Okay, I'm planning to do outline the basic data structures of autofs
v4 over the next several days. Some features are already a given:
* Multimounts/scaffolding (/net being a special case of this).
* Arbitrary mount point topology, without needing the
spawn-an-automounter hack.
* Mounting-process information passed to daemon.
* Most of the filesystem data will live in "ready-to-eat" kernel data
structures (inodes, dentries) as opposed to separate backing store.
A few frequently requested features that are *not* included, so don't
ask:
* Direct mounts. These appear to require a significant slowdown to the
entire VFS, and I do not believe this is justifiable.
* lofs. Not because I don't think it is a good idea if it can be done
right, but because it is a feature orthogonal to autofs.
* Replicated servers. This is an NFS issue, not autofs. This means
it belongs in mount(8) and the NFS kernel code. I completely refuse
to hack in a feature into autofs when it is clear it should be
provided elsewhere, *especially* when it is a feature like this when
it would be more useful if provided elsewhere. After all, why
should you be able to mount a replicated server from autofs but not
from /etc/fstab??
However, if there is some key feature that I'm missing, please speak
up now: please send an email explaining the feature and its
justification, and I will try to fit it into the design.
Oh, autofs v4 will be targetted for the 2.3 kernels and libc6. 2.3
because I am planning to ask for certain VFS changes, which probably
won't be possible in the 2.2 kernel series, and libc6 because I need a
thread-safe dynamic linker, which isn't available for libc5.
-hpa
