "garbage byte" if
774 if (M_TRAILINGSPACE(m) >= 1)
777 struct mbuf *m1 = m_get(M_DONTWAIT, MT_DATA);
784 *(mtod(m1, u_char *)) = 0;
788 m = m_gethdr(M_DONTWAIT, MT_HEADER);
794 * Fill in mbuf with extended SP header
795 * and addresses and length put into network format.
797 MH_ALIGN(m, sizeof (struct spidp));
798 m->m_len = sizeof (struct spidp);
800 si = mtod(m, struct spidp *);
801 si->si_i = *cb->s_idp;
802 si->si_s = cb->s_shdr;
803 if ((cb->s_flags & SF_PI) && (cb->s_flags & SF_HO)) {
804 register struct sphdr *sh;
805 if (m0->m_len < sizeof (*sh)) {
806 if((m0 = m_pullup(m0, sizeof(*sh))) == NULL) {
813 sh = mtod(m0, struct sphdr *);
814 si->si_dt = sh->sp_dt;
815 si->si_cc |= sh->sp_cc & SP_EM;
816 m0->m_len -= sizeof (*sh);
817 m0->m_data += sizeof (*sh);
821 if ((cb->s_flags2 & SF_NEWCALL) && recordp) {
825 if (cb->s_oobflags & SF_SOOB) {
828 * make sure OB packets convey exactly 1 byte.
829 * If the packet is 1 byte or larger, we
830 * have already guaranted there to be at least
831 * one garbage byte for the checksum, and
832 * extra bytes shouldn't hurt!
834 if (len > sizeof(*si)) {
836 len = (1 + sizeof(*si));
839 si->si_len = htons((u_short)len);
840 m->m_pkthdr.len = ((len - 1) | 1) + 1;
842 * queue stuff up for output
844 sbappendrecord(sb, m);
848 idle = (cb->s_smax == (cb->s_rack - 1));
852 off = cb->s_snxt - cb->s_rack;
853 win = min(cb->s_swnd, (cb->s_cwnd/CUNIT));
856 * If in persist timeout with window of 0, send a probe.
857 * Otherwise, if window is small but nonzero
858 * and timer expired, send what we can and go into
861 if (cb->s_force == 1 + SPPT_PERSIST) {
863 cb->s_timer[SPPT_PERSIST] = 0;
867 span = cb->s_seq - cb->s_rack;
868 len = min(span, win) - off;
872 * Window shrank after we went into it.
873 * If window shrank to 0, cancel pending
874 * restransmission and pull s_snxt back
875 * to (closed) window. We will enter persist
876 * state below. If the widndow didn't close completely,
877 * just wait for an ACK.
881 cb->s_timer[SPPT_REXMT] = 0;
882 cb->s_snxt = cb->s_rack;
887 rcv_win = sbspace(&so->so_rcv);
890 * Send if we owe peer an ACK.
892 if (cb->s_oobflags & SF_SOOB) {
894 * must transmit this out of band packet
896 cb->s_oobflags &= ~ SF_SOOB;
898 sppstat.spps_sndurg++;
901 if (cb->s_flags & SF_ACKNOW)
903 if (cb->s_state < TCPS_ESTABLISHED)
906 * Silly window can't happen in spp.
907 * Code from tcp deleted.
912 * Compare available window to amount of window
913 * known to peer (as advertised window less
914 * next expected input.) If the difference is at least two
915 * packets or at least 35% of the mximum possible window,
916 * then want to send a window update to peer.
919 u_short delta = 1 + cb->s_alo - cb->s_ack;
920 int adv = rcv_win - (delta * cb->s_mtu);
922 if ((so->so_rcv.sb_cc == 0 && adv >= (2 * cb->s_mtu)) ||
923 (100 * adv / so->so_rcv.sb_hiwat >= 35)) {
924 sppstat.spps_sndwinup++;
925 cb->s_flags |= SF_ACKNOW;
931 * Many comments from tcp_output.c are appropriate here
933 * If send window is too small, there is data to transmit, and no
934 * retransmit or persist is pending, then go to persist state.
935 * If nothing happens soon, send when timer expires:
936 * if window is nonzero, transmit what we can,
937 * otherwise send a probe.
