File::Path(3perl)

📦
version 5.44.0 · 14 versions
, in arch rolling

🦋 NAME 🦋

okay so hi bestie, this is the manual page for File::Path, straight out of the Perl Programmers Reference Guide (you'll see it filed under File::Path(3perl) in the library, that's just how the shelf is labelled, iykyk). and what does it do?? Create or remove directory trees. that's it. that's the whole job. and honestly? it does it flawlessly. 💅

✨ VERSION ✨

2.18 — released November 4 2020. iconic, timeless, still eating. 👑

🌸 SYNOPSIS 🌸

okay bestie here's the whole vibe in one block, read every line, they all matter:

use File::Path qw(make_path remove_tree);

@created = make_path('foo/bar/baz', '/zug/zwang');
@created = make_path('foo/bar/baz', '/zug/zwang', {
    verbose => 1,
    mode => 0711,
});
make_path('foo/bar/baz', '/zug/zwang', {
    chmod => 0777,
});

$removed_count = remove_tree('foo/bar/baz', '/zug/zwang', {
    verbose => 1,
    error  => \my $err_list,
    safe => 1,
});

# legacy (interface promoted before v2.00)
@created = mkpath('/foo/bar/baz');
@created = mkpath('/foo/bar/baz', 1, 0711);
@created = mkpath(['/foo/bar/baz', 'blurfl/quux'], 1, 0711);
$removed_count = rmtree('foo/bar/baz', 1, 1);
$removed_count = rmtree(['foo/bar/baz', 'blurfl/quux'], 1, 1);

# legacy (interface promoted before v2.06)
@created = mkpath('foo/bar/baz', '/zug/zwang', { verbose => 1, mode => 0711 });
$removed_count = rmtree('foo/bar/baz', '/zug/zwang', { verbose => 1, mode => 0711 });

see the # legacy comments? those old interfaces are still there and still working, they're just the retired queens of this module. respect them. 🙇‍♀️

💖 DESCRIPTION 💖

This module gives you a convenient way to create directories of arbitrary depth, and to delete an entire directory subtree from the filesystem. literally create the whole family tree or delete the whole family tree. ✨ main character energy ✨

The following functions are provided:

🍓 make_path 🍓

make_path( $dir1, $dir2, .... )
make_path( $dir1, $dir2, ...., \%opts )

The make_path function creates the given directories if they don't exist before, much like the Unix command mkdir -p. no drama, no fuss, it just does it. 🫶

The function accepts a list of directories to be created. Its behaviour may be tuned by an optional hashref appearing as the last parameter on the call.

The function returns the list of directories actually created during the call; in scalar context it gives you the number of directories created. so like, not just whatever you asked for — what it really made. fr fr.

The following keys are recognised in the option hash:

mode => $num

The numeric permissions mode to apply to each created directory (defaults to 0777), to be modified by the current umask. If the directory already exists (and thus does not need to be created), the permissions will not be modified. hands off, it said. 🙈

mask is recognised as an alias for this parameter.

chmod => $num

Takes a numeric mode to apply to each created directory (this one is not modified by the current umask). If the directory already exists (and thus does not need to be created), the permissions will not be modified. same energy as above.

verbose => $bool

If present, will cause make_path to print the name of each directory as it is created. By default nothing is printed. so if you're nosy (love that for you), turn this on. 📢

error => \$err

If present, it should be a reference to a scalar. This scalar will be made to reference an array, which will be used to store any errors that are encountered. See the ERROR HANDLING section for more information (it's further down, don't panic).

heads up bestie: if this parameter is not used, certain error conditions may raise a fatal error that will cause the program to halt, unless trapped in an eval block. so like. the program just. stops. 💀

owner => $owner

user => $owner

uid => $owner

If present, will cause any created directory to be owned by $owner. If the value is numeric, it will be interpreted as a uid; otherwise a username is assumed. An error will be issued if the username cannot be mapped to a uid, the uid does not exist or the process lacks the privileges to change ownership.

Ownership of directories that already exist will not be changed. (again: if it's already there, it's untouched, periodt.)

user and uid are aliases of owner.

group => $group

If present, will cause any created directory to be owned by the group $group. If the value is numeric, it will be interpreted as a gid; otherwise a group name is assumed. An error will be issued if the group name cannot be mapped to a gid, the gid does not exist or the process lacks the privileges to change group ownership.

