spp_usrreq.c

早期freebsd实现

		if (sns->sns_family != AF_NS)
			return;
		na = &sns->sns_addr;
		break;

	default:
		errp = (struct ns_errp *)arg;
		na = &errp->ns_err_idp.idp_dna;
		type = errp->ns_err_num;
		type = ntohs((u_short)type);
	}
	switch (type) {

	case NS_ERR_UNREACH_HOST:
		ns_pcbnotify(na, (int)nsctlerrmap[cmd], spp_abort, (long) 0);
		break;

	case NS_ERR_TOO_BIG:
	case NS_ERR_NOSOCK:
		nsp = ns_pcblookup(na, errp->ns_err_idp.idp_sna.x_port,
			NS_WILDCARD);
		if (nsp) {
			if(nsp->nsp_pcb)
				(void) spp_drop((struct sppcb *)nsp->nsp_pcb,
						(int)nsctlerrmap[cmd]);
			else
				(void) idp_drop(nsp, (int)nsctlerrmap[cmd]);
		}
		break;

	case NS_ERR_FULLUP:
		ns_pcbnotify(na, 0, spp_quench, (long) 0);
	}
}
/*
 * When a source quench is received, close congestion window
 * to one packet.  We will gradually open it again as we proceed.
 */
spp_quench(nsp)
	struct nspcb *nsp;
{
	struct sppcb *cb = nstosppcb(nsp);

	if (cb)
		cb->s_cwnd = CUNIT;
}

#ifdef notdef
int
spp_fixmtu(nsp)
register struct nspcb *nsp;
{
	register struct sppcb *cb = (struct sppcb *)(nsp->nsp_pcb);
	register struct mbuf *m;
	register struct spidp *si;
	struct ns_errp *ep;
	struct sockbuf *sb;
	int badseq, len;
	struct mbuf *firstbad, *m0;

	if (cb) {
		/* 
		 * The notification that we have sent
		 * too much is bad news -- we will
		 * have to go through queued up so far
		 * splitting ones which are too big and
		 * reassigning sequence numbers and checksums.
		 * we should then retransmit all packets from
		 * one above the offending packet to the last one
		 * we had sent (or our allocation)
		 * then the offending one so that the any queued
		 * data at our destination will be discarded.
		 */
		 ep = (struct ns_errp *)nsp->nsp_notify_param;
		 sb = &nsp->nsp_socket->so_snd;
		 cb->s_mtu = ep->ns_err_param;
		 badseq = SI(&ep->ns_err_idp)->si_seq;
		 for (m = sb->sb_mb; m; m = m->m_act) {
			si = mtod(m, struct spidp *);
			if (si->si_seq == badseq)
				break;
		 }
		 if (m == 0) return;
		 firstbad = m;
		 /*for (;;) {*/
			/* calculate length */
			for (m0 = m, len = 0; m ; m = m->m_next)
				len += m->m_len;
			if (len > cb->s_mtu) {
			}
		/* FINISH THIS
		} */
	}
}
#endif

spp_output(cb, m0)
	register struct sppcb *cb;
	struct mbuf *m0;
{
	struct socket *so = cb->s_nspcb->nsp_socket;
	register struct mbuf *m;
	register struct spidp *si = (struct spidp *) 0;
	register struct sockbuf *sb = &so->so_snd;
	int len = 0, win, rcv_win;
	short span, off, recordp = 0;
	u_short alo;
	int error = 0, sendalot;
#ifdef notdef
	int idle;
#endif
	struct mbuf *mprev;
	extern int idpcksum;

	if (m0) {
		int mtu = cb->s_mtu;
		int datalen;
		/*
		 * Make sure that packet isn't too big.
		 */
		for (m = m0; m ; m = m->m_next) {
			mprev = m;
			len += m->m_len;
			if (m->m_flags & M_EOR)
				recordp = 1;
		}
		datalen = (cb->s_flags & SF_HO) ?
				len - sizeof (struct sphdr) : len;
		if (datalen > mtu) {
			if (cb->s_flags & SF_PI) {
				m_freem(m0);
				return (EMSGSIZE);
			} else {
				int oldEM = cb->s_cc & SP_EM;

				cb->s_cc &= ~SP_EM;
				while (len > mtu) {
					/*
					 * Here we are only being called
					 * from usrreq(), so it is OK to
					 * block.
					 */
					m = m_copym(m0, 0, mtu, M_WAIT);
					if (cb->s_flags & SF_NEWCALL) {
					    struct mbuf *mm = m;
					    spp_newchecks[7]++;
					    while (mm) {
						mm->m_flags &= ~M_EOR;
						mm = mm->m_next;
					    }
					}
					error = spp_output(cb, m);
					if (error) {
						cb->s_cc |= oldEM;
						m_freem(m0);
						return(error);
					}
					m_adj(m0, mtu);
					len -= mtu;
				}
				cb->s_cc |= oldEM;
			}
		}
		/*
		 * Force length even, by adding a "garbage byte" if
		 * necessary.
		 */
		if (len & 1) {
			m = mprev;
			if (M_TRAILINGSPACE(m) >= 1)
				m->m_len++;
			else {
				struct mbuf *m1 = m_get(M_DONTWAIT, MT_DATA);

