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 1.10 2004/06/04 01:46:49 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>
58 #include "spp_timer.h"
60 #include "spp_debug.h"
62 extern u_char nsctlerrmap[]; /* from ns_input.c */
63 extern int idpcksum; /* from ns_input.c */
65 struct spp_istat spp_istat;
67 int spp_backoff[SPP_MAXRXTSHIFT+1] =
68 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
71 * SP protocol implementation.
77 spp_iss = 1; /* WRONG !! should fish it out of TODR */
79 struct spidp spp_savesi;
81 extern int sppconsdebug;
83 int spp_use_delack = 0;
84 u_short spp_newchecks[50];
88 spp_input(struct mbuf *m, struct nspcb *nsp)
91 struct spidp *si = mtod(m, struct spidp *);
97 sppstat.spps_rcvtotal++;
99 panic("No nspcb in spp_input");
104 if (cb == 0) goto bad;
106 if (m->m_len < sizeof(*si)) {
107 if ((m = m_pullup(m, sizeof(*si))) == 0) {
108 sppstat.spps_rcvshort++;
111 si = mtod(m, struct spidp *);
113 si->si_seq = ntohs(si->si_seq);
114 si->si_ack = ntohs(si->si_ack);
115 si->si_alo = ntohs(si->si_alo);
117 so = nsp->nsp_socket;
118 if (so->so_options & SO_DEBUG || traceallspps) {
119 ostate = cb->s_state;
122 if (so->so_options & SO_ACCEPTCONN) {
123 struct sppcb *ocb = cb;
125 so = sonewconn(so, 0);
130 * This is ugly, but ....
132 * Mark socket as temporary until we're
133 * committed to keeping it. The code at
134 * ``drop'' and ``dropwithreset'' check the
135 * flag dropsocket to see if the temporary
136 * socket created here should be discarded.
137 * We mark the socket as discardable until
138 * we're committed to it below in TCPS_LISTEN.
141 nsp = (struct nspcb *)so->so_pcb;
142 nsp->nsp_laddr = si->si_dna;
144 cb->s_mtu = ocb->s_mtu; /* preserve sockopts */
145 cb->s_flags = ocb->s_flags; /* preserve sockopts */
146 cb->s_flags2 = ocb->s_flags2; /* preserve sockopts */
147 cb->s_state = TCPS_LISTEN;
151 * Packet received on connection.
152 * reset idle time and keep-alive timer;
155 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
157 switch (cb->s_state) {
161 struct sockaddr_ns *sns;
162 struct ns_addr laddr;
165 * If somebody here was carying on a conversation
166 * and went away, and his pen pal thinks he can
167 * still talk, we get the misdirected packet.
169 if (spp_hardnosed && (si->si_did != 0 || si->si_seq != 0)) {
173 am = m_get(MB_DONTWAIT, MT_SONAME);
176 am->m_len = sizeof (struct sockaddr_ns);
177 sns = mtod(am, struct sockaddr_ns *);
178 sns->sns_len = sizeof(*sns);
179 sns->sns_family = AF_NS;
180 sns->sns_addr = si->si_sna;
181 laddr = nsp->nsp_laddr;
182 if (ns_nullhost(laddr))
183 nsp->nsp_laddr = si->si_dna;
184 if (ns_pcbconnect(nsp, am)) {
185 nsp->nsp_laddr = laddr;
192 dropsocket = 0; /* committed to socket */
193 cb->s_did = si->si_sid;
194 cb->s_rack = si->si_ack;
195 cb->s_ralo = si->si_alo;
196 #define THREEWAYSHAKE
198 cb->s_state = TCPS_SYN_RECEIVED;
199 cb->s_force = 1 + SPPT_KEEP;
200 sppstat.spps_accepts++;
201 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
205 * This state means that we have heard a response
206 * to our acceptance of their connection
207 * It is probably logically unnecessary in this
210 case TCPS_SYN_RECEIVED: {
211 if (si->si_did!=cb->s_sid) {
216 nsp->nsp_fport = si->si_sport;
217 cb->s_timer[SPPT_REXMT] = 0;
218 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
220 cb->s_state = TCPS_ESTABLISHED;
221 sppstat.spps_accepts++;
226 * This state means that we have gotten a response
227 * to our attempt to establish a connection.
