bin/pid-ns-init

#!/usr/bin/env python
# Copyright 2018-2021 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2

import errno
import fcntl
import functools
import os
import platform
import signal
import subprocess
import sys
import termios

from pathlib import Path

KILL_SIGNALS = (
    signal.SIGINT,
    signal.SIGTERM,
    signal.SIGHUP,
)

SIGTSTP_SIGCONT = (
    signal.SIGTSTP,
    signal.SIGCONT,
)


def forward_kill_signal(pid, signum, frame):
    if pid == 0:
        # Avoid a signal feedback loop, since signals sent to the
        # process group are also sent to the current process.
        signal.signal(signum, signal.SIG_DFL)
    os.kill(pid, signum)


def forward_sigtstp_sigcont(pid, signum, frame):
    handler = None
    if pid == 0:
        # Temporarily disable the handler in order to prevent it from
        # being called recursively, since the signal will also be sent
        # to the current process.
        handler = signal.signal(signum, signal.SIG_DFL)
    os.kill(pid, signum)
    if handler is not None:
        signal.signal(signum, handler)


def preexec_fn(uid, gid, groups, umask):
    if gid is not None:
        os.setgid(gid)
    if groups is not None:
        os.setgroups(groups)
    if uid is not None:
        os.setuid(uid)
    if umask is not None:
        os.umask(umask)

    # CPython >= 3 subprocess.Popen handles this internally.
    if platform.python_implementation() != 'CPython':
        for signum in (
                signal.SIGHUP,
                signal.SIGINT,
                signal.SIGPIPE,
                signal.SIGQUIT,
                signal.SIGTERM,
        ):
            signal.signal(signum, signal.SIG_DFL)


def main(argv):
    if len(argv) < 2:
        return 'Usage: {} <main-child-pid> or <uid> <gid> <groups> <umask> <pass_fds> <binary> <argv0> [arg]..'.format(argv[0])

    if len(argv) == 2:
        # The child process is init (pid 1) in a child pid namespace, and
        # the current process supervises from within the global pid namespace
        # (forwarding signals to init and forwarding exit status to the parent
        # process).
        main_child_pid = int(argv[1])
        setsid = False
        proc = None
    else:
        # The current process is init (pid 1) in a child pid namespace.
        uid, gid, groups, umask, pass_fds, binary, args = argv[1], argv[2], argv[3], argv[4], tuple(int(fd) for fd in argv[5].split(',')), argv[6], argv[7:]
        uid = int(uid) if uid else None
        gid = int(gid) if gid else None
        groups = tuple(int(group) for group in groups.split(',')) if groups else None
        umask = int(umask) if umask else None

        popen_kwargs = {
            'preexec_fn': functools.partial(preexec_fn, uid, gid, groups, umask),
            'pass_fds': pass_fds,
        }

        # Obtain the current nice value, which will be potentially be
        # used as the newly created session's autogroup nice value.
        nice_value = os.nice(0)

        # Isolate parent process from process group SIGSTOP (bug 675870)
        setsid = True
        os.setsid()

        # Set the previously obtained autogroup nice value again,
        # since we created a new session with os.setsid() above.
        try:
            Path("/proc/self/autogroup").write_text(str(nice_value))
        except EnvironmentError as e:
            # The process is likely not allowed to set the autogroup
            # value (Linux employs a rate limiting for unprivileged
            # changes to the autogroup value) or autogroups are not
            # enabled. Nothing we can do here, so we simply carry on.
            pass

        if sys.stdout.isatty():
            try:
                fcntl.ioctl(sys.stdout, termios.TIOCSCTTY, 0)
            except EnvironmentError as e:
                if e.errno == errno.EPERM:
                    # This means that stdout refers to the controlling terminal
                    # of the parent process, and in this case we do not want to
                    # steal it.
                    pass
                else:
                    raise
        proc = subprocess.Popen(args, executable=binary, **popen_kwargs)
        main_child_pid = proc.pid

    # If setsid has been called, use kill(0, signum) to
    # forward signals to the entire process group.
    sig_handler = functools.partial(forward_kill_signal, 0 if setsid else main_child_pid)
    for signum in KILL_SIGNALS:
        signal.signal(signum, sig_handler)

    # For correct operation of Ctrl+Z, forward SIGTSTP and SIGCONT.
    sigtstp_sigcont_handler = functools.partial(forward_sigtstp_sigcont, 0 if setsid else main_child_pid)
    for signum in SIGTSTP_SIGCONT:
        signal.signal(signum, sigtstp_sigcont_handler)

    # wait for child processes
    while True:
        try:
            pid, status = os.wait()
        except EnvironmentError as e:
            if e.errno == errno.EINTR:
                continue
            raise
        if pid == main_child_pid:
            if proc is not None:
                # Suppress warning messages like this:
                # ResourceWarning: subprocess 1234 is still running
                proc.returncode = 0

            if os.WIFEXITED(status):
                return os.WEXITSTATUS(status)
            elif os.WIFSIGNALED(status):
                signal.signal(os.WTERMSIG(status), signal.SIG_DFL)
                os.kill(os.getpid(), os.WTERMSIG(status))
            # go to the unreachable place
            break

    # this should never be reached
    return 127


if __name__ == '__main__':
    sys.exit(main(sys.argv))