forcola

defmodule Forcola do
  @moduledoc """
  Leak-free external process execution.

  Every command runs under a small Rust shim (a port program, not a NIF)
  that places the child in its own process group via `setsid` and kills
  the whole group, SIGTERM then SIGKILL, when the run times out or the
  BEAM dies. The shim detects BEAM death as stdin EOF, so cleanup happens
  even on `kill -9` of the VM.

  ## Why not `System.cmd` in a `Task`?

  `Task.shutdown` closes the Erlang port, and closing a port closes pipes;
  it sends no signal. The external process runs on until it next touches a
  closed pipe, and its children are never signaled at all. See the [process
  groups guide](process_groups.html) for the full mechanism.

  ## Modes

    * `Forcola.run/2` - bounded one-shot run, mandatory timeout
    * `Forcola.Stream` - line-by-line output as an `Enumerable`
    * `Forcola.Daemon` - long-running server under a supervision tree
    * `Forcola.Duplex` - bidirectional stdin/stdout session

  Maintainers of CLI wrapper libraries who want to offer Forcola-backed
  execution without a mandatory dependency: see the [adoption
  guide](adopting_forcola.html).
  """

  alias Forcola.{Result, Shim}

  @typedoc """
  Errors a bounded run can return.

  The spawn reason is one of `:shim_not_found`, a reason string reported
  by the shim, or `{:shim_exited, %Forcola.Result{}}`; see `run/2`.
  """
  @type run_error :: {:timeout, Result.t()} | {:spawn, term()}

  # Margin added on top of the shim's own kill_grace_ms when computing
  # the Elixir-side backstop deadline: the shim already enforces
  # timeout_ms and confirms group death within kill_grace_ms before
  # sending EXIT, so this only fires if the shim itself never reports
  # back (e.g. it's wedged or the BEAM<->shim pipe is stuck).
  @backstop_margin_ms 5_000
  @default_kill_grace_ms 5_000

  @doc """
  Run `argv` (`[binary | args]`) to completion under the shim.

  ## Options

    * `:timeout_ms` - required. On expiry the child's process group is
      killed (SIGTERM, then SIGKILL after the kill grace) and
      `{:error, {:timeout, partial_result}}` is returned with output
      captured so far. The group is confirmed dead before the call
      returns, with one exception: if the shim itself never reports back,
      an Elixir-side backstop returns a result whose status is
      `{:signal, :unconfirmed}`, meaning death was not confirmed (see
      `Forcola.Result`). A child that exits exactly at the timeout
      boundary can be reported as a timeout whose result carries the
      normal exit status, including `status: 0`.
    * `:kill_grace_ms` - SIGTERM-to-SIGKILL grace in milliseconds,
      default `5_000`. Also accepted by `Forcola.Stream.lines/2`.
    * `:cd` - working directory.
    * `:env` - list of `{name, value}` strings.
    * `:merge_stderr` - route stderr into stdout; default `false`.
    * `:user` - run the child as this user, a string username or an
      integer uid. The user's primary gid and supplementary groups are
      taken from the passwd/group database unless `:group` overrides the
      gid. See ["Running as a different user"](#run/2-running-as-a-different-user).
    * `:group` - run the child with this group as its primary gid, a
      string group name or an integer gid. Given without `:user` it sets
      the gid (and clears supplementary groups to just that gid) without
      changing the uid; given with `:user` it overrides the user's
      primary gid.

  ## Running as a different user

  `:user`/`:group` make the shim drop privileges (`setgroups`, then
  `setgid`, then `setuid`, in that order) in the child before exec. This
  is POSIX-only, a one-way drop, and requires the shim process itself to
  run with enough privilege to drop: root, or `CAP_SETUID`/`CAP_SETGID`
  on Linux. Requesting the user the shim already runs as is a no-op and
  always succeeds.

  It fails closed. If the user or group cannot be resolved, or the shim
  lacks the privilege to drop, the child is never executed and the call
  returns the mode's normal spawn error (`{:error, {:spawn, reason}}` for
  `run/2`). The command never runs as the shim's own (possibly
  privileged) user when a different user was requested. Names are
  resolved in the parent before fork; only the numeric syscalls run in
  the child.

