[lxc-users] Is an unprivileged LXC where the host user itself is mapped to 0 less secure of one where one of its subids is mapped to 0, and why?

[Fabio Tudone ([e-mail])]

Hi,

instead of creating "regular" LXC unprivileged containers where all the
users are mapped to (unprivileged) subuid/gid of my host user, I'm
considering a mapping where my host user itself will be mapped to user 0
(root). They'd be very slim single-app containers.

The reason is that in this way I don't need the rootfs directory
subtree, which resides in my user's home, to be namespace-|chmod|to a
different user and I can delete it with a plain|rm|instead of a
namespace one.

Is this kind of LXC less secure than the "regular" one, and why is it?
What could happen in the worst case?

Thanks,

-- Fabio

[Serge Hallyn]

Post by Fabio Tudone (***@paralleluniverse.co)
Hi,
instead of creating "regular" LXC unprivileged containers where all
the users are mapped to (unprivileged) subuid/gid of my host user,
I'm considering a mapping where my host user itself will be mapped
to user 0 (root). They'd be very slim single-app containers.
The reason is that in this way I don't need the rootfs directory
subtree, which resides in my user's home, to be namespace-|chmod|to
a different user and I can delete it with a plain|rm|instead of a
namespace one.
Is this kind of LXC less secure than the "regular" one, and why is
it? What could happen in the worst case?

Don't know what you mean by the regular lxc. Root in your container
will have full rights to your user-owned files on the host, but that's
it. That is no different than if you map your host uid into the
container to any other id (since root in the container will have privilege
over your host-uid-owned files in that case).

So in general I recommend against mapping your host user into the container,
but it has its uses (and i do it in one container).

[Fabio Tudone ([e-mail])]

I really meant unprivileged containers created for example by the
download template.

Usually subuids have an empty set of capabilities on the host, which
means that even if they were accessing host resources they wouldn't be
able to do much, so they are effectively "jailed". On the other hand if
a regular host's user is mapped into the container it would
theoretically be able to access host resources with the same
capabilities granted to the host's user, isn't it?

On a more practical level what could be the security implications? Are
there host resources that a malicious program could compromise when
running in a container with the capabilities of a regular host user
mapped in there? Even because of (hypothetical) system issues / bugs /
vulnerabilities. Can someone think of actual examples?

Thanks.

-- Fabio

Post by Serge Hallyn

Post by Fabio Tudone (***@paralleluniverse.co)
Hi,
instead of creating "regular" LXC unprivileged containers where all
the users are mapped to (unprivileged) subuid/gid of my host user,
I'm considering a mapping where my host user itself will be mapped
to user 0 (root). They'd be very slim single-app containers.
The reason is that in this way I don't need the rootfs directory
subtree, which resides in my user's home, to be namespace-|chmod|to
a different user and I can delete it with a plain|rm|instead of a
namespace one.
Is this kind of LXC less secure than the "regular" one, and why is
it? What could happen in the worst case?

Don't know what you mean by the regular lxc. Root in your container
will have full rights to your user-owned files on the host, but that's
it. That is no different than if you map your host uid into the
container to any other id (since root in the container will have privilege
over your host-uid-owned files in that case).
So in general I recommend against mapping your host user into the container,
but it has its uses (and i do it in one container).
_______________________________________________
lxc-users mailing list
http://lists.linuxcontainers.org/listinfo/lxc-users

[Serge Hallyn]

Post by Fabio Tudone (***@paralleluniverse.co)
I really meant unprivileged containers created for example by the
download template.
Usually subuids have an empty set of capabilities on the host, which
means that even if they were accessing host resources they wouldn't
be able to do much, so they are effectively "jailed". On the other
hand if a regular host's user is mapped into the container it would
theoretically be able to access host resources with the same
capabilities granted to the host's user, isn't it?

No. So long as the container has an lxc.id_map entry, it will run
in a private user namespace, and will have no capabilities with respect
to the host. It will have full caps wrt any uids mapped into the
container, and any resources created by the container.

Post by Fabio Tudone (***@paralleluniverse.co)
On a more practical level what could be the security implications?
Are there host resources that a malicious program could compromise
when running in a container with the capabilities of a regular host
user mapped in there? Even because of (hypothetical) system issues /
bugs / vulnerabilities. Can someone think of actual examples?

yes.

[Fabio Tudone ([e-mail])]

Post by Fabio Tudone (***@paralleluniverse.co)
On a more practical level what could be the security implications?
Are there host resources that a malicious program could compromise
when running in a container with the capabilities of a regular host
user mapped in there? Even because of (hypothetical) system issues /
bugs / vulnerabilities. Can someone think of actual examples?

yes.

Could you expand on that? What could happen for example? I'm no security
expert but I'm interested in understanding the implications.

Thanks,

-- Fabio

[Fajar A. Nugraha]

On Mon, Oct 5, 2015 at 11:58 PM, Fabio Tudone

Post by Fabio Tudone (***@paralleluniverse.co)

Post by Fabio Tudone (***@paralleluniverse.co)
On a more practical level what could be the security implications?
Are there host resources that a malicious program could compromise
when running in a container with the capabilities of a regular host
user mapped in there? Even because of (hypothetical) system issues /
bugs / vulnerabilities. Can someone think of actual examples?

yes.

Could you expand on that? What could happen for example? I'm no security
expert but I'm interested in understanding the implications.

I believe the simplest example would probably be from Stephane's blog:
https://www.stgraber.org/2014/02/09/lxc-1-0-gui-in-containers/ . User
1000 in the host is mapped to the same uid on the container, for the
purpose of easy configuration of X and sound access from the
container.

Should some security vulnerability occur that allows the user to
escape the container (or run arbitrary command inside the host), the
"escaped" user will be restricted as uid 1000, which is theoretically
still much safer compared to privileged container.

However even that non-root-on-the-host user might still cause problems:
- If you have assigned additinal permission for that user (e.g. If uid
1000 on the host is a member of "disk" group, which has write access
to block devices), the user can wreak havoc using that additional
permission
- If that user has created several containers, the "escaped" user can
compromise other containers belong to that user

So short version:
- much safer than privileged container
- can potentially still cause problems as that uid
- use different id_map (with uids not used on the host) for each
container if you want maximum security