close_fds

A small Rust library that makes it easy to close all open file descriptors.

Usage

Add to your Cargo.toml:

[dependencies] close_fds = "0.2"

In your application:

``` use closefds::closeopen_fds;

fn main() { // ... unsafe { closeopenfds(3, &[]); } // ... } ```

IMPORTANT: Please read the documentation for close_open_fds() for an explanation of why it is unsafe.

The first argument to close_open_fds() is the lowest file descriptor that should be closed; all file descriptors less than this will be left open. The second argument is a slice containing a list of additional file descriptors that should be left open. (Note: close_open_fds() will be more efficient if this list is sorted, especially if it is more than a few elements long.)

close_open_fds() always succeeds. If one method of closing the file descriptors fails, it will fall back on another.

Some other helpful functions in this crate (more details in the documentation):

• set_fds_cloexec(minfd, keep_fds): Identical to close_open_fds(), but sets the FD_CLOEXEC flag on the file descriptors instead of closing them.
• iter_open_fds(minfd): Iterates over all open file descriptors for the current process, starting at minfd.
• iter_possible_fds(minfd) (not recommended): Identical to iter_open_fds(), but may yield invalid file descriptors; the caller is responsible for checking whether they are valid.

In addition, there are "thread-safe" versions of these three functions that behave more reliably in multithreaded programs (at the cost of increased performance on some platforms). See the documentation for more details.

Note that close_open_fds() should be preferred whenever possible, as it may be able to take advantage of platform-specific optimizations that these other functions cannot.

OS support

close_fds has two OS support tiers, similar to Rust's support tiers:

Tier 1: "Guaranteed to work" (tested in CI)

• Linux (glibc and musl)
• macOS
• FreeBSD

Tier 2: "Guaranteed to build" (built, but not tested, in CI)

• NetBSD
• OpenBSD
• DragonflyBSD
• Solaris
• Illumos

Note: As stated in the license, close_fds comes with no warranty.

OS-specific notes

Here is a list of the methods that iter_open_fds(), iter_possible_fds(), close_open_fds(), and set_fds_cloexec() will try on various platforms to improve performance when listing the open file descriptors:

• Linux
  • /proc/self/fd if /proc is mounted (very efficient)
• macOS
  • /dev/fd (very efficient)
• FreeBSD
  • /dev/fd if an fdescfs appears to be mounted there (very efficient)
  • The kern.proc.nfds sysctl to get the number of open file descriptors (moderately efficient unless large numbers of file descriptors are open; not used by close_open_fds() or by the "thread-safe" functions)
• OpenBSD
  • getdtablecount() to get the number of open file descriptors (moderately efficient unless large numbers of file descriptors are open; not used by close_open_fds() or by the "thread-safe" functions)
• NetBSD
  • /proc/self/fd if /proc is mounted (very efficient)
  • fcntl(0, F_MAXFD) to get the maximum open file descriptor (moderately efficient)
• Solaris and Illumos
  • /dev/fd or /proc/self/fd if either is available (very efficient)

In certain circumstances, close_open_fds() may also call closefrom() on the BSDs and/or close_range() on Linux 5.9+/FreeBSD 12.2+, both of which are very efficient.

If none of the methods listed above are available, it will fall back on a simple loop through every possible file descriptor number -- from minfd to sysconf(_SC_OPEN_MAX). This is slow, but it will always work.

Note: The most common use case, close_open_fds(3, &[]), is very efficient on Linux (with /proc mounted, or on kernel 5.9+), macOS, all of the BSDs, and Solaris/Illumos.