References

The kernel docs and man pages this subsystem encodes. Cite specific sections in the source when interpretation is non-obvious.

Man pages

• sysctl(8) — the userspace tool whose surface this subsystem mirrors. The -a / -w / -p semantics, dot-form key naming, and the --system workflow that walks /etc/sysctl.d/*.conf are all here. Linx exposes the -r / -w equivalents (read/write); the --system conf-file applier is deferred to a consumer.
• proc(5) — the canonical reference for /proc/sys/, the layout of subtrees (net/, kernel/, vm/, fs/, …), and the per-namespace routing rules. The "Files and directories" section under /proc/sys/ is the source of truth.
• namespaces(7) — overview of which namespaces own which parts of the sysctl tree (network for net.*, UTS for kernel.hostname, IPC for kernel.shm* / kernel.msg* / kernel.sem / fs.mqueue.*).
• network_namespaces(7) — specifically the per-netns /proc/sys/net/ semantics that make net.ipv4.ip_forward meaningful inside a container.
• uts_namespaces(7), ipc_namespaces(7) — analogous references for the UTS / IPC subtrees.
• setns(2) — the syscall the cross-namespace NIF wraps for reads and writes. Particularly the "Description" of valid namespace types and the "Notes" on what passing 0 for the second argument means (autodetect from the file's type).

Kernel documentation

• Documentation/admin-guide/sysctl/ — the kernel's own per- subtree docs:
  • kernel.rst — kernel.* knobs (hostname, printk, panic, randomize_va_space, …)
  • net.rst — top-level net.* knobs; per-protocol specifics live alongside the protocol's docs
  • vm.rst — vm.* knobs (swappiness, overcommit, dirty_ratio, …)
  • fs.rst — fs.* knobs (file-max, inotify limits, mqueue, …)
  • user.rst — per-userns user.max_*_namespaces nesting limits
  • abi.rst — abi.*
• Documentation/networking/ip-sysctl.rst — the exhaustive reference for every net.ipv4.* and net.ipv6.* knob.
• Documentation/networking/...sysctl.rst — analogous per- subsystem files (tcp.rst, nf_conntrack-sysctl.rst, …).

Adjacent userspace tooling (for context, not implementation)

• procps-ng (sysctl(8)) — the canonical userspace tool. Worth skimming the source for edge cases in key normalization (the / vs . separator interchangeability, quoting of values containing whitespace).
• systemd-sysctl(8) — systemd's variant of the conf-file applier; same sysctl.d semantics with some extensions. Relevant as a reference design for an eventual consumer-side conf-file layer (not in Linx).
• nerves_system_* defconfigs — for the Nerves use case, the set of CONFIG_* kernel options that determines which sysctls exist on a given device. Linx.Sysctl.list/0 is the way to discover what's actually present at runtime.

In-repo cross-references

• Linx.Mount — the :in :: :self | {:pid, n} | {:path, p} option shape and the setns-on-a-throwaway-pthread pattern. Linx.Sysctl's cross-namespace NIF is structurally identical.

• lib/linx/mount.ex and c_src/linx_mount.c — the precedent the cross-namespace implementation mirrors end-to-end.
• Linx.User — the pure-Elixir-procfs precedent (no NIF, no Port).
• lib/linx/cgroup.ex — another pure-procfs subsystem, useful as a reference for %Linx.Cgroup.Error{}-style structured errors that Linx.Sysctl.Error mirrors.