228 * We fill in the data from the other side,
229 * telling us which port to respond to, instead of the well-
230 * known one we might have sent to in the first place.
231 * We also require that this is a response to our
235 if (si->si_did!=cb->s_sid) {
239 sppstat.spps_connects++;
240 cb->s_did = si->si_sid;
241 cb->s_rack = si->si_ack;
242 cb->s_ralo = si->si_alo;
243 cb->s_dport = nsp->nsp_fport = si->si_sport;
244 cb->s_timer[SPPT_REXMT] = 0;
245 cb->s_flags |= SF_ACKNOW;
247 cb->s_state = TCPS_ESTABLISHED;
248 /* Use roundtrip time of connection request for initial rtt */
250 cb->s_srtt = cb->s_rtt << 3;
251 cb->s_rttvar = cb->s_rtt << 1;
252 SPPT_RANGESET(cb->s_rxtcur,
253 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
254 SPPTV_MIN, SPPTV_REXMTMAX);
258 if (so->so_options & SO_DEBUG || traceallspps)
259 spp_trace(SA_INPUT, (u_char)ostate, cb, &spp_savesi, 0);
261 m->m_len -= sizeof (struct idp);
262 m->m_pkthdr.len -= sizeof (struct idp);
263 m->m_data += sizeof (struct idp);
265 if (spp_reass(cb, si)) {
268 if (cb->s_force || (cb->s_flags & (SF_ACKNOW|SF_WIN|SF_RXT)))
269 (void) spp_output(cb, (struct mbuf *)0);
270 cb->s_flags &= ~(SF_WIN|SF_RXT);
276 si->si_seq = ntohs(si->si_seq);
277 si->si_ack = ntohs(si->si_ack);
278 si->si_alo = ntohs(si->si_alo);
279 ns_error(dtom(si), NS_ERR_NOSOCK, 0);
280 if (cb->s_nspcb->nsp_socket->so_options & SO_DEBUG || traceallspps)
281 spp_trace(SA_DROP, (u_char)ostate, cb, &spp_savesi, 0);
286 if (cb == 0 || cb->s_nspcb->nsp_socket->so_options & SO_DEBUG ||
288 spp_trace(SA_DROP, (u_char)ostate, cb, &spp_savesi, 0);
292 int spprexmtthresh = 3;
295 * This is structurally similar to the tcp reassembly routine
296 * but its function is somewhat different: It merely queues
297 * packets up, and suppresses duplicates.
300 spp_reass(struct sppcb *cb, struct spidp *si)
304 struct socket *so = cb->s_nspcb->nsp_socket;
305 char packetp = cb->s_flags & SF_HI;
312 * Update our news from them.
314 if (si->si_cc & SP_SA)
315 cb->s_flags |= (spp_use_delack ? SF_DELACK : SF_ACKNOW);
316 if (SSEQ_GT(si->si_alo, cb->s_ralo))
317 cb->s_flags |= SF_WIN;
318 if (SSEQ_LEQ(si->si_ack, cb->s_rack)) {
319 if ((si->si_cc & SP_SP) && cb->s_rack != (cb->s_smax + 1)) {
320 sppstat.spps_rcvdupack++;
322 * If this is a completely duplicate ack
323 * and other conditions hold, we assume
324 * a packet has been dropped and retransmit
325 * it exactly as in tcp_input().
327 if (si->si_ack != cb->s_rack ||
328 si->si_alo != cb->s_ralo)
330 else if (++cb->s_dupacks == spprexmtthresh) {
331 u_short onxt = cb->s_snxt;
332 int cwnd = cb->s_cwnd;
334 cb->s_snxt = si->si_ack;
336 cb->s_force = 1 + SPPT_REXMT;
337 (void) spp_output(cb, (struct mbuf *)0);
338 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
340 if (cwnd >= 4 * CUNIT)
341 cb->s_cwnd = cwnd / 2;
342 if (SSEQ_GT(onxt, cb->s_snxt))
352 * If our correspondent acknowledges data we haven't sent
353 * TCP would drop the packet after acking. We'll be a little
356 if (SSEQ_GT(si->si_ack, (cb->s_smax + 1))) {
357 sppstat.spps_rcvacktoomuch++;
358 si->si_ack = cb->s_smax + 1;
360 sppstat.spps_rcvackpack++;
362 * If transmit timer is running and timed sequence
363 * number was acked, update smoothed round trip time.
