ZFS Union

[Teske, Devin]

Hi List,

I had an idea recently that I thought might be worth chasing.

I've thought for a long time that it would be really great if I could have (for the purpose of jails):

+ ZFS filesystem /vm/master
+ ZFS filesystem /vm/unit1
+ ZFS filesystem /vm/unit2
+ ZFS filesystem /vm/unit3

Unlike a ZFS snapshot/clone system, where changes in /vm/master made beyond the snapshot for which /vm/unit{1,2,3} were cloned-from do not effect /vm/unit{1,2,3}…

What if there was a way to layer /vm/unit{1,2,3} in a union manner to be the top-layer above /vm/master.

I believe that UnionFS isn't of help here specifically because I believe it to not support the following use-case example:

Step 1. touch /vm/master/foo

ASIDE: /vm/unit1/foo does not exist prior to Step 1

Step 2. See that now /vm/unit{1,2,3}/foo exists

(that was the litmus test for any layering filesystem, now comes the part that I believe UnionFS fails)

Step 3. Now, rm /vm/unit1/foo

Step 4. See that /vm/master/foo is still there, but /vm/unit1/foo remains gone

Step 5. Counter to Step 4 above, See that /vm/unit2/foo and /vm/unit3/foo still exist

In other words… the filesystem should be able to keep track of unlinked files as a black-list.

Enhancing ZFS to support union is quite sexy, because when you go down the rabbit hole of Steps 3-5 above, you realize you may want to (as an administrator) "reclaim" files from a lower layer. This could perhaps be tacked onto the "zfs" utility (whereas if enhancing UnionFS, a new utility would need to be born).

I would imagine a "zfs reclaim" (hypothetical fictional command) to allow the path from lower layers to become visible again, so-long-as a lower-layer hasn't black-listed it from an even lower-layer.

The end-run production use of this would be to allow jails to "inherit" files from a lower layer but unlike a snapshot system, continue to inherit perpetually in realtime. Going into a lower layer and making a change would immediately percolate that change to all the jails layered on top.

It would also mean nice lean deltas. Layering "/foo" on top of "/" would be a quick and dirty way of cloning your base system into a new jail where all the writes go off to that directory (something existing UnionFS technologies already do) and -- something not done by existing UnionFS technologies -- unlinked files will not appear (giving the idea that, while chroot'd or jail'd into that directory, you have more control over your universe because you "rm" a file and it goes away; but of course [hypothetically] an administrator in parent host to the jail can "reclaim" it for you from a lower layer if you want him/her to do-so).

Thoughts?
-- 
Devin

_____________
The information contained in this message is proprietary and/or confidential. If you are not the intended recipient, please: (i) delete the message and all copies; (ii) do not disclose, distribute or use the message in any manner; and (iii) notify the sender immediately. In addition, please be aware that any message addressed to our domain is subject to archiving and review by persons other than the intended recipient. Thank you.