939 if (so->so_snd.sb_cc && cb->s_timer[SPPT_REXMT] == 0 &&
940 cb->s_timer[SPPT_PERSIST] == 0) {
945 * No reason to send a packet, just return.
952 * Find requested packet.
956 cb->s_want = cb->s_snxt;
957 for (m = sb->sb_mb; m; m = m->m_act) {
958 si = mtod(m, struct spidp *);
959 if (SSEQ_LEQ(cb->s_snxt, si->si_seq))
964 if (si->si_seq == cb->s_snxt)
967 sppstat.spps_sndvoid++, si = 0;
975 alo = cb->s_ack - 1 + (rcv_win / ((short)cb->s_mtu));
976 if (SSEQ_LT(alo, cb->s_alo))
981 * must make a copy of this packet for
982 * idp_output to monkey with
984 m = m_copy(dtom(si), 0, (int)M_COPYALL);
988 si = mtod(m, struct spidp *);
989 if (SSEQ_LT(si->si_seq, cb->s_smax))
990 sppstat.spps_sndrexmitpack++;
992 sppstat.spps_sndpack++;
993 } else if (cb->s_force || cb->s_flags & SF_ACKNOW) {
995 * Must send an acknowledgement or a probe
998 sppstat.spps_sndprobe++;
999 if (cb->s_flags & SF_ACKNOW)
1000 sppstat.spps_sndacks++;
1001 m = m_gethdr(M_DONTWAIT, MT_HEADER);
1005 * Fill in mbuf with extended SP header
1006 * and addresses and length put into network format.
1008 MH_ALIGN(m, sizeof (struct spidp));
1009 m->m_len = sizeof (*si);
1010 m->m_pkthdr.len = sizeof (*si);
1011 si = mtod(m, struct spidp *);
1012 si->si_i = *cb->s_idp;
1013 si->si_s = cb->s_shdr;
1014 si->si_seq = cb->s_smax + 1;
1015 si->si_len = htons(sizeof (*si));
1019 if (so->so_options & SO_DEBUG || traceallspps)
1020 spp_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1024 * Stuff checksum and output datagram.
1026 if ((si->si_cc & SP_SP) == 0) {
1027 if (cb->s_force != (1 + SPPT_PERSIST) ||
1028 cb->s_timer[SPPT_PERSIST] == 0) {
1030 * If this is a new packet and we are not currently
1031 * timing anything, time this one.
1033 if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1034 cb->s_smax = si->si_seq;
1035 if (cb->s_rtt == 0) {
1036 sppstat.spps_segstimed++;
1037 cb->s_rtseq = si->si_seq;
1042 * Set rexmt timer if not currently set,
1043 * Initial value for retransmit timer is smoothed
1044 * round-trip time + 2 * round-trip time variance.
1045 * Initialize shift counter which is used for backoff
1046 * of retransmit time.
1048 if (cb->s_timer[SPPT_REXMT] == 0 &&
1049 cb->s_snxt != cb->s_rack) {
1050 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
1051 if (cb->s_timer[SPPT_PERSIST]) {
1052 cb->s_timer[SPPT_PERSIST] = 0;
1056 } else if (SSEQ_LT(cb->s_smax, si->si_seq)) {
1057 cb->s_smax = si->si_seq;
1059 } else if (cb->s_state < TCPS_ESTABLISHED) {
1061 cb->s_rtt = 1; /* Time initial handshake */
1062 if (cb->s_timer[SPPT_REXMT] == 0)
1063 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
1067 * Do not request acks when we ack their data packets or
1068 * when we do a gratuitous window update.
1070 if (((si->si_cc & SP_SP) == 0) || cb->s_force)
1072 si->si_seq = htons(si->si_seq);
1073 si->si_alo = htons(alo);
1074 si->si_ack = htons(cb->s_ack);
1078 len = ntohs(si->si_len);
1081 si->si_sum = ns_cksum(m, len);
1083 si->si_sum = 0xffff;
1086 if (so->so_options & SO_DEBUG || traceallspps)
1087 spp_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
1089 if (so->so_options & SO_DONTROUTE)
1090 error = ns_output(m, (struct route *)0, NS_ROUTETOIF);
1092 error = ns_output(m, &cb->s_nspcb->nsp_route, 0);
1097 sppstat.spps_sndtotal++;
1099 * Data sent (as far as we can tell).