364 * See discussion of algorithm in tcp_input.c
366 if (cb->s_rtt && SSEQ_GT(si->si_ack, cb->s_rtseq)) {
367 sppstat.spps_rttupdated++;
368 if (cb->s_srtt != 0) {
370 delta = cb->s_rtt - (cb->s_srtt >> 3);
371 if ((cb->s_srtt += delta) <= 0)
375 delta -= (cb->s_rttvar >> 2);
376 if ((cb->s_rttvar += delta) <= 0)
380 * No rtt measurement yet
382 cb->s_srtt = cb->s_rtt << 3;
383 cb->s_rttvar = cb->s_rtt << 1;
387 SPPT_RANGESET(cb->s_rxtcur,
388 ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1,
389 SPPTV_MIN, SPPTV_REXMTMAX);
392 * If all outstanding data is acked, stop retransmit
393 * timer and remember to restart (more output or persist).
394 * If there is more data to be acked, restart retransmit
395 * timer, using current (possibly backed-off) value;
397 if (si->si_ack == cb->s_smax + 1) {
398 cb->s_timer[SPPT_REXMT] = 0;
399 cb->s_flags |= SF_RXT;
400 } else if (cb->s_timer[SPPT_PERSIST] == 0)
401 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
403 * When new data is acked, open the congestion window.
404 * If the window gives us less than ssthresh packets
405 * in flight, open exponentially (maxseg at a time).
406 * Otherwise open linearly (maxseg^2 / cwnd at a time).
409 if (cb->s_cwnd > cb->s_ssthresh)
410 incr = max(incr * incr / cb->s_cwnd, 1);
411 cb->s_cwnd = min(cb->s_cwnd + incr, cb->s_cwmx);
413 * Trim Acked data from output queue.
415 while ((m = so->so_snd.sb_mb) != NULL) {
416 if (SSEQ_LT((mtod(m, struct spidp *))->si_seq, si->si_ack))
417 sbdroprecord(&so->so_snd);
422 cb->s_rack = si->si_ack;
424 if (SSEQ_LT(cb->s_snxt, cb->s_rack))
425 cb->s_snxt = cb->s_rack;
426 if (SSEQ_LT(cb->s_swl1, si->si_seq) || (cb->s_swl1 == si->si_seq &&
427 (SSEQ_LT(cb->s_swl2, si->si_ack) ||
428 (cb->s_swl2 == si->si_ack && SSEQ_LT(cb->s_ralo, si->si_alo))))) {
429 /* keep track of pure window updates */
430 if ((si->si_cc & SP_SP) && cb->s_swl2 == si->si_ack
431 && SSEQ_LT(cb->s_ralo, si->si_alo)) {
432 sppstat.spps_rcvwinupd++;
433 sppstat.spps_rcvdupack--;
435 cb->s_ralo = si->si_alo;
436 cb->s_swl1 = si->si_seq;
437 cb->s_swl2 = si->si_ack;
438 cb->s_swnd = (1 + si->si_alo - si->si_ack);
439 if (cb->s_swnd > cb->s_smxw)
440 cb->s_smxw = cb->s_swnd;
441 cb->s_flags |= SF_WIN;
444 * If this packet number is higher than that which
445 * we have allocated refuse it, unless urgent
447 if (SSEQ_GT(si->si_seq, cb->s_alo)) {
448 if (si->si_cc & SP_SP) {
449 sppstat.spps_rcvwinprobe++;
452 sppstat.spps_rcvpackafterwin++;
453 if (si->si_cc & SP_OB) {
454 if (SSEQ_GT(si->si_seq, cb->s_alo + 60)) {
455 ns_error(dtom(si), NS_ERR_FULLUP, 0);
457 } /* else queue this packet; */
459 /*register struct socket *so = cb->s_nspcb->nsp_socket;
460 if (so->so_state && SS_NOFDREF) {
461 ns_error(dtom(si), NS_ERR_NOSOCK, 0);
466 ns_error(dtom(si), NS_ERR_FULLUP, 0);
471 * If this is a system packet, we don't need to
472 * queue it up, and won't update acknowledge #
474 if (si->si_cc & SP_SP) {
478 * We have already seen this packet, so drop.