  Any exit status is `{:ok, %Forcola.Result{}}`; callers branch on
  `:status`. A non-zero exit is a result, not an error.

  ## Spawn errors

    * `{:error, {:spawn, :shim_not_found}}` - no shim binary exists for
      this target (neither downloaded nor built).
    * `{:error, {:spawn, reason}}` where `reason` is a string - the shim
      reported the spawn failure, for example a missing or
      non-executable command.
    * `{:error, {:spawn, {:shim_exited, %Forcola.Result{}}}}` - the shim
      exited without reporting an exit or error, for example because it
      was SIGKILLed. The result's status is `{:signal, :unconfirmed}`. A
      SIGKILLed shim gets no chance to kill the group, so the child may
      survive, reparented to pid 1.
  """
  @spec run([String.t(), ...], keyword()) :: {:ok, Result.t()} | {:error, run_error()}
  def run([_binary | _] = argv, opts) do
    timeout_ms = Keyword.fetch!(opts, :timeout_ms)
    kill_grace_ms = Keyword.get(opts, :kill_grace_ms, @default_kill_grace_ms)

    case Shim.open() do
      {:ok, port} ->
        try do
          spawn_and_collect(port, argv, opts, timeout_ms, kill_grace_ms)
        after
          close_port(port)
        end

      {:error, :not_found} ->
        {:error, {:spawn, :shim_not_found}}
    end
  end

  defp spawn_and_collect(port, argv, opts, timeout_ms, kill_grace_ms) do
    payload = Shim.encode_spawn(argv, Keyword.put(opts, :kill_grace_ms, kill_grace_ms))
    Shim.send_frame(port, Shim.tag_spawn(), payload)

    # Elixir-side backstop only: the shim itself enforces timeout_ms and
    # confirms group death within kill_grace_ms before reporting EXIT, so
    # under normal operation the shim's own EXIT frame arrives first.
    # This deadline exists in case the shim never reports back at all.
    deadline =
      System.monotonic_time(:millisecond) + timeout_ms + kill_grace_ms + @backstop_margin_ms

    collect(port, %{stdout: [], stderr: []}, deadline)
  end

  defp collect(port, acc, deadline) do
    remaining = deadline - System.monotonic_time(:millisecond)

    receive do
      {^port, {:data, <<tag, payload::binary>>}} ->
        handle_frame(port, tag, payload, acc, deadline)

      {^port, {:exit_status, _status}} ->
        # The shim exited without ever sending EXIT/ERROR (e.g. it
        # crashed). Death is implied by the OS reaping the port program
        # itself, but the child's own group state cannot be confirmed
        # from here.
        {:error, {:spawn, {:shim_exited, result(acc, {:signal, :unconfirmed})}}}
    after
      max(remaining, 0) ->
        {:error, {:timeout, result(acc, {:signal, :unconfirmed})}}
    end
  end

  defp handle_frame(port, tag, payload, acc, deadline) do
    cond do
      tag == Shim.tag_stdout() ->
        collect(port, %{acc | stdout: [acc.stdout | payload]}, deadline)

      tag == Shim.tag_stderr() ->
        collect(port, %{acc | stderr: [acc.stderr | payload]}, deadline)

      tag == Shim.tag_exit() ->
        {status, timed_out} = Shim.decode_exit(payload)
        res = result(acc, status)
        if timed_out, do: {:error, {:timeout, res}}, else: {:ok, res}

      tag == Shim.tag_error() ->
        reason = Shim.decode_error(payload)
        {:error, {:spawn, reason}}

      true ->
        collect(port, acc, deadline)
    end
  end

  defp result(acc, status) do
    %Result{
      status: status,
      stdout: IO.iodata_to_binary(acc.stdout),
      stderr: IO.iodata_to_binary(acc.stderr)
    }
  end

  defp close_port(port) do
    if port_open?(port) do
      Port.close(port)
    end
  catch
    :error, :badarg -> :ok
  end

  defp port_open?(port) do
    Port.info(port) != nil
  end
end