1100 * If this advertises a larger window than any other segment,
1101 * then remember the size of the advertized window.
1102 * Any pending ACK has now been sent.
1105 cb->s_flags &= ~(SF_ACKNOW|SF_DELACK);
1106 if (SSEQ_GT(alo, cb->s_alo))
1114 int spp_do_persist_panics = 0;
1117 register struct sppcb *cb;
1119 register t = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1120 extern int spp_backoff[];
1122 if (cb->s_timer[SPPT_REXMT] && spp_do_persist_panics)
1123 panic("spp_output REXMT");
1125 * Start/restart persistance timer.
1127 SPPT_RANGESET(cb->s_timer[SPPT_PERSIST],
1128 t*spp_backoff[cb->s_rxtshift],
1129 SPPTV_PERSMIN, SPPTV_PERSMAX);
1130 if (cb->s_rxtshift < SPP_MAXRXTSHIFT)
1134 spp_ctloutput(req, so, level, name, value)
1138 struct mbuf **value;
1140 register struct mbuf *m;
1141 struct nspcb *nsp = sotonspcb(so);
1142 register struct sppcb *cb;
1143 int mask, error = 0;
1145 if (level != NSPROTO_SPP) {
1146 /* This will have to be changed when we do more general
1147 stacking of protocols */
1148 return (idp_ctloutput(req, so, level, name, value));
1154 cb = nstosppcb(nsp);
1161 m = m_get(M_DONTWAIT, MT_DATA);
1166 case SO_HEADERS_ON_INPUT:
1170 case SO_HEADERS_ON_OUTPUT:
1173 m->m_len = sizeof(short);
1174 *mtod(m, short *) = cb->s_flags & mask;
1178 m->m_len = sizeof(u_short);
1179 *mtod(m, short *) = cb->s_mtu;
1182 case SO_LAST_HEADER:
1183 m->m_len = sizeof(struct sphdr);
1184 *mtod(m, struct sphdr *) = cb->s_rhdr;
1187 case SO_DEFAULT_HEADERS:
1188 m->m_len = sizeof(struct spidp);
1189 *mtod(m, struct sphdr *) = cb->s_shdr;
1199 if (value == 0 || *value == 0) {
1206 case SO_HEADERS_ON_INPUT:
1210 case SO_HEADERS_ON_OUTPUT:
1213 if (cb->s_flags & SF_PI) {
1214 ok = mtod(*value, int *);
1216 cb->s_flags |= mask;
1218 cb->s_flags &= ~mask;
1219 } else error = EINVAL;
1223 cb->s_mtu = *(mtod(*value, u_short *));
1228 ok = mtod(*value, int *);
1230 cb->s_flags2 |= SF_NEWCALL;
1233 cb->s_flags2 &= ~SF_NEWCALL;
1239 case SO_DEFAULT_HEADERS:
1241 register struct sphdr *sp
1242 = mtod(*value, struct sphdr *);
1243 cb->s_dt = sp->sp_dt;
1244 cb->s_cc = sp->sp_cc & SP_EM;
1259 spp_usrreq(so, req, m, nam, controlp)
1262 struct mbuf *m, *nam, *controlp;
1264 struct nspcb *nsp = sotonspcb(so);
1265 register struct sppcb *cb;
1267 int error = 0, ostate;
1269 register struct sockbuf *sb;
1271 if (req == PRU_CONTROL)
1272 return (ns_control(so, (int)m, (caddr_t)nam,
1273 (struct ifnet *)controlp));
1275 if (req != PRU_ATTACH) {
1280 cb = nstosppcb(nsp);
1282 ostate = cb ? cb->s_state : 0;
1291 error = ns_pcballoc(so, &nspcb);