480 if (SSEQ_LT(si->si_seq, cb->s_ack)) {
482 sppstat.spps_rcvduppack++;
483 if (si->si_seq == cb->s_ack - 1)
488 * Loop through all packets queued up to insert in
489 * appropriate sequence.
491 for (q = cb->s_q.si_next; q!=&cb->s_q; q = q->si_next) {
492 if (si->si_seq == SI(q)->si_seq) {
493 sppstat.spps_rcvduppack++;
496 if (SSEQ_LT(si->si_seq, SI(q)->si_seq)) {
497 sppstat.spps_rcvoopack++;
501 insque(si, q->si_prev);
503 * If this packet is urgent, inform process
505 if (si->si_cc & SP_OB) {
506 cb->s_iobc = ((char *)si)[1 + sizeof(*si)];
508 cb->s_oobflags |= SF_IOOB;
511 #define SPINC sizeof(struct sphdr)
513 * Loop through all packets queued up to update acknowledge
514 * number, and present all acknowledged data to user;
515 * If in packet interface mode, show packet headers.
517 for (q = cb->s_q.si_next; q!=&cb->s_q; q = q->si_next) {
518 if (SI(q)->si_seq == cb->s_ack) {
521 if (SI(q)->si_cc & SP_OB) {
522 cb->s_oobflags &= ~SF_IOOB;
523 if (so->so_rcv.sb_cc)
524 so->so_oobmark = so->so_rcv.sb_cc;
526 so->so_state |= SS_RCVATMARK;
531 sppstat.spps_rcvpack++;
533 if (cb->s_flags2 & SF_NEWCALL) {
534 struct sphdr *sp = mtod(m, struct sphdr *);
535 u_char dt = sp->sp_dt;
537 if (dt != cb->s_rhdr.sp_dt) {
539 m_getclr(MB_DONTWAIT, MT_CONTROL);
544 cb->s_rhdr.sp_dt = dt;
545 mm->m_len = 5; /*XXX*/
548 *(u_char *)(&s[2]) = dt;
549 sbappend(&so->so_rcv, mm);
552 if (sp->sp_cc & SP_OB) {
553 MCHTYPE(m, MT_OOBDATA);
556 so->so_state &= ~SS_RCVATMARK;
561 m->m_pkthdr.len -= SPINC;
563 if ((sp->sp_cc & SP_EM) || packetp) {
564 sbappendrecord(&so->so_rcv, m);
567 sbappend(&so->so_rcv, m);
571 sbappendrecord(&so->so_rcv, m);
573 cb->s_rhdr = *mtod(m, struct sphdr *);
576 m->m_pkthdr.len -= SPINC;
577 sbappend(&so->so_rcv, m);
582 if (wakeup) sorwakeup(so);
587 spp_ctlinput(int cmd, caddr_t arg)
590 struct ns_errp *errp = 0;
592 struct sockaddr_ns *sns;
595 if (cmd < 0 || cmd > PRC_NCMDS)
597 type = NS_ERR_UNREACH_HOST;
606 case PRC_HOSTUNREACH:
607 sns = (struct sockaddr_ns *)arg;
608 if (sns->sns_family != AF_NS)
614 errp = (struct ns_errp *)arg;
615 na = &errp->ns_err_idp.idp_dna;
616 type = errp->ns_err_num;
617 type = ntohs((u_short)type);
621 case NS_ERR_UNREACH_HOST:
622 ns_pcbnotify(na, (int)nsctlerrmap[cmd], spp_abort, (long) 0);
627 nsp = ns_pcblookup(na, errp->ns_err_idp.idp_sna.x_port,
631 (void) spp_drop((struct sppcb *)nsp->nsp_pcb,
632 (int)nsctlerrmap[cmd]);
634 (void) idp_drop(nsp, (int)nsctlerrmap[cmd]);
639 ns_pcbnotify(na, 0, spp_quench, (long) 0);
643 * When a source quench is received, close congestion window
644 * to one packet. We will gradually open it again as we proceed.
