Thus spoke H. Peter Anvin on 24-Feb-99 :
>>
>> My orginal solution was, to make /home a autofs which manages only symbolic
>> links to an /import automounter mount point. It worked find: each filesystem
>> was mounted only once. But since the symlink were not removable, I was screw
>> once a user has to be moved, or I did a typo at installing a new user.
>
> This I'd rather solve by Solaris-style pseudomounts, again due to the
> namespace issue (it may not be an issue for you, but it is a cause of
> frequent complaints.) Solaris allows you to specify a root for the
> kernel to mount, and the kernel will internally maintain an NFS mount
> for multiple users, exposed only at a higher level. Probably doesn't
> solve the #superblocks issue, though, but that's better handled by
> allocating superblocks dynamically anyway.
>
> I tend to believe klugy solutions just make the problem worse by
> reducing the pressure to fix the real problem...
On the other hand the AIX automounter allows you to do exactly what I want
via shared mounts
Sharing Mounts
The hostname is the server name where the file system will mount, pathname is
the directory path to mount, and subdir, when supplied, is where the symbolic
link is made to a sub-directory. This can be used to prevent duplicate mounts
when multiple directories in the same remote file system can be accessed. For
example, consider the following map for /home:
able prime:/home/prime:able
baker prime:/home/prime:baker
When a user attempts to access a file in /home/able, the automount command
mounts prime:/home/prime, but creates a symbolic link called /home/able to the
able sub-directory in the temporarily mounted file system. If a user
immediately tries to access a file in /home/baker, the automount command
creates a symbolic link that points to the baker sub-directory. The
/home/prime directory would already be mounted.
Why do you think, putting extra code in the kernel is better than
allowing a solution on deamon level? By no means I want to trigger a
discussion of a monolitic kernel vesus the mach approach. The monolitic won,
but learned how to handle modules ;-)
And I tend to believe, that, as always in live, there is not only one unique
solution to a particular problem.
Frithjof
"If you see someone without a smile, give him one of yours"
Frithjof Anders
Institut fuer Festkoerperphysik
Technische Universitaet Darmstadt
Hochschulstr. 6
64289 Darmstadt, GERMANY
Tel +49 (6151) 16-5235 email: [EMAIL PROTECTED]
FAX +49 (6151) 16-3681