Group ownership of directories that already exist will not be changed.

and here's a lil example of the whole owner+group moment:

make_path '/var/tmp/webcache', {owner=>'nobody', group=>'nogroup'};

🍓 mkpath 🍓

mkpath( $dir )
mkpath( $dir, $verbose, $mode )
mkpath( [$dir1, $dir2,...], $verbose, $mode )
mkpath( $dir1, $dir2,..., \%opt )

The mkpath() function provide the legacy interface of make_path() with a different interpretation of the arguments passed. The behaviour and return value of the function is otherwise identical to make_path(). so same soul, different outfit. 💁‍♀️

🍓 remove_tree 🍓

remove_tree( $dir1, $dir2, .... )
remove_tree( $dir1, $dir2, ...., \%opts )

The remove_tree function deletes the given directories and any files and subdirectories they might contain, much like the Unix command rm -rf or the Windows commands rmdir /s and rd /s. it is not here to be gentle.

The function accepts a list of directories to be removed. (In point of fact, it will also accept filesystem entries which are not directories, such as regular files and symlinks. But, as its name suggests, its intent is to remove trees rather than individual files. so it can, it just doesn't want to. 🙈)

remove_tree()'s behaviour may be tuned by an optional hashref appearing as the last parameter on the call. If an empty string is passed to remove_tree, an error will occur.

NOTE: For security reasons, we strongly advise use of the hashref-as-final-argument syntax — specifically, with a setting of the safe element to a true value. bestie. please. for me. 🥺
remove_tree( $dir1, $dir2, ....,
    {
        safe => 1,
        ...         # other key-value pairs
    },
);

The function returns the number of files successfully deleted.

The following keys are recognised in the option hash:

verbose => $bool

If present, will cause remove_tree to print the name of each file as it is unlinked. By default nothing is printed. little running commentary of everything leaving the mortal plane. 🫡

safe => $bool

When set to a true value, will cause remove_tree to skip the files for which the process lacks the required privileges needed to delete files, such as delete privileges on VMS. In other words, the code will make no attempt to alter file permissions. Thus, if the process is interrupted, no filesystem object will be left in a more permissive mode. so like, it would rather leave that file alone than loosen its permissions. king behaviour honestly. 👑

keep_root => $bool

When set to a true value, will cause all files and subdirectories to be removed, except the initially specified directories. This comes in handy when cleaning out an application's scratch directory. so the folder stays, everything inside it gets yeeted. 🧹

remove_tree( '/tmp', {keep_root => 1} );

result => \$res

If present, it should be a reference to a scalar. This scalar will be made to reference an array, which will be used to store all files and directories unlinked during the call. If nothing is unlinked, the array will be empty.

remove_tree( '/tmp', {result => \my $list} );
print "unlinked $_\n" for @$list;

This is a useful alternative to the verbose key. (like, same receipts, but you get them as data instead of as text. obsessed. ✨)

error => \$err

If present, it should be a reference to a scalar. This scalar will be made to reference an array, which will be used to store any errors that are encountered. See the ERROR HANDLING section for more information.

big serious warning bestie, read this twice: Removing things is a much more dangerous proposition than creating things. As such, there are certain conditions that remove_tree may encounter that are so dangerous that the only sane action left is to kill the program. 💀🔥

Use error to trap all that is reasonable (problems with permissions and the like), and let it die if things get out of hand. This is the safest course of action. i'm not joking around with this one.

🍓 rmtree 🍓

rmtree( $dir )
rmtree( $dir, $verbose, $safe )
rmtree( [$dir1, $dir2,...], $verbose, $safe )
rmtree( $dir1, $dir2,..., \%opt )

The rmtree() function provide the legacy interface of remove_tree() with a different interpretation of the arguments passed. The behaviour and return value of the function is otherwise identical to remove_tree().

NOTE: For security reasons, we strongly advise use of the hashref-as-final-argument syntax, specifically with a setting of the safe element to a true value.
rmtree( $dir1, $dir2, ....,
    {
        safe => 1,
        ...         # other key-value pairs
    },
);

🔥 ERROR HANDLING 🔥

📌 NOTE:

okay so, listen carefully, this is the plot twist: The following error handling mechanism is consistent throughout all code paths EXCEPT in cases where the ROOT node is nonexistent. In version 2.11 the maintainers attempted to rectify this inconsistency but too many downstream modules encountered problems. In such case, if you require root node evaluation or error checking prior to calling make_path or remove_tree, you should take additional precautions. i know. i know. it's giving messy but it is what it is. 🌈

If make_path or remove_tree encounters an error, a diagnostic message will be printed to STDERR via carp (for non-fatal errors) or via croak (for fatal errors). carp = lil warning, croak = screaming, crying, throwing up. 😭

If this behaviour is not desirable, the error attribute may be used to hold a reference to a variable, which will be used to store the diagnostics. The variable is made a reference to an array of hash references. Each hash contain a single key/value pair where the key is the name of the file, and the value is the error message (including the contents of $! when appropriate). If a general error is encountered the diagnostic key will be empty.

