Re: Summary of how linux can best avoid the need for streams

Hans Reiser (reiser@ceic.com)
Tue, 29 Jun 1999 23:54:31 +0000 (/etc/localtime)


Jamie Lokier writes:
> Hans Reiser wrote:
> > * inheritance of stat data
> > * overloading directory names so that if they are opened as files they are
> > files, and if accessed as directories they are directories.
>
> Probably ok in kernel.
>
> > * inheritance of file bodies
> > * a syntax based on rdf for writing to files which have inheritance
> > (solving this stumbling block was important for me)
> > * filters, such that "dirname/..tar" generates a tar file when read, and
> > "dirname/..cat" concatenates for read, and "filename/..filtername" runs
> > filtername on the file/directory filename.
>
> Can I suggest this is handled in user-space, but using kernel hooks?

You likely are right in this division of features to user and kernel
space.

>
> That is, the kernel is asked to open a file as directory or whatever,
> and just like the hacks to cd into a tar file, the kernel invokes a
> user-space solution to create & cache the fake files.
>
> I suggest there are a lot of things that could be done by extending the
> userfs approach to try dynamically allocating mount points. Kernel
> support could be very simple.
>
> - Treating an archive file as a directory (podfuk?)

Non sequitur: reiserfs is more compact than tar for small files..., but
not more compact than gzip, we'll work on it...

> - Viewing a structured storage directory as a flat file
> - URL access (http, ftp)
>
> > All of these features are valuable in and of themselves. Together they
> > collectively eliminate all need for structured storage. Jeremy
> > Allison's needs for emulating W2k are a good test of whether that goal
> > has been reached, and so I use his needs as one test.
> >
> > Hans
> >

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/