Re: [PATCH] splice support #2

[Ingo Molnar <mingo@elte.hu>]

* Jens Axboe <axboe@suse.de> wrote:

> On Thu, Mar 30 2006, Ingo Molnar wrote:
> > 
> > * Jens Axboe <axboe@suse.de> wrote:
> > 
> > > Hi,
> > > 
> > > This patch should resolve all issues mentioned so far. I'd still like 
> > > to implement the page moving, but that should just be a separate 
> > > patch.
> > 
> > neat stuff. One question: why do we require fdin or fdout to be a pipe?  
> > Is there any fundamental problem with implementing what Larry's original 
> > paper described too: straight pagecache -> socket transfers? Without a
> > pipe intermediary forced inbetween. It only adds unnecessary overhead.
> 
> No, not a fundamental problem. I think I even hid that in some comment 
> in there, at least if it's decipharable by someone else than myself... 
> Basically I think it would be nice in the future to tidy this a little 
> bit and separate the actual container from the pipe itself - and have 
> the pipe just fill/use the same container.

why is there a container needed at all? If i splice pagecache->socket, 
we can use sendpage to send it off immediately. There is no need for any 
container - both the pagecache and sendpage use struct page, and when we 
iterate to create a container we might as well ->sendpage() those pages 
off immediately instead.

I agree with the purpose of making sys_splice() generic and in 
particular usable in scripts/shells where pipes are commonly used, but 
we should also fulfill the original promise (outlined 15 years ago or 
so) and not limit this to pipes. That way i could improve TUX to make 
use of it for example ;)

	Ingo