skip to navigation
skip to content

Python Wiki

Python Insider Blog

Python 2 or 3?

Help Fund Python

[Python resources in languages other than English]

Non-English Resources

Add an event to this calendar.

Times are shown in UTC/GMT.

Add an event to this calendar.

PEP:433
Title:Easier suppression of file descriptor inheritance
Version:14d54024c533
Last-Modified:2013-08-28 21:58:48 +0200 (Wed, 28 Aug 2013)
Author:Victor Stinner <victor.stinner at gmail.com>
Status:Superseded
Type:Standards Track
Content-Type:text/x-rst
Created:10-January-2013
Python-Version:3.4
Superseded-By:446

Abstract

Add a new optional cloexec parameter on functions creating file descriptors, add different ways to change default values of this parameter, and add four new functions:

  • os.get_cloexec(fd)
  • os.set_cloexec(fd, cloexec=True)
  • sys.getdefaultcloexec()
  • sys.setdefaultcloexec(cloexec)

Rationale

A file descriptor has a close-on-exec flag which indicates if the file descriptor will be inherited or not.

On UNIX, if the close-on-exec flag is set, the file descriptor is not inherited: it will be closed at the execution of child processes; otherwise the file descriptor is inherited by child processes.

On Windows, if the close-on-exec flag is set, the file descriptor is not inherited; the file descriptor is inherited by child processes if the close-on-exec flag is cleared and if CreateProcess() is called with the bInheritHandles parameter set to TRUE (when subprocess.Popen is created with close_fds=False for example). Windows does not have "close-on-exec" flag but an inheritance flag which is just the opposite value. For example, setting close-on-exec flag means clearing the HANDLE_FLAG_INHERIT flag of an handle.

Status in Python 3.3

On UNIX, the subprocess module closes file descriptors greater than 2 by default since Python 3.2 [1]. All file descriptors created by the parent process are automatically closed in the child process.

xmlrpc.server.SimpleXMLRPCServer sets the close-on-exec flag of the listening socket, the parent class socketserver.TCPServer does not set this flag.

There are other cases creating a subprocess or executing a new program where file descriptors are not closed: functions of the os.spawn*() and the os.exec*() families and third party modules calling exec() or fork() + exec(). In this case, file descriptors are shared between the parent and the child processes which is usually unexpected and causes various issues.

This PEP proposes to continue the work started with the change in the subprocess in Python 3.2, to fix the issue in any code, and not just code using subprocess.

Inherited file descriptors issues

Closing the file descriptor in the parent process does not close the related resource (file, socket, ...) because it is still open in the child process.

The listening socket of TCPServer is not closed on exec(): the child process is able to get connection from new clients; if the parent closes the listening socket and create a new listening socket on the same address, it would get an "address already is used" error.

Not closing file descriptors can lead to resource exhaustion: even if the parent closes all files, creating a new file descriptor may fail with "too many files" because files are still open in the child process.

See also the following issues:

Security

Leaking file descriptors is a major security vulnerability. An untrusted child process can read sensitive data like passwords and take control of the parent process though leaked file descriptors. It is for example a known vulnerability to escape from a chroot.

See also the CERT recommandation: FIO42-C. Ensure files are properly closed when they are no longer needed.

Example of vulnerabilities:

Atomicity

Using fcntl() to set the close-on-exec flag is not safe in a multithreaded application. If a thread calls fork() and exec() between the creation of the file descriptor and the call to fcntl(fd, F_SETFD, new_flags): the file descriptor will be inherited by the child process. Modern operating systems offer functions to set the flag during the creation of the file descriptor, which avoids the race condition.

Portability

Python 3.2 added socket.SOCK_CLOEXEC flag, Python 3.3 added os.O_CLOEXEC flag and os.pipe2() function. It is already possible to set atomically close-on-exec flag in Python 3.3 when opening a file and creating a pipe or socket.

