>
> On 16-Mar-99 H. Peter Anvin wrote:
> > There are additional problems, as well:
> >
> > - autofs v3 will allow processes to access the tree after only the
> > root has been accessed.
>
> I'm not sure I follow you. You can't name parts of the tree without going
> through the root.
>
Think about what happens when you mount the root filesystem, but
before another filesystem gets mounted. Now another process comes
along and tries to mount a side filesystem.
> > - I think you might find that adding directory trees to autofs v3 is a
> > bit more challenging than you thought, due to the fact that in autofs
> > v3, all directories but the root are synthetic empty directories.
>
> That's not too hard to fix. It just requires that autofs has more
> general purpose directories. I'm currently adding an inode hash
> table to keep track of all autofs inodes, and moving the directory
> hash into each directory inode. This cleans up handling of symlinks
> too. I'm still looking at the best way of dealing with expiration.
This is when you really want multithreading. Believe me!
-hpa
