Re: Directories traversal (was Re: the deadly hypercube of death, or: handling permissions)

[Marcus Brinkmann]

At Fri, 28 Apr 2006 13:44:35 +0200,
Pierre THIERRY <address@hidden> wrote:
> 
> Scribit Marcus Brinkmann dies 28/04/2006 hora 01:54:
> > > Where will these user mutable filesystem reside?
> > Whereever we want it to be.  It could be in the form of a union
> > filesystem.
> 
> Have you already thought about how exactly this could be implemented? In
> Unix, users typically occupy a partition, where they can have quotas or
> not. The FS deals with allocating space to each user. If each user has
> it's FS, we must at least add an allocator layer between the storage
> itself and the user. This could be LVM or some sort of it.

The current take is to use the space bank, which provides an overall
system quota on the amount of memory a user (or a group of users
collectively) can consume.

From this quota, you have to run all your programs, including the file
server that implements your "home directory".

> Speaking about the users' dir, I was also wondering about how backups
> could be made of a FS that contains arbitrary translators.

You backup a consistent snapshot of the complete persistent storage
(not sure how you do _that_ but it's certainly possible).  Maybe
Jonathan can give some more info.  Do you need a second-level snapshot
mechanism to take a snapshot of the disk?

> In the case of $HOME, it would probably be a waste if you backup the
> entire space allocated to the user, in same sense that is a waste to
> backup the content of a block device instead of the files contained in
> it. I think either the user gives a transitive read-only capacity to
> it's $HOME, or a capacity whose invocation returns a compact view of the
> underlying data of it's FS. It could even be encrypted, if the backup
> user should not be able to read $HOME.

If all important information can be serialized, you can do that.

> > Think outside the box.  We are not talking about a Unixish system
> > here.
> 
> I think it could be very helpful if we also discuss some times the
> higher-level organization of Hurd-NG, because I'm pretty sure I'm not
> the only one that has not a clear view of how it could be in practice.

I would love to.  There are open questions there as well that are
important to address.
 
> As we were talking about use cases, it would also probably help finding
> ones, and also finding contradictions or usability issues in the design.

Yes, very important.

Thanks,
Marcus