lazy-umount cwd and ..

lazy-umount cwd and ..

[Tim Hockin]

I notice that a process that is in a dir which gets lazy-unmounted
suddenly sees it's current dir change and its '..' dir points back to
itself.

I'm not sure it's a huge deal, but we have a half-patch floating around
that changes the behavior such that the unmounted mnt->mnt_parent is
retained and unreferenced when the mnt is finally released.  This seems to
make any process which is in the unmounted mount not see anything
different, but does not let any new processes into the mount.

Minor, but friendly.

Should I bother to polish this patch off and send it, or is it just not
something we want to care about?

Tim

Re: lazy-umount cwd and ..

[viro]

On Wed, May 05, 2004 at 09:44:33PM -0700, Tim Hockin wrote:
> I notice that a process that is in a dir which gets lazy-unmounted
> suddenly sees it's current dir change and its '..' dir points back to
> itself.
> 
> I'm not sure it's a huge deal, but we have a half-patch floating around
> that changes the behavior such that the unmounted mnt->mnt_parent is
> retained and unreferenced when the mnt is finally released.  This seems to
> make any process which is in the unmounted mount not see anything
> different, but does not let any new processes into the mount.
> 
> Minor, but friendly.
> 
> Should I bother to polish this patch off and send it, or is it just not
> something we want to care about?

No.  This is simply wrong - one of the situations when you want lazy-umount
is getting a stuck filesystem (e.g. NFS mounted hard) and wanting to get
it out of the way, so that stuff it's mounted on could be unmounted clean.

So we definitely don't want to keep anything pinned down.

Re: lazy-umount cwd and ..

[Tim Hockin]

On Thu, May 06, 2004 at 07:46:17AM +0100, viro@parcelfarce.linux.theplanet.co.uk wrote:
> On Wed, May 05, 2004 at 09:44:33PM -0700, Tim Hockin wrote:
> > Should I bother to polish this patch off and send it, or is it just not
> > something we want to care about?
> 
> No.  This is simply wrong - one of the situations when you want lazy-umount
> is getting a stuck filesystem (e.g. NFS mounted hard) and wanting to get
> it out of the way, so that stuff it's mounted on could be unmounted clean.
> 
> So we definitely don't want to keep anything pinned down.

I'll buy that.  I guess it's not worth dreaming up something else to make
that be less surprising to apps with cwd in a lazy umounted mount.

Cheers.