interfaces: allow loopback as a block-device

[sabaini]

Allowing loopback devices in the block-devices interface would be convenient for testing and proof-of-concept setups for users of snaps that consume block devices such as MicroCeph

Use case 1: allow a snap to utilize a preconfigured loopback bdev, for
instance allow MicroCeph to use a loopback device as an OSD for
testing or other non-performance critical work

Use case 2: allow a snap to create a loopback device automatically;
e.g. for the above scenario MicroCeph could automatically set up a
loop bdev for use as an OSD

[mvo5]

This seems okay to me, but let's see what security and Samuele think.

[pedronis]

what's the use case in mind here? I wonder if a separate interface only about loopback devices would (also) make sense

[sabaini]

what's the use case in mind here? I wonder if a separate interface only about loopback devices would (also) make sense

There's two (related) use cases I'm thinking of, both relating to our MicroCeph snap. To lower the barrier of entry for MicroCeph users it would be great if users could deploy a fully functional Ceph without needing a separate physical block device -- which atm they need for creating an OSD. If they need to run MicroCeph virtualized even more so since e.g. Multipass doesn't even support separate block devices

Use case 1: allow a snap to utilize a preconfigured loopback bdev, for
instance allow MicroCeph to use a loopback device as an OSD for
testing or other non-performance critical work

Use case 2: allow a snap to create a loopback device automatically;
e.g. for the above scenario MicroCeph could automatically set up a
loop bdev for use as an OSD

[sabaini]

This shows a failure to build but I don't suppose this is due to this PR, right?

[alexmurray]

LGTM - block-devices is already super-privileged and adding support for loopback devices doesn't appreciably increase the attack surface here.

[pedronis]

+1