				if (m1 == 0) {
					m_freem(m0);
					return (ENOBUFS);
				}
				m1->m_len = 1;
				*(mtod(m1, u_char *)) = 0;
				m->m_next = m1;
			}
		}
		m = m_gethdr(M_DONTWAIT, MT_HEADER);
		if (m == 0) {
			m_freem(m0);
			return (ENOBUFS);
		}
		/*
		 * Fill in mbuf with extended SP header
		 * and addresses and length put into network format.
		 */
		MH_ALIGN(m, sizeof (struct spidp));
		m->m_len = sizeof (struct spidp);
		m->m_next = m0;
		si = mtod(m, struct spidp *);
		si->si_i = *cb->s_idp;
		si->si_s = cb->s_shdr;
		if ((cb->s_flags & SF_PI) && (cb->s_flags & SF_HO)) {
			register struct sphdr *sh;
			if (m0->m_len < sizeof (*sh)) {
				if((m0 = m_pullup(m0, sizeof(*sh))) == NULL) {
					(void) m_free(m);
					m_freem(m0);
					return (EINVAL);
				}
				m->m_next = m0;
			}
			sh = mtod(m0, struct sphdr *);
			si->si_dt = sh->sp_dt;
			si->si_cc |= sh->sp_cc & SP_EM;
			m0->m_len -= sizeof (*sh);
			m0->m_data += sizeof (*sh);
			len -= sizeof (*sh);
		}
		len += sizeof(*si);
		if ((cb->s_flags2 & SF_NEWCALL) && recordp) {
			si->si_cc  |= SP_EM;
			spp_newchecks[8]++;
		}
		if (cb->s_oobflags & SF_SOOB) {
			/*
			 * Per jqj@cornell:
			 * make sure OB packets convey exactly 1 byte.
			 * If the packet is 1 byte or larger, we
			 * have already guaranted there to be at least
			 * one garbage byte for the checksum, and
			 * extra bytes shouldn't hurt!
			 */
			if (len > sizeof(*si)) {
				si->si_cc |= SP_OB;
				len = (1 + sizeof(*si));
			}
		}
		si->si_len = htons((u_short)len);
		m->m_pkthdr.len = ((len - 1) | 1) + 1;
		/*
		 * queue stuff up for output
		 */
		sbappendrecord(sb, m);
		cb->s_seq++;
	}
#ifdef notdef
	idle = (cb->s_smax == (cb->s_rack - 1));
#endif
again:
	sendalot = 0;
	off = cb->s_snxt - cb->s_rack;
	win = min(cb->s_swnd, (cb->s_cwnd/CUNIT));

	/*
	 * If in persist timeout with window of 0, send a probe.
	 * Otherwise, if window is small but nonzero
	 * and timer expired, send what we can and go into
	 * transmit state.
	 */
	if (cb->s_force == 1 + SPPT_PERSIST) {
		if (win != 0) {
			cb->s_timer[SPPT_PERSIST] = 0;
			cb->s_rxtshift = 0;
		}
	}
	span = cb->s_seq - cb->s_rack;
	len = min(span, win) - off;

	if (len < 0) {
		/*
		 * Window shrank after we went into it.
		 * If window shrank to 0, cancel pending
		 * restransmission and pull s_snxt back
		 * to (closed) window.  We will enter persist
		 * state below.  If the widndow didn't close completely,
		 * just wait for an ACK.
		 */
		len = 0;
		if (win == 0) {
			cb->s_timer[SPPT_REXMT] = 0;
			cb->s_snxt = cb->s_rack;
		}
	}
	if (len > 1)
		sendalot = 1;
	rcv_win = sbspace(&so->so_rcv);

	/*
	 * Send if we owe peer an ACK.
	 */
	if (cb->s_oobflags & SF_SOOB) {
		/*
		 * must transmit this out of band packet
		 */
		cb->s_oobflags &= ~ SF_SOOB;
		sendalot = 1;
		sppstat.spps_sndurg++;
		goto found;
	}
	if (cb->s_flags & SF_ACKNOW)
		goto send;
	if (cb->s_state < TCPS_ESTABLISHED)
		goto send;
	/*
	 * Silly window can't happen in spp.
	 * Code from tcp deleted.
	 */
	if (len)
		goto send;
	/*
	 * Compare available window to amount of window
	 * known to peer (as advertised window less
	 * next expected input.)  If the difference is at least two
	 * packets or at least 35% of the mximum possible window,
	 * then want to send a window update to peer.
	 */
	if (rcv_win > 0) {
		u_short delta =  1 + cb->s_alo - cb->s_ack;
		int adv = rcv_win - (delta * cb->s_mtu);
		
		if ((so->so_rcv.sb_cc == 0 && adv >= (2 * cb->s_mtu)) ||
		    (100 * adv / so->so_rcv.sb_hiwat >= 35)) {
			sppstat.spps_sndwinup++;
			cb->s_flags |= SF_ACKNOW;
			goto send;
		}

	}
	/*
	 * Many comments from tcp_output.c are appropriate here
	 * including . . .
	 * If send window is too small, there is data to transmit, and no
	 * retransmit or persist is pending, then go to persist state.
	 * If nothing happens soon, send when timer expires:
	 * if window is nonzero, transmit what we can,
	 * otherwise send a probe.
	 */
	if (so->so_snd.sb_cc && cb->s_timer[SPPT_REXMT] == 0 &&
		cb->s_timer[SPPT_PERSIST] == 0) {
			cb->s_rxtshift = 0;
			spp_setpersist(cb);
	}
	/*
	 * No reason to send a packet, just return.
	 */
	cb->s_outx = 1;
	return (0);

send:
	/*
	 * Find requested packet.
	 */
	si = 0;
	if (len > 0) {
		cb->s_want = cb->s_snxt;
		for (m = sb->sb_mb; m; m = m->m_act) {
			si = mtod(m, struct spidp *);
			if (SSEQ_LEQ(cb->s_snxt, si->si_seq))
				break;
		}
	found:
		if (si) {
			if (si->si_seq == cb->s_snxt)
					cb->s_snxt++;
				else
					sppstat.spps_sndvoid++, si = 0;
		}
	}
	/*
	 * update window
	 */
	if (rcv_win < 0)
		rcv_win = 0;
	alo = cb->s_ack - 1 + (rcv_win / ((short)cb->s_mtu));
	if (SSEQ_LT(alo, cb->s_alo)) 
		alo = cb->s_alo;

	if (si) {
		/*
		 * must make a copy of this packet for
		 * idp_output to monkey with
		 */
		m = m_copy(dtom(si), 0, (int)M_COPYALL);
		if (m == NULL) {
			return (ENOBUFS);
		}
		si = mtod(m, struct spidp *);
		if (SSEQ_LT(si->si_seq, cb->s_smax))
			sppstat.spps_sndrexmitpack++;
		else
			sppstat.spps_sndpack++;
	} else if (cb->s_force || cb->s_flags & SF_ACKNOW) {
		/*
		 * Must send an acknowledgement or a probe
		 */
		if (cb->s_force)
			sppstat.spps_sndprobe++;
		if (cb->s_flags & SF_ACKNOW)
			sppstat.spps_sndacks++;
		m = m_gethdr(M_DONTWAIT, MT_HEADER);
		if (m == 0)
			return (ENOBUFS);
		/*
		 * Fill in mbuf with extended SP header
		 * and addresses and length put into network format.
		 */
		MH_ALIGN(m, sizeof (struct spidp));
		m->m_len = sizeof (*si);
		m->m_pkthdr.len = sizeof (*si);
		si = mtod(m, struct spidp *);
		si->si_i = *cb->s_idp;
		si->si_s = cb->s_shdr;
		si->si_seq = cb->s_smax + 1;
		si->si_len = htons(sizeof (*si));
		si->si_cc |= SP_SP;
	} else {
		cb->s_outx = 3;
		if (so->so_options & SO_DEBUG || traceallspps)
			spp_trace(SA_OUTPUT, cb->s_state, cb, si, 0);
		return (0);
	}
	/*
	 * Stuff checksum and output datagram.
	 */
	if ((si->si_cc & SP_SP) == 0) {
		if (cb->s_force != (1 + SPPT_PERSIST) ||
		    cb->s_timer[SPPT_PERSIST] == 0) {
			/*
			 * If this is a new packet and we are not currently 
			 * timing anything, time this one.
			 */
			if (SSEQ_LT(cb->s_smax, si->si_seq)) {
				cb->s_smax = si->si_seq;
				if (cb->s_rtt == 0) {
					sppstat.spps_segstimed++;
					cb->s_rtseq = si->si_seq;
					cb->s_rtt = 1;
				}
			}
			/*
			 * Set rexmt timer if not currently set,
			 * Initial value for retransmit timer is smoothed
			 * round-trip time + 2 * round-trip time variance.
			 * Initialize shift counter which is used for backoff
			 * of retransmit time.
			 */
			if (cb->s_timer[SPPT_REXMT] == 0 &&
			    cb->s_snxt != cb->s_rack) {
				cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
				if (cb->s_timer[SPPT_PERSIST]) {
					cb->s_timer[SPPT_PERSIST] = 0;
					cb->s_rxtshift = 0;
				}
			}
		} else if (SSEQ_LT(cb->s_smax, si->si_seq)) {
			cb->s_smax = si->si_seq;
		}
	} else if (cb->s_state < TCPS_ESTABLISHED) {
		if (cb->s_rtt == 0)
			cb->s_rtt = 1; /* Time initial handshake */
		if (cb->s_timer[SPPT_REXMT] == 0)
			cb->s_timer[SPPT_REXMT] = cb->s_rxtcur;
	}
	{
		/*
		 * Do not request acks when we ack their data packets or
		 * when we do a gratuitous window update.
		 */
		if (((si->si_cc & SP_SP) == 0) || cb->s_force)
				si->si_cc |= SP_SA;
		si->si_seq = htons(si->si_seq);
		si->si_alo = htons(alo);
		si->si_ack = htons(cb->s_ack);