1294 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1295 error = soreserve(so, (u_long) 3072, (u_long) 3072);
1299 nsp = sotonspcb(so);
1301 mm = m_getclr(M_DONTWAIT, MT_PCB);
1308 cb = mtod(mm, struct sppcb *);
1309 mm = m_getclr(M_DONTWAIT, MT_HEADER);
1311 (void) m_free(dtom(m));
1315 cb->s_idp = mtod(mm, struct idp *);
1316 cb->s_state = TCPS_LISTEN;
1319 cb->s_q.si_next = cb->s_q.si_prev = &cb->s_q;
1321 cb->s_mtu = 576 - sizeof (struct spidp);
1322 cb->s_cwnd = sbspace(sb) * CUNIT / cb->s_mtu;
1323 cb->s_ssthresh = cb->s_cwnd;
1324 cb->s_cwmx = sbspace(sb) * CUNIT /
1325 (2 * sizeof (struct spidp));
1326 /* Above is recomputed when connecting to account
1327 for changed buffering or mtu's */
1328 cb->s_rtt = SPPTV_SRTTBASE;
1329 cb->s_rttvar = SPPTV_SRTTDFLT << 2;
1330 SPPT_RANGESET(cb->s_rxtcur,
1331 ((SPPTV_SRTTBASE >> 2) + (SPPTV_SRTTDFLT << 2)) >> 1,
1332 SPPTV_MIN, SPPTV_REXMTMAX);
1333 nsp->nsp_pcb = (caddr_t) cb;
1341 if (cb->s_state > TCPS_LISTEN)
1342 cb = spp_disconnect(cb);
1348 error = ns_pcbbind(nsp, nam);
1352 if (nsp->nsp_lport == 0)
1353 error = ns_pcbbind(nsp, (struct mbuf *)0);
1355 cb->s_state = TCPS_LISTEN;
1359 * Initiate connection to peer.
1360 * Enter SYN_SENT state, and mark socket as connecting.
1361 * Start keep-alive timer, setup prototype header,
1362 * Send initial system packet requesting connection.
1365 if (nsp->nsp_lport == 0) {
1366 error = ns_pcbbind(nsp, (struct mbuf *)0);
1370 error = ns_pcbconnect(nsp, nam);
1374 sppstat.spps_connattempt++;
1375 cb->s_state = TCPS_SYN_SENT;
1378 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
1379 cb->s_force = 1 + SPPTV_KEEP;
1381 * Other party is required to respond to
1382 * the port I send from, but he is not
1383 * required to answer from where I am sending to,
1384 * so allow wildcarding.
1385 * original port I am sending to is still saved in
1389 error = spp_output(cb, (struct mbuf *) 0);
1397 * We may decide later to implement connection closing
1398 * handshaking at the spp level optionally.
1399 * here is the hook to do it:
1401 case PRU_DISCONNECT:
1402 cb = spp_disconnect(cb);
1406 * Accept a connection. Essentially all the work is
1407 * done at higher levels; just return the address
1408 * of the peer, storing through addr.
1411 struct sockaddr_ns *sns = mtod(nam, struct sockaddr_ns *);
1413 nam->m_len = sizeof (struct sockaddr_ns);
1414 sns->sns_family = AF_NS;
1415 sns->sns_addr = nsp->nsp_faddr;
1421 cb = spp_usrclosed(cb);
1423 error = spp_output(cb, (struct mbuf *) 0);
1427 * After a receive, possibly send acknowledgment
1428 * updating allocation.