647 spp_quench(struct nspcb *nsp)
649 struct sppcb *cb = nstosppcb(nsp);
657 spp_fixmtu(struct nspcb *nsp)
659 struct sppcb *cb = (struct sppcb *)(nsp->nsp_pcb);
665 struct mbuf *firstbad, *m0;
669 * The notification that we have sent
670 * too much is bad news -- we will
671 * have to go through queued up so far
672 * splitting ones which are too big and
673 * reassigning sequence numbers and checksums.
674 * we should then retransmit all packets from
675 * one above the offending packet to the last one
676 * we had sent (or our allocation)
677 * then the offending one so that the any queued
678 * data at our destination will be discarded.
680 ep = (struct ns_errp *)nsp->nsp_notify_param;
681 sb = &nsp->nsp_socket->so_snd;
682 cb->s_mtu = ep->ns_err_param;
683 badseq = SI(&ep->ns_err_idp)->si_seq;
684 for (m = sb->sb_mb; m; m = m->m_act) {
685 si = mtod(m, struct spidp *);
686 if (si->si_seq == badseq)
692 /* calculate length */
693 for (m0 = m, len = 0; m ; m = m->m_next)
695 if (len > cb->s_mtu) {
704 spp_output(struct sppcb *cb, struct mbuf *m0)
706 struct socket *so = cb->s_nspcb->nsp_socket;
708 struct spidp *si = (struct spidp *) 0;
709 struct sockbuf *sb = &so->so_snd;
710 int len = 0, win, rcv_win;
711 short span, off, recordp = 0;
713 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, MB_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(MB_DONTWAIT, MT_DATA);
784 *(mtod(m1, u_char *)) = 0;
788 m = m_gethdr(MB_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)) {
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(MB_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 spp_setpersist(struct sppcb *cb)
1119 int t = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1121 if (cb->s_timer[SPPT_REXMT] && spp_do_persist_panics)
1122 panic("spp_output REXMT");
1124 * Start/restart persistance timer.
1126 SPPT_RANGESET(cb->s_timer[SPPT_PERSIST],
1127 t*spp_backoff[cb->s_rxtshift],
1128 SPPTV_PERSMIN, SPPTV_PERSMAX);
1129 if (cb->s_rxtshift < SPP_MAXRXTSHIFT)
1134 spp_ctloutput(int req, struct socket *so, int level,
1135 int name, struct mbuf **value)
1138 struct nspcb *nsp = sotonspcb(so);
1140 int mask, error = 0;
1142 if (level != NSPROTO_SPP) {
1143 /* This will have to be changed when we do more general
1144 stacking of protocols */
1145 return (idp_ctloutput(req, so, level, name, value));
1151 cb = nstosppcb(nsp);
1158 m = m_get(MB_DONTWAIT, MT_DATA);
1163 case SO_HEADERS_ON_INPUT:
1167 case SO_HEADERS_ON_OUTPUT:
1170 m->m_len = sizeof(short);
1171 *mtod(m, short *) = cb->s_flags & mask;
1175 m->m_len = sizeof(u_short);
1176 *mtod(m, short *) = cb->s_mtu;
1179 case SO_LAST_HEADER:
1180 m->m_len = sizeof(struct sphdr);
1181 *mtod(m, struct sphdr *) = cb->s_rhdr;
1184 case SO_DEFAULT_HEADERS:
1185 m->m_len = sizeof(struct spidp);
1186 *mtod(m, struct sphdr *) = cb->s_shdr;
1196 if (value == 0 || *value == 0) {
1203 case SO_HEADERS_ON_INPUT:
1207 case SO_HEADERS_ON_OUTPUT:
1210 if (cb->s_flags & SF_PI) {
1211 ok = mtod(*value, int *);
1213 cb->s_flags |= mask;
1215 cb->s_flags &= ~mask;
1216 } else error = EINVAL;
1220 cb->s_mtu = *(mtod(*value, u_short *));
1225 ok = mtod(*value, int *);
1227 cb->s_flags2 |= SF_NEWCALL;
1230 cb->s_flags2 &= ~SF_NEWCALL;
1236 case SO_DEFAULT_HEADERS:
1239 = mtod(*value, struct sphdr *);
1240 cb->s_dt = sp->sp_dt;
1241 cb->s_cc = sp->sp_cc & SP_EM;
1255 #warning "spp_usrreq not converted to FreeBSD usrreq style! watch out!"
