> > BTW, where are the zerocopy patches for pipes? Maybe I'm missing 
> > something but it seems that pipes inside the kernel are still 
> > implememented by copying into the kernel and then copying out. 
> > Whatever method the zerocopy pipes use is probably what I'm looking 
> > for though. 
> 
> Ask DaveM or look through l-k archives for URL of recent variant... 

I've attached the latest patch. I didn't send it to Linus yet - it isn't
really a bug fix, and there are a few open issures:

* I still hope that zero-copy (actually single copy) is possible for
PIPE_BUF (4096) byte transfers.

Unfortunately that would be a user space visible change:

void main()
{
	int pipes[2];
	pipe(pipes);
	write(pipes[1],buf,PIPE_BUF);
}

would block, every other Unix version I tested doesn't block. (or
write(,,PIPE_BUF) after poll(POLLOUT) - I'm sure that there are apps
where the O_NONBLOCK is missing)

But since glibc by default uses 4096 byte writes it would speed up the
transfers.

* > 64 kB chunks

* memory pressure. Not that important for 64 kB transfers, but required
for larger buffers.

I tested the patch with 2.4.1, and it compiles with 2.4.2-ac11.

--
	Manfred