1431 cb->s_flags |= SF_RVD;
1432 (void) spp_output(cb, (struct mbuf *) 0);
1433 cb->s_flags &= ~SF_RVD;
1437 (void) spp_drop(cb, ECONNABORTED);
1447 if ((cb->s_oobflags & SF_IOOB) || so->so_oobmark ||
1448 (so->so_state & SS_RCVATMARK)) {
1450 *mtod(m, caddr_t) = cb->s_iobc;
1457 if (sbspace(&so->so_snd) < -512) {
1461 cb->s_oobflags |= SF_SOOB;
1465 u_short *p = mtod(controlp, u_short *);
1467 if ((p[0] == 5) && p[1] == 1) { /* XXXX, for testing */
1468 cb->s_shdr.sp_dt = *(u_char *)(&p[2]);
1474 error = spp_output(cb, m);
1479 ns_setsockaddr(nsp, nam);
1483 ns_setpeeraddr(nsp, nam);
1487 cb = spp_timers(cb, (int)nam);
1488 req |= ((int)nam) << 8;
1500 if (cb && (so->so_options & SO_DEBUG || traceallspps))
1501 spp_trace(SA_USER, (u_char)ostate, cb, (struct spidp *)0, req);
1503 if (controlp != NULL)
1511 spp_usrreq_sp(so, req, m, nam, controlp)
1514 struct mbuf *m, *nam, *controlp;
1516 int error = spp_usrreq(so, req, m, nam, controlp);
1518 if (req == PRU_ATTACH && error == 0) {
1519 struct nspcb *nsp = sotonspcb(so);
1520 ((struct sppcb *)nsp->nsp_pcb)->s_flags |=
1521 (SF_HI | SF_HO | SF_PI);
1527 * Create template to be used to send spp packets on a connection.
1528 * Called after host entry created, fills
1529 * in a skeletal spp header (choosing connection id),
1530 * minimizing the amount of work necessary when the connection is used.
1533 register struct sppcb *cb;
1535 register struct nspcb *nsp = cb->s_nspcb;
1536 register struct idp *idp = cb->s_idp;
1537 register struct sockbuf *sb = &(nsp->nsp_socket->so_snd);
1539 idp->idp_pt = NSPROTO_SPP;
1540 idp->idp_sna = nsp->nsp_laddr;
1541 idp->idp_dna = nsp->nsp_faddr;
1542 cb->s_sid = htons(spp_iss);
1543 spp_iss += SPP_ISSINCR/2;
1545 cb->s_cwnd = (sbspace(sb) * CUNIT) / cb->s_mtu;
1546 cb->s_ssthresh = cb->s_cwnd; /* Try to expand fast to full complement
1548 cb->s_cwmx = (sbspace(sb) * CUNIT) / (2 * sizeof(struct spidp));
1549 cb->s_cwmx = max(cb->s_cwmx, cb->s_cwnd);
1550 /* But allow for lots of little packets as well */
1554 * Close a SPIP control block:
1555 * discard spp control block itself
1556 * discard ns protocol control block
1557 * wake up any sleepers
1561 register struct sppcb *cb;
1563 register struct spidp_q *s;
1564 struct nspcb *nsp = cb->s_nspcb;
1565 struct socket *so = nsp->nsp_socket;
1566 register struct mbuf *m;
1568 s = cb->s_q.si_next;
1569 while (s != &(cb->s_q)) {
1571 m = dtom(s->si_prev);
1575 (void) m_free(dtom(cb->s_idp));
1576 (void) m_free(dtom(cb));
1578 soisdisconnected(so);
1580 sppstat.spps_closed++;
1581 return ((struct sppcb *)0);
1584 * Someday we may do level 3 handshaking
1585 * to close a connection or send a xerox style error.
1586 * For now, just close.
1590 register struct sppcb *cb;
1592 return (spp_close(cb));
1596 register struct sppcb *cb;
1598 return (spp_close(cb));
1601 * Drop connection, reporting
1602 * the specified error.
1606 register struct sppcb *cb;
1609 struct socket *so = cb->s_nspcb->nsp_socket;
1612 * someday, in the xerox world
1613 * we will generate error protocol packets
1614 * announcing that the socket has gone away.
1616 if (TCPS_HAVERCVDSYN(cb->s_state)) {
1617 sppstat.spps_drops++;
1618 cb->s_state = TCPS_CLOSED;
1619 /*(void) tcp_output(cb);*/
1621 sppstat.spps_conndrops++;
1622 so->so_error = errno;
1623 return (spp_close(cb));
1630 (void) spp_close((struct sppcb *)nsp->nsp_pcb);
1633 int spp_backoff[SPP_MAXRXTSHIFT+1] =
1634 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
1636 * Fast timeout routine for processing delayed acks
1640 register struct nspcb *nsp;
1641 register struct sppcb *cb;
1644 nsp = nspcb.nsp_next;
1646 for (; nsp != &nspcb; nsp = nsp->nsp_next)
1647 if ((cb = (struct sppcb *)nsp->nsp_pcb) &&
1648 (cb->s_flags & SF_DELACK)) {
1649 cb->s_flags &= ~SF_DELACK;
1650 cb->s_flags |= SF_ACKNOW;
1651 sppstat.spps_delack++;
1652 (void) spp_output(cb, (struct mbuf *) 0);
1658 * spp protocol timeout routine called every 500 ms.
