Kernel requirements

Bailey adapts to what the running kernel offers. Nothing here is a hard requirement for the tool to run; each missing feature removes a layer, and bailey reports what it could not enforce.

Summary

Feature	Needed for	Since	Without it
Landlock	Filesystem policy	5.13	No filesystem or network enforcement, warned
Landlock ABI 4	Network rules	6.7	Network policy not enforced
Landlock ABI 5	Device ioctl restrictions	6.10	That subset is skipped
Unprivileged user namespaces	--isolate	long-standing, often disabled	Falls back to Landlock and seccomp, warned
cgroup v2 with delegation	Resource limits	long-standing	Limits skipped, warned
seccomp filters	Syscall denylist	long-standing	Run fails to establish
BTF (/sys/kernel/btf/vmlinux)	Audit backend	5.x with CONFIG_DEBUG_INFO_BTF	Audit unavailable

Checking

# Landlock active?
grep landlock /sys/kernel/security/lsm

# Unprivileged user namespaces?
cat /proc/sys/user/max_user_namespaces
sysctl kernel.unprivileged_userns_clone 2>/dev/null

# cgroup v2?
stat -fc %T /sys/fs/cgroup

# BTF?
test -e /sys/kernel/btf/vmlinux && echo yes

Landlock ABI negotiation

Landlock's access rights grew across versions, and asking for a right the kernel does not know is an error rather than a no-op. Bailey requests its target ABI in best-effort mode: the kernel applies the rights it knows and the rest are dropped, and bailey reports what was not enforced.

The practical consequence is that on a kernel older than 6.7 your network policy is not applied even though the config is valid. bailey show prints the policy as written; it does not tell you what the kernel will do with it.

Distribution notes

Unprivileged user namespaces are the most commonly disabled feature. Debian and Ubuntu have shipped restrictions at various points, and AppArmor profiles can block the unshare even when the sysctl permits it. Bailey probes by attempting the real setup in a throwaway child rather than reading the sysctl, so its report reflects what will actually happen.

BTF requires CONFIG_DEBUG_INFO_BTF=y. Most mainstream distribution kernels enable it; minimal or custom kernels often do not.

Landlock must be in the kernel's lsm= boot parameter list. Most distributions include it by default on kernels that support it.

Architecture

x86_64 and aarch64 are supported for the seccomp filter. Other architectures fail to build the filter and therefore fail to establish a sandbox.

The eBPF audit programs attach by BTF id and take their argument layout from the running kernel, so they are not tied to one architecture. They have been exercised on x86_64.