Overview

Copy Markdown View Source

Linx.Mount shapes the filesystem hierarchy a process sees — mounting, unmounting, bind-mounting, and pivoting the root — and reads the mount table back, in the BEAM's own mount namespace or inside another process's.

A mount namespace is the set of mounts a process sees as its filesystem tree. Linx.Mount wraps the classic syscalls that build and reshape it — mount(2), umount2(2), pivot_root(2) — plus the read-side parser for /proc/.../mountinfo. It deliberately wraps the classic calls rather than the newer fsopen/fsmount family: they map one-to-one onto the mount and umount commands operators already know, and each is a single-shot call that a NIF can wrap safely without forking.

The headline capability is pivot_root: swapping a workload onto a custom rootfs before it runs — exactly what a container runtime does. The picky kernel constraints around it (new root must be a mount point, no shared propagation on ancestors, CWD inside the new root) are satisfied by a setup ritual the examples spell out.

Where it fits

Linx.Mount is a checkpoint subsystem. A child spawned with namespaces: [:mount, :pid] gets a copy of the host's mount table — so it still sees the host's /proc. Mounting a fresh /proc into the child's namespace before proceed/1 is the fix, alongside Linx.User, Linx.Cgroup, and the rest.

What makes it lifecycle-agnostic is the :in option. Every mutating verb takes in: :self | {:pid, n} | {:path, p}, targeting any process whose namespace files exist — parked at a checkpoint, fully running, or any other live pid. The mechanism is the same throwaway-pthread setns(2) trick Linx.Netlink and Linx.Sysctl use, with one twist: the worker unshare(CLONE_FS)s first, since the kernel refuses a mount-namespace setns from a thread sharing its fs_struct with the BEAM's schedulers. A container engine is the consumer that sequences make-rprivate → bind → pivot_root → umount-old around the checkpoint.

Flow

flowchart TD
    spawn["child parked at checkpoint<br/>namespaces: [:mount, :pid, …]"] --> priv
    subgraph ops["mount ops via in: {:pid, host_pid}"]
        priv["make / rprivate<br/>(detach shared propagation)"] --> bind["bind new_root onto itself<br/>(make it a mount point)"]
        bind --> pivot["pivot_root(new_root, old_root)<br/>new_root becomes /"]
        pivot --> umount["umount /old_root<br/>(flags: [:detach])"]
    end
    umount --> proceed["Linx.Process.proceed/1<br/>workload execs inside the new rootfs"]

Learn more