Re: Hurd FS hierarchy (was Re: LD_LIBRARY_PATH troubles)

[Richard Kreuter]

On Sat, Mar 16, 2002 at 11:55:40AM -0500, Marcus Brinkmann wrote:
> On Sat, Mar 16, 2002 at 11:07:53AM -0500, Richard Kreuter wrote:
> >   The main difference at the moment between FHS and the hierarchy now
> > used is that /usr is a symlink to /. 
> 
> The trick is that the FHS doesn't demand that /usr/foo is physically
> different from /foo.  We are compliant in this are because of the
> symlink, which ensures that whereever the FHS says that a filename is
> /usr/foo, you can get the file by using /usr/foo.  That you can also
> access it with /foo directly doesn't bother the FHS in the least.

  Erm, the FHS prohibits "introducing a new subdirectory of the root
filesystem" (rationale, section 3.1), on grounds having to do with the
partition containing the host-specific, static boot and recovery
files.  I think that the FHS doesn't consistently and clearly
distinguish the notions of "filesystem on a store accessible at
directory <foo>", from "file hierarchy rooted at directory <foo>".
For example, FHS 3.1 rationale states "Software must never create or
require special files or subdirectories in the root directory",
assuming that the contents of the root directory are exactly the
contents of a single filesystem on a 'root partition'.  I'm not
convinced that the current /usr -> / link breaks FHS compliance; I
mean here that the FHS was not constructed with general enough ideas
about things.  Perhaps not enough of the non-Thomas-Bushnell
contributors to the FHS weren't sufficiently familiar with the
potential of the Hurd?  :)

  In any case, when shadowfs is done, setting the root of the /usr
hierarchy on the root hierarchy's node while maintaining a small store
for the static, host-specific boot and recovery files can accommodate
the stated rationale for the limits of the 'root filesystem' of FHS
parlance.  (Of course, you know this already; but mailing lists are
public forums and public records, so I'm belaboring the point for
those who don't.)  I don't know much about unionfs in the BSD's; maybe
the existing functions of unionfs can demonstrate the distinctions
between a file hierarchy and store containing a filesystem?

Thanks,
Richard
kreuter at ausar.rutgers.edu