1258 spp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam,
1259 struct mbuf *controlp)
1262 #ifdef OBSOLETE /* not converted to FreeBSD usrreq style XXX */
1263 struct nspcb *nsp = sotonspcb(so);
1264 struct sppcb *cb = NULL;
1270 if (req == PRU_CONTROL)
1271 return (ns_control(so, (int)m, (caddr_t)nam,
1272 (struct ifnet *)controlp));
1274 if (req != PRU_ATTACH) {
1279 cb = nstosppcb(nsp);
1281 ostate = cb ? cb->s_state : 0;
1290 error = ns_pcballoc(so, &nspcb);
1293 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1294 error = soreserve(so, (u_long) 3072, (u_long) 3072);
1298 nsp = sotonspcb(so);
1301 mm = m_getclr(MB_DONTWAIT, MT_CONTROL);
1308 cb = mtod(mm, struct sppcb *);
1309 mm = m_getclr(MB_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)
1513 spp_usrreq_sp(struct socket *so, int req, struct mbuf *m,
1514 struct mbuf *nam, struct mbuf *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 spp_template(struct sppcb *cb)
1535 struct nspcb *nsp = cb->s_nspcb;
1536 struct idp *idp = cb->s_idp;
1537 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
1560 spp_close(struct sppcb *cb)
1563 struct nspcb *nsp = cb->s_nspcb;
1564 struct socket *so = nsp->nsp_socket;
1567 s = cb->s_q.si_next;
1568 while (s != &(cb->s_q)) {
1570 m = dtom(s->si_prev);
1574 (void) m_free(dtom(cb->s_idp));
1575 (void) m_free(dtom(cb));
1577 soisdisconnected(so);
1579 sppstat.spps_closed++;
1580 return ((struct sppcb *)0);
1583 * Someday we may do level 3 handshaking
1584 * to close a connection or send a xerox style error.
1585 * For now, just close.
1588 spp_usrclosed(struct sppcb *cb)
1590 return (spp_close(cb));
1594 spp_disconnect(struct sppcb *cb)
1596 return (spp_close(cb));
1600 * Drop connection, reporting
1601 * the specified error.
1604 spp_drop(struct sppcb *cb, int errno)
1606 struct socket *so = cb->s_nspcb->nsp_socket;
1609 * someday, in the xerox world
1610 * we will generate error protocol packets
1611 * announcing that the socket has gone away.
1613 if (TCPS_HAVERCVDSYN(cb->s_state)) {
1614 sppstat.spps_drops++;
1615 cb->s_state = TCPS_CLOSED;
1616 /*(void) tcp_output(cb);*/
1618 sppstat.spps_conndrops++;
1619 so->so_error = errno;
1620 return (spp_close(cb));
1624 spp_abort(struct nspcb *nsp)
1626 spp_close((struct sppcb *)nsp->nsp_pcb);
1630 * Fast timeout routine for processing delayed acks
1639 nsp = nspcb.nsp_next;
1641 for (; nsp != &nspcb; nsp = nsp->nsp_next)
1642 if ((cb = (struct sppcb *)nsp->nsp_pcb) &&
1643 (cb->s_flags & SF_DELACK)) {
1644 cb->s_flags &= ~SF_DELACK;
1645 cb->s_flags |= SF_ACKNOW;
1646 sppstat.spps_delack++;
1647 (void) spp_output(cb, (struct mbuf *) 0);
1653 * spp protocol timeout routine called every 500 ms.