		if (idpcksum) {
			si->si_sum = 0;
			len = ntohs(si->si_len);
			if (len & 1)
				len++;
			si->si_sum = ns_cksum(m, len);
		} else
			si->si_sum = 0xffff;

		cb->s_outx = 4;
		if (so->so_options & SO_DEBUG || traceallspps)
			spp_trace(SA_OUTPUT, cb->s_state, cb, si, 0);

		if (so->so_options & SO_DONTROUTE)
			error = ns_output(m, (struct route *)0, NS_ROUTETOIF);
		else
			error = ns_output(m, &cb->s_nspcb->nsp_route, 0);
	}
	if (error) {
		return (error);
	}
	sppstat.spps_sndtotal++;
	/*
	 * Data sent (as far as we can tell).
	 * If this advertises a larger window than any other segment,
	 * then remember the size of the advertized window.
	 * Any pending ACK has now been sent.
	 */
	cb->s_force = 0;
	cb->s_flags &= ~(SF_ACKNOW|SF_DELACK);
	if (SSEQ_GT(alo, cb->s_alo))
		cb->s_alo = alo;
	if (sendalot)
		goto again;
	cb->s_outx = 5;
	return (0);
}

int spp_do_persist_panics = 0;

spp_setpersist(cb)
	register struct sppcb *cb;
{
	register t = ((cb->s_srtt >> 2) + cb->s_rttvar) >> 1;
	extern int spp_backoff[];

	if (cb->s_timer[SPPT_REXMT] && spp_do_persist_panics)
		panic("spp_output REXMT");
	/*
	 * Start/restart persistance timer.
	 */
	SPPT_RANGESET(cb->s_timer[SPPT_PERSIST],
	    t*spp_backoff[cb->s_rxtshift],
	    SPPTV_PERSMIN, SPPTV_PERSMAX);
	if (cb->s_rxtshift < SPP_MAXRXTSHIFT)
		cb->s_rxtshift++;
}
/*ARGSUSED*/
spp_ctloutput(req, so, level, name, value)
	int req;
	struct socket *so;
	int name;
	struct mbuf **value;
{
	register struct mbuf *m;
	struct nspcb *nsp = sotonspcb(so);
	register struct sppcb *cb;
	int mask, error = 0;

	if (level != NSPROTO_SPP) {
		/* This will have to be changed when we do more general
		   stacking of protocols */
		return (idp_ctloutput(req, so, level, name, value));
	}
	if (nsp == NULL) {
		error = EINVAL;
		goto release;
	} else
		cb = nstosppcb(nsp);

	switch (req) {

	case PRCO_GETOPT:
		if (value == NULL)
			return (EINVAL);
		m = m_get(M_DONTWAIT, MT_DATA);
		if (m == NULL)
			return (ENOBUFS);
		switch (name) {

		case SO_HEADERS_ON_INPUT:
			mask = SF_HI;
			goto get_flags;

		case SO_HEADERS_ON_OUTPUT:
			mask = SF_HO;
		get_flags:
			m->m_len = sizeof(short);
			*mtod(m, short *) = cb->s_flags & mask;
			break;

		case SO_MTU:
			m->m_len = sizeof(u_short);
			*mtod(m, short *) = cb->s_mtu;
			break;

		case SO_LAST_HEADER:
			m->m_len = sizeof(struct sphdr);
			*mtod(m, struct sphdr *) = cb->s_rhdr;
			break;

		case SO_DEFAULT_HEADERS:
			m->m_len = sizeof(struct spidp);
			*mtod(m, struct sphdr *) = cb->s_shdr;
			break;

		default:
			error = EINVAL;
		}
		*value = m;
		break;

	case PRCO_SETOPT:
		if (value == 0 || *value == 0) {
			error = EINVAL;
			break;
		}
		switch (name) {
			int *ok;

		case SO_HEADERS_ON_INPUT: