From: Herbert Poetzl (herbert_at_13thfloor.at)
Date: Tue 03 Aug 2004 - 17:06:34 BST
On Tue, Aug 03, 2004 at 05:05:16PM +0200, Jörn Engel wrote:
> To avoid sending a pure test message and being banned from the list,
> I'll have to add something useful as well... cowlinks.
> Matt and Herbert have told me they really want that project to get
> finished asap - reason enough for me to give it some more focus again.
> The patches on the webpage got updated and now include a fix from
> Herbert. Thanks for that one! If noone beats me to it, I'll check
> with Andrew Morton whether the three infrastructure patches can be
> included into -mm.
sounds good, btw, I heard the copious locking was
removed? (matt told so, if I got him right)
> Madcow is called mad because it's just that. You have to decide
> whether you want hard links or cow links. Either get rid of hard
> links completely and confuse some programs. Or use them for some
> directory trees only and get problems whenever moving between
> madcow-trees and regular trees. Not good.
maybe some information how the COW issue is currently
addressed in linux-vserver. we are using something
called immutable linkage invert or immutable unlink
(or hundred other names folks invented ;) but it
basically works like this:
you create a hardlink of a file for several servers
(sometimes to a template, sometimes just between
the servers), then set the immutable attribute (which
can not be reverted inside a virtual server), and
an additional attribute (now called iunlink) which
permits the user to unlink such an otherwise immutable
hardlink reference ...
this is sufficient for most package updates and
similar but of course a smart solution (CoW*) would
be appreciated ...
btw, the process of identifying identical files and
combining them into an immutable but unlinkable hard
reference is (here) called unification, and is done
by some userspace tools ...
> A real cowlink patch is started but still quite ugly. It's a bit
> embarrasing, but if people don't care and want to test it anyway,
> feedback is always welcome. I'll try to add some design details to
> the webpage soon.
> So if anyone want to participate in any way - coding, testing,
> discussion, userspace programs, documentation, whatever - please let
> me know. Help is always appreciated.
I'm willing to help where it is required, as much
as my time permits ...
> Oh yes, the web page is here:
> Schrödinger's cat is <BLINK>not</BLINK> dead.
> -- Illiad
> Vserver mailing list
Vserver mailing list