The problem is that these flags and functions are not portable: only recent versions of operating systems support them. O_CLOEXEC and SOCK_CLOEXEC flags are ignored by old Linux versions and so FD_CLOEXEC flag must be checked using fcntl(fd, F_GETFD). If the kernel ignores O_CLOEXEC or SOCK_CLOEXEC flag, a call to fcntl(fd, F_SETFD, flags) is required to set close-on-exec flag.

Note

OpenBSD older 5.2 does not close the file descriptor with close-on-exec flag set if fork() is used before exec(), but it works correctly if exec() is called without fork(). Try openbsd_bug.py.

Scope

Applications still have to close explicitly file descriptors after a fork(). The close-on-exec flag only closes file descriptors after exec(), and so after fork() + exec().

This PEP only change the close-on-exec flag of file descriptors created by the Python standard library, or by modules using the standard library. Third party modules not using the standard library should be modified to conform to this PEP. The new os.set_cloexec() function can be used for example.

Note

See Close file descriptors after fork for a possible solution for fork() without exec().

Proposal

Add a new optional cloexec parameter on functions creating file descriptors and different ways to change default value of this parameter.

Add new functions:

  • os.get_cloexec(fd:int) -> bool: get the close-on-exec flag of a file descriptor. Not available on all platforms.
  • os.set_cloexec(fd:int, cloexec:bool=True): set or clear the close-on-exec flag on a file descriptor. Not available on all platforms.
  • sys.getdefaultcloexec() -> bool: get the current default value of the cloexec parameter
  • sys.setdefaultcloexec(cloexec: bool): set the default value of the cloexec parameter

Add a new optional cloexec parameter to:

  • asyncore.dispatcher.create_socket()
  • io.FileIO
  • io.open()
  • open()
  • os.dup()
  • os.dup2()
  • os.fdopen()
  • os.open()
  • os.openpty()
  • os.pipe()
  • select.devpoll()
  • select.epoll()
  • select.kqueue()
  • socket.socket()
  • socket.socket.accept()
  • socket.socket.dup()
  • socket.socket.fromfd
  • socket.socketpair()

The default value of the cloexec parameter is sys.getdefaultcloexec().

Add a new command line option -e and an environment variable PYTHONCLOEXEC to the set close-on-exec flag by default.

subprocess clears the close-on-exec flag of file descriptors of the pass_fds parameter.

All functions creating file descriptors in the standard library must respect the default value of the cloexec parameter: sys.getdefaultcloexec().

File descriptors 0 (stdin), 1 (stdout) and 2 (stderr) are expected to be inherited, but Python does not handle them differently. When os.dup2() is used to replace standard streams, cloexec=False must be specified explicitly.

Drawbacks of the proposal:

  • It is not more possible to know if the close-on-exec flag will be set or not on a newly created file descriptor just by reading the source code.
  • If the inheritance of a file descriptor matters, the cloexec parameter must now be specified explicitly, or the library or the application will not work depending on the default value of the cloexec parameter.

Alternatives

Inheritance enabled by default, default no configurable

Add a new optional parameter cloexec on functions creating file descriptors. The default value of the cloexec parameter is False, and this default cannot be changed. File descriptor inheritance enabled by default is also the default on POSIX and on Windows. This alternative is the most convervative option.

This option does not solve issues listed in the Rationale section, it only provides an helper to fix them. All functions creating file descriptors have to be modified to set cloexec=True in each module used by an application to fix all these issues.

Inheritance enabled by default, default can only be set to True

This alternative is based on the proposal: the only difference is that sys.setdefaultcloexec() does not take any argument, it can only be used to set the default value of the cloexec parameter to True.

Disable inheritance by default

This alternative is based on the proposal: the only difference is that the default value of the cloexec parameter is True (instead of False).

If a file must be inherited by child processes, cloexec=False parameter can be used.

Advantages of setting close-on-exec flag by default:

Drawbacks of setting close-on-exec flag by default:

  • It violates the principle of least surprise. Developers using the os module may expect that Python respects the POSIX standard and so that close-on-exec flag is not set by default.
  • The os module is written as a thin wrapper to system calls (to functions of the C standard library). If atomic flags to set close-on-exec flag are not supported (see Appendix: Operating system support), a single Python function call may call 2 or 3 system calls (see Performances section).
  • Extra system calls, if any, may slow down Python: see Performances.

Backward compatibility: only a few programs rely on inheritance of file descriptors, and they only pass a few file descriptors, usually just one. These programs will fail immediatly with EBADF error, and it will be simple to fix them: add cloexec=False parameter or use os.set_cloexec(fd, False).

The subprocess module will be changed anyway to clear close-on-exec flag on file descriptors listed in the pass_fds parameter of Popen constructor. So it possible that these programs will not need any fix if they use the subprocess module.

Close file descriptors after fork

This PEP does not fix issues with applications using fork() without exec(). Python needs a generic process to register callbacks which would be called after a fork, see #16500: Add an atfork module [2]. Such registry could be used to close file descriptors just after a fork().

Drawbacks:

  • It does not solve the problem on Windows: fork() does not exist on Windows
  • This alternative does not solve the problem for programs using exec() without fork().
  • A third party module may call directly the C function fork() which will not call "atfork" callbacks.
  • All functions creating file descriptors must be changed to register a callback and then unregister their callback when the file is closed. Or a list of all open file descriptors must be maintained.
  • The operating system is a better place than Python to close automatically file descriptors. For example, it is not easy to avoid a race condition between closing the file and unregistering the callback closing the file.

open(): add "e" flag to mode

A new "e" mode would set close-on-exec flag (best-effort).

This alternative only solves the problem for open(). socket.socket() and os.pipe() do not have a mode parameter for example.

Since its version 2.7, the GNU libc supports "e" flag for fopen(). It uses O_CLOEXEC if available, or use fcntl(fd, F_SETFD, FD_CLOEXEC). With Visual Studio, fopen() accepts a "N" flag which uses O_NOINHERIT.

Bikeshedding on the name of the new parameter

  • inherit, inherited: closer to Windows definition
  • sensitive
  • sterile: "Does not produce offspring."

Applications using inheritance of file descriptors

Most developers don't know that file descriptors are inherited by default. Most programs do not rely on inheritance of file descriptors. For example, subprocess.Popen was changed in Python 3.2 to close all file descriptors greater than 2 in the child process by default. No user complained about this behavior change yet.

Network servers using fork may want to pass the client socket to the child process. For example, on UNIX a CGI server pass the socket client through file descriptors 0 (stdin) and 1 (stdout) using dup2().

To access a restricted resource like creating a socket listening on a TCP port lower than 1024 or reading a file containing sensitive data like passwords, a common practice is: start as the root user, create a file descriptor, create a child process, drop privileges (ex: change the current user), pass the file descriptor to the child process and exit the parent process.

Security is very important in such use case: leaking another file descriptor would be a critical security vulnerability (see Security). The root process may not exit but monitors the child process instead, and restarts a new child process and pass the same file descriptor if the previous child process crashed.

Example of programs taking file descriptors from the parent process using a command line option:

  • gpg: --status-fd <fd>, --logger-fd <fd>, etc.
  • openssl: -pass fd:<fd>
  • qemu: -add-fd <fd>
  • valgrind: --log-fd=<fd>, --input-fd=<fd>, etc.
  • xterm: -S <fd>

On Linux, it is possible to use "/dev/fd/<fd>" filename to pass a file descriptor to a program expecting a filename.

Performances

Setting close-on-exec flag may require additional system calls for each creation of new file descriptors. The number of additional system calls depends on the method used to set the flag:

  • O_NOINHERIT: no additional system call
  • O_CLOEXEC: one additional system call, but only at the creation of the first file descriptor, to check if the flag is supported. If the flag is not supported, Python has to fallback to the next method.
  • ioctl(fd, FIOCLEX): one additional system call per file descriptor
  • fcntl(fd, F_SETFD, flags): two additional system calls per file descriptor, one to get old flags and one to set new flags

On Linux, setting the close-on-flag has a low overhead on performances. Results of bench_cloexec.py on Linux 3.6:

  • close-on-flag not set: 7.8 us
  • O_CLOEXEC: 1% slower (7.9 us)
  • ioctl(): 3% slower (8.0 us)
  • fcntl(): 3% slower (8.0 us)

Implementation

os.get_cloexec(fd)

Get the close-on-exec flag of a file descriptor.

Pseudo-code:

if os.name == 'nt':
    def get_cloexec(fd):
        handle = _winapi._get_osfhandle(fd);
        flags = _winapi.GetHandleInformation(handle)
        return not(flags & _winapi.HANDLE_FLAG_INHERIT)
else:
    try:
        import fcntl
    except ImportError:
        pass
    else:
        def get_cloexec(fd):
            flags = fcntl.fcntl(fd, fcntl.F_GETFD)
            return bool(flags & fcntl.FD_CLOEXEC)

os.set_cloexec(fd, cloexec=True)

Set or clear the close-on-exec flag on a file descriptor. The flag is set after the creation of the file descriptor and so it is not atomic.

Pseudo-code:

if os.name == 'nt':
    def set_cloexec(fd, cloexec=True):
        handle = _winapi._get_osfhandle(fd);
        mask = _winapi.HANDLE_FLAG_INHERIT
        if cloexec:
            flags = 0
        else:
            flags = mask
        _winapi.SetHandleInformation(handle, mask, flags)
else:
    fnctl = None
    ioctl = None
    try:
        import ioctl
    except ImportError:
        try:
            import fcntl
        except ImportError:
            pass
    if ioctl is not None and hasattr('FIOCLEX', ioctl):
        def set_cloexec(fd, cloexec=True):
            if cloexec:
                ioctl.ioctl(fd, ioctl.FIOCLEX)
            else:
                ioctl.ioctl(fd, ioctl.FIONCLEX)
    elif fnctl is not None:
        def set_cloexec(fd, cloexec=True):
            flags = fcntl.fcntl(fd, fcntl.F_GETFD)
            if cloexec:
                flags |= FD_CLOEXEC
            else:
                flags &= ~FD_CLOEXEC
            fcntl.fcntl(fd, fcntl.F_SETFD, flags)

ioctl is preferred over fcntl because it requires only one syscall, instead of two syscalls for fcntl.

Note

fcntl(fd, F_SETFD, flags) only supports one flag (FD_CLOEXEC), so it would be possible to avoid fcntl(fd, F_GETFD). But it may drop other flags in the future, and so it is safer to keep the two functions calls.

Note

fopen() function of the GNU libc ignores the error if fcntl(fd, F_SETFD, flags) failed.

open()

  • Windows: open() with O_NOINHERIT flag [atomic]
  • open() with O_CLOEXEC flag [atomic]
  • open() + os.set_cloexec(fd, True) [best-effort]

os.dup()

  • Windows: DuplicateHandle() [atomic]
  • fcntl(fd, F_DUPFD_CLOEXEC) [atomic]
  • dup() + os.set_cloexec(fd, True) [best-effort]

os.dup2()

  • fcntl(fd, F_DUP2FD_CLOEXEC, fd2) [atomic]
  • dup3() with O_CLOEXEC flag [atomic]
  • dup2() + os.set_cloexec(fd, True) [best-effort]

os.pipe()

  • Windows: CreatePipe() with SECURITY_ATTRIBUTES.bInheritHandle=TRUE, or _pipe() with O_NOINHERIT flag [atomic]
  • pipe2() with O_CLOEXEC flag [atomic]
  • pipe() + os.set_cloexec(fd, True) [best-effort]

socket.socket()

  • Windows: WSASocket() with WSA_FLAG_NO_HANDLE_INHERIT flag [atomic]
  • socket() with SOCK_CLOEXEC flag [atomic]
  • socket() + os.set_cloexec(fd, True) [best-effort]

socket.socketpair()

  • socketpair() with SOCK_CLOEXEC flag [atomic]
  • socketpair() + os.set_cloexec(fd, True) [best-effort]

socket.socket.accept()

  • accept4() with SOCK_CLOEXEC flag [atomic]
  • accept() + os.set_cloexec(fd, True) [best-effort]

Backward compatibility

There is no backward incompatible change. The default behaviour is unchanged: the close-on-exec flag is not set by default.

Appendix: Operating system support

Windows

Windows has an O_NOINHERIT flag: "Do not inherit in child processes".

For example, it is supported by open() and _pipe().

The flag can be cleared using SetHandleInformation(fd, HANDLE_FLAG_INHERIT, 0).

CreateProcess() has an bInheritHandles parameter: if it is FALSE, the handles are not inherited. If it is TRUE, handles with HANDLE_FLAG_INHERIT flag set are inherited. subprocess.Popen uses close_fds option to define bInheritHandles.

ioctl

Functions:

  • ioctl(fd, FIOCLEX, 0): set the close-on-exec flag
  • ioctl(fd, FIONCLEX, 0): clear the close-on-exec flag

Availability: Linux, Mac OS X, QNX, NetBSD, OpenBSD, FreeBSD.

fcntl

Functions:

  • flags = fcntl(fd, F_GETFD); fcntl(fd, F_SETFD, flags | FD_CLOEXEC): set the close-on-exec flag
  • flags = fcntl(fd, F_GETFD); fcntl(fd, F_SETFD, flags & ~FD_CLOEXEC): clear the close-on-exec flag

Availability: AIX, Digital UNIX, FreeBSD, HP-UX, IRIX, Linux, Mac OS X, OpenBSD, Solaris, SunOS, Unicos.

Atomic flags

New flags:

  • O_CLOEXEC: available on Linux (2.6.23), FreeBSD (8.3), OpenBSD 5.0, Solaris 11, QNX, BeOS, next NetBSD release (6.1?). This flag is part of POSIX.1-2008.
  • SOCK_CLOEXEC flag for socket() and socketpair(), available on Linux 2.6.27, OpenBSD 5.2, NetBSD 6.0.
  • WSA_FLAG_NO_HANDLE_INHERIT flag for WSASocket(): supported on Windows 7 with SP1, Windows Server 2008 R2 with SP1, and later
  • fcntl(): F_DUPFD_CLOEXEC flag, available on Linux 2.6.24, OpenBSD 5.0, FreeBSD 9.1, NetBSD 6.0, Solaris 11. This flag is part of POSIX.1-2008.
  • fcntl(): F_DUP2FD_CLOEXEC flag, available on FreeBSD 9.1 and Solaris 11.
  • recvmsg(): MSG_CMSG_CLOEXEC, available on Linux 2.6.23, NetBSD 6.0.

On Linux older than 2.6.23, O_CLOEXEC flag is simply ignored. So we have to check that the flag is supported by calling fcntl(). If it does not work, we have to set the flag using ioctl() or fcntl().

On Linux older than 2.6.27, if the SOCK_CLOEXEC flag is set in the socket type, socket() or socketpair() fail and errno is set to EINVAL.

On Windows XPS3, WSASocket() with with WSAEPROTOTYPE when WSA_FLAG_NO_HANDLE_INHERIT flag is used.

New functions:

  • dup3(): available on Linux 2.6.27 (and glibc 2.9)
  • pipe2(): available on Linux 2.6.27 (and glibc 2.9)
  • accept4(): available on Linux 2.6.28 (and glibc 2.10)

If accept4() is called on Linux older than 2.6.28, accept4() returns -1 (fail) and errno is set to ENOSYS.

Footnotes

[1]On UNIX since Python 3.2, subprocess.Popen() closes all file descriptors by default: close_fds=True. It closes file descriptors in range 3 inclusive to local_max_fd exclusive, where local_max_fd is fcntl(0, F_MAXFD) on NetBSD, or sysconf(_SC_OPEN_MAX) otherwise. If the error pipe has a descriptor smaller than 3, ValueError is raised.