Idmaps for user namespace

Introduction

LXD runs safe containers. This is achieved mostly through the use of user namespaces which make it possible to run containers unprivileged, greatly limiting the attack surface.

User namespaces work by mapping a set of UIDs and GIDs on the host to a set of UIDs and GIDs in the container.

For example, we can define that the host UIDs and GIDs from 100000 to 165535 may be used by LXD and should be mapped to UID/GID 0 through 65535 in the container.

As a result a process running as UID 0 in the container will actually be running as UID 100000.

Allocations should always be of at least 65536 UIDs and GIDs to cover the POSIX range including root (0) and nobody (65534).

Kernel support

User namespaces require a kernel >= 3.12, LXD will start even on older kernels but will refuse to start containers.

Allowed ranges

On most hosts, LXD will check /etc/subuid and /etc/subgid for allocations for the lxd user and on first start, set the default profile to use the first 65536 UIDs and GIDs from that range.

If the range is shorter than 65536 (which includes no range at all), then LXD will fail to create or start any container until this is corrected.

If some but not all of /etc/subuid, /etc/subgid, newuidmap (path lookup) and newgidmap (path lookup) can be found on the system, LXD will fail the startup of any container until this is corrected as this shows a broken shadow setup.

If none of those files can be found, then LXD will assume a 1000000000 UID/GID range starting at a base UID/GID of 1000000.

This is the most common case and is usually the recommended setup when not running on a system which also hosts fully unprivileged containers (where the container runtime itself runs as a user).

Varying ranges between hosts

The source map is sent when moving containers between hosts so that they can be remapped on the receiving host.

Different idmaps per container

LXD supports using different idmaps per container, to further isolate containers from each other. This is controlled with two per-container configuration keys, security.idmap.isolated and security.idmap.size.

Containers with security.idmap.isolated will have a unique ID range computed for them among the other containers with security.idmap.isolated set (if none is available, setting this key will simply fail).

Containers with security.idmap.size set will have their ID range set to this size. Isolated containers without this property set default to a ID range of size 65536; this allows for POSIX compliance and a nobody user inside the container.

To select a specific map, the security.idmap.base key will let you override the auto-detection mechanism and tell LXD what host UID/GID you want to use as the base for the container.

These properties require a container reboot to take effect.

Custom idmaps

LXD also supports customizing bits of the idmap, e.g. to allow users to bind mount parts of the host’s file system into a container without the need for any UID-shifting file system. The per-container configuration key for this is raw.idmap, and looks like:

both 1000 1000
uid 50-60 500-510
gid 100000-110000 10000-20000

The first line configures both the UID and GID 1000 on the host to map to UID 1000 inside the container (this can be used for example to bind mount a user’s home directory into a container).

The second and third lines map only the UID or GID ranges into the container, respectively. The second entry per line is the source ID, i.e. the ID on the host, and the third entry is the range inside the container. These ranges must be the same size.

This property requires a container reboot to take effect.