Am I right in thinking that a UID in one container is not necessarily
equivalent to the numerically equivalent UID in another container?

If that's the case then the key management code will need changing as it
assumes all keys belonging to one numeric UID eat out of the same quota and
the numeric UIDs are used in security checks.

Furthermore, processes in one container can access keys created by a process
in another container by ID.  Is this desirable or not?

David
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

yes. If I understand correct key management requires containerization (i.e. "virtualization")
as well other subsystems like IPC dealing with IDs.

Processes from one container should not be able to access keys from another container.

David Howells wrote:
> Am I right in thinking that a UID in one container is not necessarily
> equivalent to the numerically equivalent UID in another container?
> 
> If that's the case then the key management code will need changing as it
> assumes all keys belonging to one numeric UID eat out of the same quota and
> the numeric UIDs are used in security checks.
> 
> Furthermore, processes in one container can access keys created by a process
> in another container by ID.  Is this desirable or not?
> 
> David
> _______________________________________________
> Containers mailing list
> Containers@lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/containers
> 
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

Quoting David Howells (dhowells@redhat.com):
> 
> Am I right in thinking that a UID in one container is not necessarily
> equivalent to the numerically equivalent UID in another container?
> 
> If that's the case then the key management code will need changing as it
> assumes all keys belonging to one numeric UID eat out of the same quota and
> the numeric UIDs are used in security checks.
> 
> Furthermore, processes in one container can access keys created by a process
> in another container by ID.  Is this desirable or not?
> 
> David

Yes, the confusion comes from using the word 'container' which doesn't
really exist.  The user namespaces (CLONE_NEWUSER) are what provide
separate of uids.  We want uid 5 in one user namespace to have
completely separate set of keys from uid 5 in another user namespace.

This isn't yet a crucial thing to get right as the user namespaces are
only partially implemented, but it's certainly a good thing to be looking
at and fix when convenient to do so.  It looks like maybe just adding
a struct user_namespace * to a struct key should suffice.

-serge
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

Serge E. Hallyn <serue@us.ibm.com> wrote:

> It looks like maybe just adding a struct user_namespace * to a struct key
> should suffice.

That's not quite sufficient.  The per-UID key_user structs also need to be
differentiated.  Unfortunately, I can't just merge it into user_struct as I
then end up with a reference loop user_struct -> uid_keyring -> user_struct.

Rooting the key_user trees in user_namespace will probably do the trick.

A couple of questions:

 (1) A process may inherit a session keyring over clone().  Should this be
     discarded if CLONE_NEWUSER is set?  Or would I need to copy it?

 (2) In a recent patch, I've given the root user its own quota limits.  Is UID
     0 always the root user in any container?  Or would it make more sense
     just to scrap the per-root quota limits?

David
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

Quoting David Howells (dhowells@redhat.com):
> Serge E. Hallyn <serue@us.ibm.com> wrote:
> 
> > It looks like maybe just adding a struct user_namespace * to a struct key
> > should suffice.
> 
> That's not quite sufficient.  The per-UID key_user structs also need to be
> differentiated.  Unfortunately, I can't just merge it into user_struct as I
> then end up with a reference loop user_struct -> uid_keyring -> user_struct.
> 
> Rooting the key_user trees in user_namespace will probably do the trick.
> 
> A couple of questions:
> 
>  (1) A process may inherit a session keyring over clone().  Should this be
>      discarded if CLONE_NEWUSER is set?  Or would I need to copy it?

Someone else may have stronger feelings about this.  Personally so long
as a container setup program has a way of discarding the keyring
manually I think that's fine.

>  (2) In a recent patch, I've given the root user its own quota limits.  Is UID
>      0 always the root user in any container?  Or would it make more sense
>      just to scrap the per-root quota limits?

Yeah uid 0 may not have a bunch of privileges, but it is still the root
user.

thanks,
-serge
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers