On 11/02/15 13:35, sf...@users.sourceforge.net wrote:
> OmegaPhil:
>> The crux of the problem is aufs returned success when unmounting, when
>> it shouldn't have as not everything had released references to the
>> volume. That was my question.
> 
> Still I don't understand what you wrote.
> It would be better if you write a simple script or C program
> (in system-programming) to reproduce the problem on a generic system
> other than yours.
> 
> For example,
> 
> - you have an entry in /etc/fstab, let's call the fs-type is sillyfs.
> - your cgmanager is running as usual, and it holds the sillyfs mounted.
> - you try unmounting the sillyfs in another namespace and FAILED.
> 
> Assuming this behaviour is what you want, if you replace sillyfs by
> aufs, you would succeed unmouting in the last step. And you think this
> aufs behaviour is a problem, right? If so, I'd ask you to represent it
> in a simple script. And does there such sillyfs exist in real world?
> What is it?
> 
> 
> J. R. Okajima


I'm not taking this further at the moment, I'm simply letting you know
that currently aufs does not know when a volume genuinely unused, and
therefore OK to unmount (the underlying problem of cgmanager breaking
stuff is now easy to work around, but it exposes this more serious issue).

I suspect this is due to bad handling of namespace stuff (that I know
nothing about) on behalf of cgmanager, but even in such a case, when an
unmount succeeds, that is taken by the user as a guarantee that nothing
is hooked into the volume(s) managed by aufs, or any other filesystem.

If anyone is interested, this is trivial to reproduce in Debian Testing
currently (cgmanger v0.35-1 or below):

1. aufs volume maintained in fstab.
2. Ensure sysvinit is in place, not systemd: sudo aptitude install
sysvinit-core
3. Start the system up into runlevel 2, unmount the aufs volume - this
will succeed.
4. Try to move the old mountpoint - it will fail.
5. sudo /etc/init.d/cgmanager restart
6. You can move the mountpoint directory now.

Thanks for your time earlier - I'll be active on the cgmanager bug.

Attachment: signature.asc
Description: OpenPGP digital signature

------------------------------------------------------------------------------
Dive into the World of Parallel Programming. The Go Parallel Website,
sponsored by Intel and developed in partnership with Slashdot Media, is your
hub for all things parallel software development, from weekly thought
leadership blogs to news, videos, case studies, tutorials and more. Take a
look and join the conversation now. http://goparallel.sourceforge.net/

Reply via email to