Coda File System

Re: Win 95 stuff (was Re: coda-4.6.0-pre1 available)

From: Derek Fawcus <>
Date: Thu, 14 May 1998 09:12:01 +0100
On Thu, May 14, 1998 at 03:56:31AM -0400, Michael Callahan wrote:
> On Thu, 14 May 1998, Derek Fawcus wrote:
> > On Wed, May 13, 1998 at 06:10:31PM -0400, Michael Callahan wrote:
> >   Already got them.  Plus a few others including the two Nutshell books on
> > Win95 and NT filesystems.  There's a bit of overlap,  but between them they
> > seem to cover everything.
> Ha!  In the windows world, nothing covers everything!

   How true.  But one can always reverse enginner the executables... :-)

  [ snip ]

> IIRC win3.1 is out of the question: the IFSMGR didn't appear until 3.11.

  Yeah it's not there.  However the support mechanism (vxd's and interrupt
hooking is),  so in theory one _could_ do it.  However... it's probably
not worth the effort.

> 3.11: I'm not sure, but skeptical.   IFSMGR was undocumented and in God
> knows what odd state, and I'm not sure in what state in-kernel networking
> was, either.

  The documentation for the 3.11 IFSMGR is basically the win95 stuff then
suck it and see.  The networking,   believe works (after a fashion),
certainly I used the vxd based TCP/IP stack under 3.11

> Btw, MMAP.VXD doesn't actually do mmap'ing of files: it just allows a
> process to have a flat memory address space in which it can allocate pages
> whereever it likes.  The DPMI services in W95 don't let you do this: you
> can ask for a memory block, but you have no control over the linear
> address at which it appears.  To support RVM, it must be possible to
> allocate memory blocks at the same location in multiple runs of the
> program.

  Ah - so thats what the comment was about.  I've just been reading the
RVM manual,  and so was starting to wonder.  I thought (from the manual)
the identical mapping requirement was a application imposed choice - so
it would appear Venus does this.  From what you've said above,  it actually
looks as if a 3.11 version is feasible.  However it may _require_ that
the VFAT stuff & 32 bit file access is used - if I remember correctly
the file API hooking in 3.11 is not as complete and calling into DOS for
file access may well be a problem.

  I guess that for the mini cache you're using the ring 0 file services
to access the underlying cache files?  

Derek Fawcus                                          
Spider Software Ltd.                                        +44 (0) 131 475 7034
Received on 1998-05-14 04:14:12