1659 * Updates the timers in all active pcb's and
1660 * causes finite state machine actions if timers expire.
1664 register struct nspcb *ip, *ipnxt;
1665 register struct sppcb *cb;
1670 * Search through tcb's and update active timers.
1672 ip = nspcb.nsp_next;
1677 while (ip != &nspcb) {
1679 ipnxt = ip->nsp_next;
1682 for (i = 0; i < SPPT_NTIMERS; i++) {
1683 if (cb->s_timer[i] && --cb->s_timer[i] == 0) {
1684 (void) spp_usrreq(cb->s_nspcb->nsp_socket,
1685 PRU_SLOWTIMO, (struct mbuf *)0,
1686 (struct mbuf *)i, (struct mbuf *)0,
1688 if (ipnxt->nsp_prev != ip)
1698 spp_iss += SPP_ISSINCR/PR_SLOWHZ; /* increment iss */
1702 * SPP timer processing.
1705 spp_timers(cb, timer)
1706 register struct sppcb *cb;
1712 cb->s_force = 1 + timer;
1716 * 2 MSL timeout in shutdown went off. TCP deletes connection
1720 printf("spp: SPPT_2MSL went off for no reason\n");
1721 cb->s_timer[timer] = 0;
1725 * Retransmission timer went off. Message has not
1726 * been acked within retransmit interval. Back off
1727 * to a longer retransmit interval and retransmit one packet.
1730 if (++cb->s_rxtshift > SPP_MAXRXTSHIFT) {
1731 cb->s_rxtshift = SPP_MAXRXTSHIFT;
1732 sppstat.spps_timeoutdrop++;
1733 cb = spp_drop(cb, ETIMEDOUT);
1736 sppstat.spps_rexmttimeo++;
1737 rexmt = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1738 rexmt *= spp_backoff[cb->s_rxtshift];
1739 SPPT_RANGESET(cb->s_rxtcur, rexmt, SPPTV_MIN, SPPTV_REXMTMAX);
1740 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
1742 * If we have backed off fairly far, our srtt
1743 * estimate is probably bogus. Clobber it
1744 * so we'll take the next rtt measurement as our srtt;
1745 * move the current srtt into rttvar to keep the current
1746 * retransmit times until then.
1748 if (cb->s_rxtshift > SPP_MAXRXTSHIFT / 4 ) {
1749 cb->s_rttvar += (cb->s_srtt >> 2);
1752 cb->s_snxt = cb->s_rack;
1754 * If timing a packet, stop the timer.
1758 * See very long discussion in tcp_timer.c about congestion
1759 * window and sstrhesh
1761 win = min(cb->s_swnd, (cb->s_cwnd/CUNIT)) / 2;
1765 cb->s_ssthresh = win * CUNIT;
1766 (void) spp_output(cb, (struct mbuf *) 0);
1770 * Persistance timer into zero window.
1771 * Force a probe to be sent.
1774 sppstat.spps_persisttimeo++;
1776 (void) spp_output(cb, (struct mbuf *) 0);
1780 * Keep-alive timer went off; send something
1781 * or drop connection if idle for too long.
1784 sppstat.spps_keeptimeo++;
1785 if (cb->s_state < TCPS_ESTABLISHED)
1787 if (cb->s_nspcb->nsp_socket->so_options & SO_KEEPALIVE) {
1788 if (cb->s_idle >= SPPTV_MAXIDLE)
1790 sppstat.spps_keepprobe++;
1791 (void) spp_output(cb, (struct mbuf *) 0);
1794 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
1797 sppstat.spps_keepdrops++;
1798 cb = spp_drop(cb, ETIMEDOUT);
1804 int SppcbSize = sizeof (struct sppcb);
1805 int NspcbSize = sizeof (struct nspcb);