Proxmox LXC Security | Unprivileged Container Guide

LXC containers share the host kernel, so their security depends on how root inside the container maps to the host. In a
privileged container, container root is host root in the eyes of the kernel; a container escape is a host compromise.
Unprivileged containers map container root to a high, harmless host UID, which is why they are the safer default.

Our LXC creation guide shows how to build one; this guide shows how to keep it contained.
For VM-versus-container tradeoffs see Docker vs LXC.

Privileged Versus Unprivileged

The difference is a UID mapping. Unprivileged containers shift every container UID by 100000 on the host, so a process running as root inside owns nothing outside the container.

architectureuid-map.txt

  container uid 0     -> host uid 100000   (unprivileged)
  container uid 1000  -> host uid 101000
  container uid 0     -> host uid 0        (privileged: avoid)

Privileged vs unprivileged — The UID mapping is the boundary

Create Containers Unprivileged by Default

In the wizard keep Unprivileged container ticked. On the CLI pass --unprivileged 1.

bashpct create

pct create 210 local:vztmpl/debian-12-standard_12.7-1_amd64.tar.zst \
  --hostname app01 --unprivileged 1 --cores 2 --memory 2048 --swap 512 \
  --rootfs local-lvm:8 --net0 name=eth0,bridge=vmbr0,firewall=1

Creating an unprivileged CT — Least privilege from the start

Understand the ID Mapping

The default mapping covers 65536 IDs starting at 100000. Custom mappings are only needed when a container must share files with the host at the same numeric IDs.

bash/etc/pve/lxc/210.conf (custom, optional)

lxc.idmap: u 0 100000 65536
lxc.idmap: g 0 100000 65536

WarningAny custom mapping that maps container IDs to real host IDs weakens isolation. Add only what you can justify.

UID and GID mapping — Container IDs shifted on the host

Bind Mounts and File Permissions

A host directory mounted into an unprivileged container must be owned by the shifted IDs, otherwise the container sees it as owned by nobody. Mount read-only when the container only reads.

bashhost

mkdir -p /srv/shared && chown -R 100000:100000 /srv/shared

bashmount point

pct set 210 -mp0 /srv/shared,mp=/mnt/shared,ro=1

Bind mount permissions — Own by the shifted UID

Restrict Container Features

Every extra feature widens the attack surface. Enable nesting only for Docker-in-LXC, and keyctl/fuse only when an application requires them.

bashreview

pct config 210 | grep -E 'features|unprivileged'
# only if truly needed:
pct set 210 --features nesting=1

Container features — Off unless required

Isolate the Network

Place tiers on separate VLANs and enable the per-container firewall with a default-drop policy, allowing only required ports (see the firewall guide and VLAN guide).

bashvlan tag + firewall

pct set 210 --net0 name=eth0,bridge=vmbr0,tag=30,firewall=1

Limit CPU, Memory and Disk

A runaway process in one container should not starve the host or its neighbours.

bashlimits

pct set 210 --cores 2 --cpulimit 1.5 --memory 2048 --swap 512

Audit Existing Containers

List every container and find those still privileged. Migrate them by backing up and restoring as unprivileged (see restore guide).

bashfind privileged CTs

pct list
grep -L '^unprivileged: 1' /etc/pve/lxc/*.conf

Audit privileged containers — Find and migrate

Quick Reference

Unprivileged by default; nesting/keyctl/fuse only when required
Bind mounts owned by shifted IDs; firewall on; resource limits set

Proxmox LXC Security: Unprivileged Containers, ID Mapping and Limits