Subject: Re: RFC: new mode bits in stat structure
To: Jason Thorpe <thorpej@nas.nasa.gov>
From: Chris G. Demetriou <cgd@pa.dec.com>
List: tech-kern
Date: 07/01/1998 08:38:10
> On Tue, 30 Jun 1998 20:10:59 +0200 (MEST) 
>  Zdenek Salvet <salvet@horn.ics.muni.cz> wrote:
> 
>  > The problem of "ls" display can be handled in general and elegant manner
>  > using fs-specific "ls plugins" (see /usr/lib/fs/ on Solaris or mount_* and
>  > fsck_* families). The overhead would be only one dlopen() per listed
>  > filesystem type.
> 
> ...and I will note that on ELF platforms, this only works if ls(1) is
> dynamically linked (because the interpreter is loaded by the kernel.)

However, it could be possible to load the interpreter, if it's not
already loaded, from user-land code.  I mean, just because that's not
the default mode of operation doesn't mean it can't be done!

Similarly, in statically-linked a.out binaries, you have no loader
loaded, though it's loaded from crt0, but if the code were written
'right' then you could load it later.

You still wouldn't be able to have the symbol information for the main
program, but a carefully written plugin API would keep that from being
a problem.



cgd