1654 * Updates the timers in all active pcb's and
1655 * causes finite state machine actions if timers expire.
1660 struct nspcb *ip, *ipnxt;
1666 * Search through tcb's and update active timers.
1668 ip = nspcb.nsp_next;
1673 while (ip != &nspcb) {
1675 ipnxt = ip->nsp_next;
1678 for (i = 0; i < SPPT_NTIMERS; i++) {
1679 if (cb->s_timer[i] && --cb->s_timer[i] == 0) {
1680 (void) spp_usrreq(cb->s_nspcb->nsp_socket,
1681 PRU_SLOWTIMO, (struct mbuf *)0,
1682 (struct mbuf *)i, (struct mbuf *)0);
1683 if (ipnxt->nsp_prev != ip)
1693 spp_iss += SPP_ISSINCR/PR_SLOWHZ; /* increment iss */
1697 * SPP timer processing.
1700 spp_timers(struct sppcb *cb, int timer)
1705 cb->s_force = 1 + timer;
1709 * 2 MSL timeout in shutdown went off. TCP deletes connection
1713 printf("spp: SPPT_2MSL went off for no reason\n");
1714 cb->s_timer[timer] = 0;
1718 * Retransmission timer went off. Message has not
1719 * been acked within retransmit interval. Back off
1720 * to a longer retransmit interval and retransmit one packet.
1723 if (++cb->s_rxtshift > SPP_MAXRXTSHIFT) {
1724 cb->s_rxtshift = SPP_MAXRXTSHIFT;
1725 sppstat.spps_timeoutdrop++;
1726 cb = spp_drop(cb, ETIMEDOUT);
1729 sppstat.spps_rexmttimeo++;
1730 rexmt = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
1731 rexmt *= spp_backoff[cb->s_rxtshift];
1732 SPPT_RANGESET(cb->s_rxtcur, rexmt, SPPTV_MIN, SPPTV_REXMTMAX);
1733 cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
1735 * If we have backed off fairly far, our srtt
1736 * estimate is probably bogus. Clobber it
1737 * so we'll take the next rtt measurement as our srtt;
1738 * move the current srtt into rttvar to keep the current
1739 * retransmit times until then.
1741 if (cb->s_rxtshift > SPP_MAXRXTSHIFT / 4 ) {
1742 cb->s_rttvar += (cb->s_srtt >> 2);
1745 cb->s_snxt = cb->s_rack;
1747 * If timing a packet, stop the timer.
1751 * See very long discussion in tcp_timer.c about congestion
1752 * window and sstrhesh
1754 win = min(cb->s_swnd, (cb->s_cwnd/CUNIT)) / 2;
1758 cb->s_ssthresh = win * CUNIT;
1759 (void) spp_output(cb, (struct mbuf *) 0);
1763 * Persistance timer into zero window.
1764 * Force a probe to be sent.
1767 sppstat.spps_persisttimeo++;
1769 (void) spp_output(cb, (struct mbuf *) 0);
1773 * Keep-alive timer went off; send something
1774 * or drop connection if idle for too long.
1777 sppstat.spps_keeptimeo++;
1778 if (cb->s_state < TCPS_ESTABLISHED)
1780 if (cb->s_nspcb->nsp_socket->so_options & SO_KEEPALIVE) {
1781 if (cb->s_idle >= SPPTV_MAXIDLE)
1783 sppstat.spps_keepprobe++;
1784 (void) spp_output(cb, (struct mbuf *) 0);
1787 cb->s_timer[SPPT_KEEP] = SPPTV_KEEP;
1790 sppstat.spps_keepdrops++;
1791 cb = spp_drop(cb, ETIMEDOUT);
1797 int SppcbSize = sizeof (struct sppcb);
1798 int NspcbSize = sizeof (struct nspcb);