On Mon, 7 Apr 1997, H. Peter Anvin wrote:
> >
> > > I have made a test release of autofs, a kernel-assisted automounter
> > > for Linux. Currently it only supports YP (NIS) maps; I'm planning to
> > > add support for a few more kinds of maps Really Soon Now[TM]. Expect
> > > the versions to change quickly as the code stabilizes.
> > >
> > > The distribution is at:
> > >
> > > ftp.kernel.org:/pub/linux/daemons/autofs
> >
> > What is the difference between autofs and amd? Dan told me that it
> > was supposed to be a fix for some brokeness in amd, but he could not
> > tell me which brokeness he was talking about.
> >
>
>
> 1. amd does everything in user space. Consequently, it (a) has to
> simulate an NFS daemon and (b) cause two context switches even for
> simple path lookups (i.e. the target is already mounted!) This is
> *SLOW*.
>
> 2. amd is a big, monolithic monster, using a very odd format for maps.
>
> 3. amd is just plain broken code (we have had no end of trouble with
> it here.)
>
> 4. "pwd" in an amd-mounted directory gives a pathname that may not
> exist later during execution.
>
> #1 and #4 are the reasons to bring it into the kernel. #2 and #3 are
> just the need to get a clean code base.
>
> -hpa
>