Richard Stallman wrote:
>
> AutoFS handles automounting (/var/autofs/misc/floppy under debian) but
> not fully, because the floppy is always mounted as root, preventing the
> user from using the floppy. This could be worked around by making the
[...]
> Actually I think it would be fine to make the floppy world-writable.
> Except on the few machines where there are lots of users actually
> using floppies, that is what I would always do anyway. And on those
> few machines, maybe it is better to require explicit mount commands.
I always assume a network environment in which other users are present
on the machine. These people can watch for the VFS: floppy changed
mesages and since it is mounted world-readable and owned by root (DOS
floppies are this way; an ext2 floppy certainly wouldn't be) they can
snoop on your potentially personal data. Having to explicitly umount
the floppy when done doesn't solve this problem, and brings the whole
issue of usability back up.
> Condition 3 needs a VFS hook, I believe. Yes, autofs will auto unmount
> after a specific amount of time, but after ejecting the floppy, I can
> still do a 'ls'. The buffer cache should be cleared when it is ejected,
> and I believe the VFS gets this signal.
>
> This seems like a real issue. Programs that were actually trying to
> do I/O on the floppy will have to lose, of course--unless people write
> some very hairy feature to remember various floppies when they are not
> on line. But the system should continue gracefully even though a disk
> is removed.
The system currently does continue gracefully - it just keeps old data
laying around, and you can't insert a new disk and have it immediately
recognized. I don't remember explicity testing this, but I believe a
new disk is recognized upon a 'cache-miss'. I don't know what happens
if you eject a DOS disk and insert an ext2 diskette, but it handles it
reasonably. If you try it and see old data, simply umount the directory
and cd back into it.
I agree with you that there should be no to minimal overhead by adding a
VFS hook - all that has to be done is force a umount of the device, and
(although I haven't looked at the code) I see no reason why this can't
be done. The hook already exists, it's just not being used by anything
other than syslog. One issue may be that it gets the signal too late
(eg, at access time, not insertion).
Either way, it seems that the real solution is the supermount patches,
which do not appear to be part of the stock kernel. These things have
been floating around since before kernel 2.0 it seems so I don't know
why they are still not integrated.
Here is a recent (~10 month old) announcement:
http://207.178.22.52/articles/business/021.html
This appears to be the homepage (updated Apr 18, 00):
http://www.geocities.com/SiliconValley/Lab/8144/supermount.html
And here is my google search for more info:
http://www.google.com/search?q=linux+supermount&meta=lr%3D%26hl%3Den
Finally, I'd like to apologize for dropping your [RMS] email address
from this discussion. I figured that you get enough email as is, and
probably wouldn't want to follow the discourse. I realize that was not
my decision to make, and apologize.
Christopher
