tcp.c

这是一个同样来自贝尔实验室的和UNIX有着渊源的操作系统, 其简洁的设计和实现易于我们学习和理解

}

void
tcpackproc(void *a)
{
	Tcptimer *t, *tp, *timeo;
	Proto *tcp;
	Tcppriv *priv;
	int loop;

	tcp = a;
	priv = tcp->priv;

	for(;;) {
		tsleep(&up->sleep, return0, 0, MSPTICK);

		qlock(&priv->tl);
		timeo = nil;
		loop = 0;
		for(t = priv->timers; t != nil; t = tp) {
			if(loop++ > 10000)
				panic("tcpackproc1");
			tp = t->next;
 			if(t->state == TcptimerON) {
				t->count--;
				if(t->count == 0) {
					timerstate(priv, t, TcptimerDONE);
					t->readynext = timeo;
					timeo = t;
				}
			}
		}
		qunlock(&priv->tl);

		loop = 0;
		for(t = timeo; t != nil; t = t->readynext) {
			if(loop++ > 10000)
				panic("tcpackproc2");
			if(t->state == TcptimerDONE && t->func != nil && !waserror()){
				(*t->func)(t->arg);
				poperror();
			}
		}

		limborexmit(tcp);
	}
}

void
tcpgo(Tcppriv *priv, Tcptimer *t)
{
	if(t == nil || t->start == 0)
		return;

	qlock(&priv->tl);
	t->count = t->start;
	timerstate(priv, t, TcptimerON);
	qunlock(&priv->tl);
}

void
tcphalt(Tcppriv *priv, Tcptimer *t)
{
	if(t == nil)
		return;

	qlock(&priv->tl);
	timerstate(priv, t, TcptimerOFF);
	qunlock(&priv->tl);
}

int
backoff(int n)
{
	return 1 << n;
}

void
localclose(Conv *s, char *reason)	/* called with tcb locked */
{
	Tcpctl *tcb;
	Reseq *rp,*rp1;
	Tcppriv *tpriv;

	tpriv = s->p->priv;
	tcb = (Tcpctl*)s->ptcl;

	iphtrem(&tpriv->ht, s);

	tcphalt(tpriv, &tcb->timer);
	tcphalt(tpriv, &tcb->rtt_timer);
	tcphalt(tpriv, &tcb->acktimer);
	tcphalt(tpriv, &tcb->katimer);

	/* Flush reassembly queue; nothing more can arrive */
	for(rp = tcb->reseq; rp != nil; rp = rp1) {
		rp1 = rp->next;
		freeblist(rp->bp);
		free(rp);
	}
	tcb->reseq = nil;

	if(tcb->state == Syn_sent)
		Fsconnected(s, reason);
	if(s->state == Announced)
		wakeup(&s->listenr);

	qhangup(s->rq, reason);
	qhangup(s->wq, reason);

	tcpsetstate(s, Closed);
}

/* mtu (- TCP + IP hdr len) of 1st hop */
int
tcpmtu(Proto *tcp, uchar *addr, int version, int *scale)
{
	Ipifc *ifc;
	int mtu;

	ifc = findipifc(tcp->f, addr, 0);
	switch(version){
	default:
	case V4:
		mtu = DEF_MSS;
		if(ifc != nil)
			mtu = ifc->maxtu - ifc->m->hsize - (TCP4_PKT + TCP4_HDRSIZE);
		break;
	case V6:
		mtu = DEF_MSS6;
		if(ifc != nil)
			mtu = ifc->maxtu - ifc->m->hsize - (TCP6_PKT + TCP6_HDRSIZE);
		break;
	}
	if(ifc != nil){
		if(ifc->mbps > 100)
			*scale = HaveWS | 3;
		else if(ifc->mbps > 10)
			*scale = HaveWS | 1;
		else
			*scale = HaveWS | 0;
	} else
		*scale = HaveWS | 0;

	return mtu;
}

void
inittcpctl(Conv *s, int mode)
{
	Tcpctl *tcb;
	Tcp4hdr* h4;
	Tcp6hdr* h6;
	int mss;

	tcb = (Tcpctl*)s->ptcl;

	memset(tcb, 0, sizeof(Tcpctl));

	tcb->ssthresh = 65535;
	tcb->srtt = tcp_irtt<<LOGAGAIN;
	tcb->mdev = 0;

	/* setup timers */
	tcb->timer.start = tcp_irtt / MSPTICK;
	tcb->timer.func = tcptimeout;
	tcb->timer.arg = s;
	tcb->rtt_timer.start = MAX_TIME;
	tcb->acktimer.start = TCP_ACK / MSPTICK;
	tcb->acktimer.func = tcpacktimer;
	tcb->acktimer.arg = s;
	tcb->katimer.start = DEF_KAT / MSPTICK;
	tcb->katimer.func = tcpkeepalive;
	tcb->katimer.arg = s;

	mss = DEF_MSS;

	/* create a prototype(pseudo) header */
	if(mode != TCP_LISTEN){
		if(ipcmp(s->laddr, IPnoaddr) == 0)
			findlocalip(s->p->f, s->laddr, s->raddr);

		switch(s->ipversion){
		case V4:
			h4 = &tcb->protohdr.tcp4hdr;
			memset(h4, 0, sizeof(*h4));
			h4->proto = IP_TCPPROTO;
			hnputs(h4->tcpsport, s->lport);
			hnputs(h4->tcpdport, s->rport);
			v6tov4(h4->tcpsrc, s->laddr);
			v6tov4(h4->tcpdst, s->raddr);
			break;
		case V6:
			h6 = &tcb->protohdr.tcp6hdr;
			memset(h6, 0, sizeof(*h6));
			h6->proto = IP_TCPPROTO;
			hnputs(h6->tcpsport, s->lport);
			hnputs(h6->tcpdport, s->rport);
			ipmove(h6->tcpsrc, s->laddr);
			ipmove(h6->tcpdst, s->raddr);
			mss = DEF_MSS6;
			break;
		default:
			panic("inittcpctl: version %d", s->ipversion);
		}
	}

	tcb->mss = tcb->cwind = mss;

	/* default is no window scaling */
	tcb->window = QMAX;
	tcb->rcv.wnd = QMAX;
	tcb->rcv.scale = 0;
	tcb->snd.scale = 0;
	qsetlimit(s->rq, QMAX);
}

/*
 *  called with s qlocked
 */
void
tcpstart(Conv *s, int mode)
{
	Tcpctl *tcb;
	Tcppriv *tpriv;
	char kpname[KNAMELEN];

	tpriv = s->p->priv;

	if(tpriv->ackprocstarted == 0){
		qlock(&tpriv->apl);
		if(tpriv->ackprocstarted == 0){
			sprint(kpname, "#I%dtcpack", s->p->f->dev);
			kproc(kpname, tcpackproc, s->p);
			tpriv->ackprocstarted = 1;
		}
		qunlock(&tpriv->apl);
	}

	tcb = (Tcpctl*)s->ptcl;

	inittcpctl(s, mode);

	iphtadd(&tpriv->ht, s);
	switch(mode) {
	case TCP_LISTEN:
		tpriv->stats[PassiveOpens]++;
		tcb->flags |= CLONE;
		tcpsetstate(s, Listen);
		break;

	case TCP_CONNECT:
		tpriv->stats[ActiveOpens]++;
		tcb->flags |= ACTIVE;
		tcpsndsyn(s, tcb);
		tcpsetstate(s, Syn_sent);
		tcpoutput(s);
		break;
	}
}

static char*
tcpflag(ushort flag)
{
	static char buf[128];

	sprint(buf, "%d", flag>>10);	/* Head len */
	if(flag & URG)
		strcat(buf, " URG");
	if(flag & ACK)
		strcat(buf, " ACK");
	if(flag & PSH)
		strcat(buf, " PSH");
	if(flag & RST)
		strcat(buf, " RST");
	if(flag & SYN)
		strcat(buf, " SYN");
	if(flag & FIN)
		strcat(buf, " FIN");

	return buf;
}

Block *
htontcp6(Tcp *tcph, Block *data, Tcp6hdr *ph, Tcpctl *tcb)
{
	int dlen;
	Tcp6hdr *h;
	ushort csum;
	ushort hdrlen, optpad = 0;
	uchar *opt;

	hdrlen = TCP6_HDRSIZE;
	if(tcph->flags & SYN){
		if(tcph->mss)
			hdrlen += MSS_LENGTH;
		if(tcph->ws)
			hdrlen += WS_LENGTH;
		optpad = hdrlen & 3;
		if(optpad)
			optpad = 4 - optpad;
		hdrlen += optpad;
	}

	if(data) {
		dlen = blocklen(data);
		data = padblock(data, hdrlen + TCP6_PKT);
		if(data == nil)
			return nil;
	}
	else {
		dlen = 0;
		data = allocb(hdrlen + TCP6_PKT + 64);	/* the 64 pad is to meet mintu's */
		if(data == nil)
			return nil;
		data->wp += hdrlen + TCP6_PKT;
	}

	/* copy in pseudo ip header plus port numbers */
	h = (Tcp6hdr *)(data->rp);
	memmove(h, ph, TCP6_TCBPHDRSZ);

	/* compose pseudo tcp header, do cksum calculation */
	hnputl(h->vcf, hdrlen + dlen);
	h->ploadlen[0] = h->ploadlen[1] = h->proto = 0;
	h->ttl = ph->proto;

	/* copy in variable bits */
	hnputl(h->tcpseq, tcph->seq);
	hnputl(h->tcpack, tcph->ack);
	hnputs(h->tcpflag, (hdrlen<<10) | tcph->flags);
	hnputs(h->tcpwin, tcph->wnd>>(tcb != nil ? tcb->snd.scale : 0));
	hnputs(h->tcpurg, tcph->urg);

	if(tcph->flags & SYN){
		opt = h->tcpopt;
		if(tcph->mss != 0){
			*opt++ = MSSOPT;
			*opt++ = MSS_LENGTH;
			hnputs(opt, tcph->mss);
			opt += 2;
		}
		if(tcph->ws != 0){
			*opt++ = WSOPT;
			*opt++ = WS_LENGTH;
			*opt++ = tcph->ws;
		}
		while(optpad-- > 0)
			*opt++ = NOOPOPT;
	}

	if(tcb != nil && tcb->nochecksum){
		h->tcpcksum[0] = h->tcpcksum[1] = 0;
	} else {
		csum = ptclcsum(data, TCP6_IPLEN, hdrlen+dlen+TCP6_PHDRSIZE);
		hnputs(h->tcpcksum, csum);
	}

	/* move from pseudo header back to normal ip header */
	memset(h->vcf, 0, 4);
	h->vcf[0] = IP_VER6;
	hnputs(h->ploadlen, hdrlen+dlen);
	h->proto = ph->proto;

	return data;
}

Block *
htontcp4(Tcp *tcph, Block *data, Tcp4hdr *ph, Tcpctl *tcb)
{
	int dlen;
	Tcp4hdr *h;
	ushort csum;
	ushort hdrlen, optpad = 0;
	uchar *opt;

	hdrlen = TCP4_HDRSIZE;
	if(tcph->flags & SYN){
		if(tcph->mss)
			hdrlen += MSS_LENGTH;
		if(tcph->ws)
			hdrlen += WS_LENGTH;
		optpad = hdrlen & 3;
		if(optpad)
			optpad = 4 - optpad;
		hdrlen += optpad;
	}

	if(data) {
		dlen = blocklen(data);
		data = padblock(data, hdrlen + TCP4_PKT);
		if(data == nil)
			return nil;
	}
	else {
		dlen = 0;
		data = allocb(hdrlen + TCP4_PKT + 64);	/* the 64 pad is to meet mintu's */
		if(data == nil)
			return nil;
		data->wp += hdrlen + TCP4_PKT;
	}

	/* copy in pseudo ip header plus port numbers */
	h = (Tcp4hdr *)(data->rp);
	memmove(h, ph, TCP4_TCBPHDRSZ);

	/* copy in variable bits */
	hnputs(h->tcplen, hdrlen + dlen);
	hnputl(h->tcpseq, tcph->seq);
	hnputl(h->tcpack, tcph->ack);
	hnputs(h->tcpflag, (hdrlen<<10) | tcph->flags);
	hnputs(h->tcpwin, tcph->wnd>>(tcb != nil ? tcb->snd.scale : 0));
	hnputs(h->tcpurg, tcph->urg);

	if(tcph->flags & SYN){
		opt = h->tcpopt;
		if(tcph->mss != 0){
			*opt++ = MSSOPT;
			*opt++ = MSS_LENGTH;
			hnputs(opt, tcph->mss);
			opt += 2;
		}
		if(tcph->ws != 0){
			*opt++ = WSOPT;
			*opt++ = WS_LENGTH;
			*opt++ = tcph->ws;
		}
		while(optpad-- > 0)
			*opt++ = NOOPOPT;
	}

	if(tcb != nil && tcb->nochecksum){
		h->tcpcksum[0] = h->tcpcksum[1] = 0;
	} else {
		csum = ptclcsum(data, TCP4_IPLEN, hdrlen+dlen+TCP4_PHDRSIZE);
		hnputs(h->tcpcksum, csum);
	}

	return data;
}

int
ntohtcp6(Tcp *tcph, Block **bpp)
{
	Tcp6hdr *h;
	uchar *optr;
	ushort hdrlen;
	ushort optlen;
	int n;

	*bpp = pullupblock(*bpp, TCP6_PKT+TCP6_HDRSIZE);
	if(*bpp == nil)
		return -1;

	h = (Tcp6hdr *)((*bpp)->rp);
	tcph->source = nhgets(h->tcpsport);
	tcph->dest = nhgets(h->tcpdport);
	tcph->seq = nhgetl(h->tcpseq);
	tcph->ack = nhgetl(h->tcpack);
	hdrlen = (h->tcpflag[0]>>2) & ~3;
	if(hdrlen < TCP6_HDRSIZE) {
		freeblist(*bpp);
		return -1;
	}

	tcph->flags = h->tcpflag[1];
	tcph->wnd = nhgets(h->tcpwin);
	tcph->urg = nhgets(h->tcpurg);
	tcph->mss = 0;
	tcph->ws = 0;
	tcph->len = nhgets(h->ploadlen) - hdrlen;

	*bpp = pullupblock(*bpp, hdrlen+TCP6_PKT);
	if(*bpp == nil)
		return -1;

	optr = h->tcpopt;
	n = hdrlen - TCP6_HDRSIZE;
	while(n > 0 && *optr != EOLOPT) {
		if(*optr == NOOPOPT) {
			n--;
			optr++;
			continue;
		}
		optlen = optr[1];
		if(optlen < 2 || optlen > n)
			break;
		switch(*optr) {
		case MSSOPT:
			if(optlen == MSS_LENGTH)
				tcph->mss = nhgets(optr+2);
			break;
		case WSOPT:
			if(optlen == WS_LENGTH && *(optr+2) <= 14)
				tcph->ws = HaveWS | *(optr+2);
			break;
		}
		n -= optlen;
		optr += optlen;
	}
	return hdrlen;
}

int
ntohtcp4(Tcp *tcph, Block **bpp)
{
	Tcp4hdr *h;
	uchar *optr;
	ushort hdrlen;
	ushort optlen;
	int n;

	*bpp = pullupblock(*bpp, TCP4_PKT+TCP4_HDRSIZE);
	if(*bpp == nil)
		return -1;

	h = (Tcp4hdr *)((*bpp)->rp);
	tcph->source = nhgets(h->tcpsport);
	tcph->dest = nhgets(h->tcpdport);
	tcph->seq = nhgetl(h->tcpseq);
	tcph->ack = nhgetl(h->tcpack);

	hdrlen = (h->tcpflag[0]>>2) & ~3;
	if(hdrlen < TCP4_HDRSIZE) {
		freeblist(*bpp);
		return -1;
	}

	tcph->flags = h->tcpflag[1];
	tcph->wnd = nhgets(h->tcpwin);
	tcph->urg = nhgets(h->tcpurg);
	tcph->mss = 0;
	tcph->ws = 0;
	tcph->len = nhgets(h->length) - (hdrlen + TCP4_PKT);

	*bpp = pullupblock(*bpp, hdrlen+TCP4_PKT);
	if(*bpp == nil)
		return -1;

	optr = h->tcpopt;
	n = hdrlen - TCP4_HDRSIZE;
	while(n > 0 && *optr != EOLOPT) {
		if(*optr == NOOPOPT) {
			n--;
			optr++;
			continue;
		}
		optlen = optr[1];
		if(optlen < 2 || optlen > n)
			break;
		switch(*optr) {
		case MSSOPT:
			if(optlen == MSS_LENGTH)
				tcph->mss = nhgets(optr+2);
			break;
		case WSOPT:
			if(optlen == WS_LENGTH && *(optr+2) <= 14)
				tcph->ws = HaveWS | *(optr+2);
			break;
		}
		n -= optlen;
		optr += optlen;
	}
	return hdrlen;
}

/*
 *  For outgiing calls, generate an initial sequence
 *  number and put a SYN on the send queue
 */
void
tcpsndsyn(Conv *s, Tcpctl *tcb)
{
	tcb->iss = (nrand(1<<16)<<16)|nrand(1<<16);
	tcb->rttseq = tcb->iss;
	tcb->snd.wl2 = tcb->iss;
	tcb->snd.una = tcb->iss;
	tcb->snd.ptr = tcb->rttseq;
	tcb->snd.nxt = tcb->rttseq;
	tcb->flgcnt++;
	tcb->flags |= FORCE;
	tcb->sndsyntime = NOW;

	/* set desired mss and scale */
	tcb->mss = tcpmtu(s->p, s->laddr, s->ipversion, &tcb->scale);
}

void
sndrst(Proto *tcp, uchar *source, uchar *dest, ushort length, Tcp *seg, uchar version, char *reason)
{
	Block *hbp;
	uchar rflags;
	Tcppriv *tpriv;
	Tcp4hdr ph4;
	Tcp6hdr ph6;

	netlog(tcp->f, Logtcp, "sndrst: %s", reason);

	tpriv = tcp->priv;

	if(seg->flags & RST)
		return;

	/* make pseudo header */
	switch(version) {
	case V4:
		memset(&ph4, 0, sizeof(ph4));
		ph4.vihl = IP_VER4;
		v6tov4(ph4.tcpsrc, dest);
		v6tov4(ph4.tcpdst, source);
		ph4.proto = IP_TCPPROTO;
		hnputs(ph4.tcplen, TCP4_HDRSIZE);
		hnputs(ph4.tcpsport, seg->dest);
		hnputs(ph4.tcpdport, seg->source);
		break;
	case V6:
		memset(&ph6, 0, sizeof(ph6));
		ph6.vcf[0] = IP_VER6;
		ipmove(ph6.tcpsrc, dest);
		ipmove(ph6.tcpdst, source);
		ph6.proto = IP_TCPPROTO;
		hnputs(ph6.ploadlen, TCP6_HDRSIZE);
		hnputs(ph6.tcpsport, seg->dest);
		hnputs(ph6.tcpdport, seg->source);
		break;
	default:
		panic("sndrst: version %d", version);
	}

	tpriv->stats[OutRsts]++;
	rflags = RST;

	/* convince the other end that this reset is in band */
	if(seg->flags & ACK) {
		seg->seq = seg->ack;