git: 9front

Info  •  Files  •  Log  •  Branches

devether: don't drop packets when bypassing

As we'r already using qsetbypass() on the output queue
to flush it, we could as well just pass the queued
pacekts to the bypass Netfile.

This also gets rid of the qpass() in etheroq().

--- a/sys/src/9/port/devether.c

+++ b/sys/src/9/port/devether.c

@@ -30,6 +30,21 @@

 	freeb(b);

 }

+static void

+bypass(void *arg, Block *b)

+{

+	Ether *ether;

+	Netfile *f;

+

+	ether = (Ether*)arg;

+	if((f = ether->bypass) == nil){

+		freeb(b);

+		return;

+	}

+	if(qpass(f->in, b) < 0)

+		ether->soverflows++;

+}

+

 Chan*

 etherattach(char* spec)

 {

@@ -276,8 +291,6 @@

 static void

 etheroq(Ether* ether, Block* bp, Netfile **from)

 {

-	Netfile *x;

-

 	if((*from)->bridge == 0)

 		memmove(((Etherpkt*)bp->rp)->s, ether->ea, Eaddrlen);

@@ -292,11 +305,6 @@

 		return;

 	if(ether->dmat != nil)

 		dmatproxy(bp, 1, ether->ea, ether->dmat);

-	if((x = ether->bypass) != nil){

-		if(qpass(x->in, bp) < 0)

-			ether->soverflows++;

-		return;

-	}

 	ether->outpackets++;

 	qbwrite(ether->oq, bp);

 }

@@ -314,10 +322,10 @@

 		if(nn >= 0){

 			/* got bypassed? */

 			if(ether->f[NETID(chan->qid.path)]->bypass){

-				/* flush and bypass output queue */

-				qsetbypass(ether->oq, drop);

 				/* ignore mbps and use large input queue size */

 				netifsetlimit(ether, MB);

+				/* bypass output queue */

+				qsetbypass(ether->oq, bypass);

 			}

 			return nn;

 		}

-- 

⑨