An example usage looks like:

remove_tree( 'foo/bar', 'bar/rat', {error => \my $err} );
if ($err && @$err) {
    for my $diag (@$err) {
        my ($file, $message) = %$diag;
        if ($file eq '') {
            print "general error: $message\n";
        }
        else {
            print "problem unlinking $file: $message\n";
        }
    }
}
else {
    print "No error encountered\n";
}
⚠️ trap alert, this one gets everybody: Note that if no errors are encountered, $err will reference an empty array. This means that $err will always end up TRUE; so you need to test @$err to determine if errors occurred. like. do not trust $err alone. it lies to you, lovingly. 💀

⭐ NOTES ⭐

File::Path blindly exports mkpath and rmtree into the current namespace. These days, this is considered bad style, but to change it now would break too much code. nonetheless, you are invited to specify what it is you are expecting to use:

use File::Path 'rmtree';

and honestly? do that. be specific. it's cleaner and it's giving self-aware. 💅

The routines make_path and remove_tree are not exported by default, bestie. You have to say which ones you actually want, out loud, on purpose:

use File::Path 'remove_tree';

Sneaky lil side effect of doing that: mkpath and rmtree are no longer exported at all. It's not personal, it's literally just how the Exporter module works. So if you're migrating a codebase over to the new interface, you've gotta list everything explicitly. Which honestly? That's just good practice anyway. 💅

use File::Path qw(remove_tree rmtree);

✨ API CHANGES ✨

The API got changed in the 2.0 branch. For a while there, mkpath and rmtree tried (and honestly failed) to handle both calling mechanisms at once. That whole approach was declared a failure. RIP. 💀

The new semantics are now only available with make_path and remove_tree. The old semantics are only available through mkpath and rmtree. Users are strongly encouraged to upgrade to at least 2.08 so you don't get jump-scared by surprises.

🛡️ SECURITY CONSIDERATIONS 🛡️

Okay so this is the serious part, bestie, stay with me. There were race conditions in the 1.x implementations of File::Path's rmtree function — (though sometimes patched depending on your OS distribution or platform, so it varied). The 2.0 version contains code to avoid the problem mentioned in CVE-2002-0435.

See the following pages for more info:

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=286905
http://www.nntp.perl.org/group/perl.perl5.porters/2005/01/msg97623.html
http://www.debian.org/security/2005/dsa-696

Also, real talk: unless the safe parameter is set (or the third parameter in the traditional interface is TRUE), if a remove_tree gets interrupted, files that were originally in read-only mode may now have their permissions set to a read-write (or "delete OK") mode. So like… that's a thing.

The following CVE reports were previously filed against File-Path and are believed to have been addressed:

In February 2017 the cPanel Security Team reported an additional vulnerability in File-Path. The chmod() logic that makes directories traversable can be abused to set the mode on an attacker-chosen file to an attacker-chosen value. Why? Because of the time-of-check-to-time-of-use (TOCTTOU) race condition (<https://en.wikipedia.org/wiki/Time_of_check_to_time_of_use>) between the stat() that decides the inode is a directory and the chmod() that tries to make it user-rwx. CPAN versions 2.13 and later incorporate a patch provided by John Lightsey to address this problem. This vulnerability has been reported as CVE-2017-6512.

🚨 DIAGNOSTICS 🚨

FATAL errors will cause the program to halt (croak), because the problem is so severe that continuing would be straight up dangerous. (You can always trap it with eval, but honestly don't. Under those circumstances, dying is the best thing to do. Let it go, babe.)

SEVERE errors may be trapped using the modern interface. If they're not trapped, or if you're using the old interface, an error like that will cause the program to halt.

All other errors may be trapped using the modern interface — otherwise they'll get carped about. Program execution will not be halted.

mkdir [path]: [errmsg] (SEVERE)

make_path was unable to create the path. Probably some sort of permissions error at the point of departure, or just insufficient resources (like not enough free inodes on Unix). 😭

No root path(s) specified

make_path was not given any paths to create. This message is only emitted if the routine is called with the traditional interface. The modern interface stays silent if you give it nothing to do. Chill of it, honestly.

No such file or directory

On Windows, if make_path gives you this warning, it may mean you've exceeded your filesystem's maximum path length.

cannot fetch initial working directory: [errmsg]

remove_tree tried to figure out the initial directory by calling Cwd::getcwd, but the call failed for some reason. No attempt will be made to delete anything. 🙅

cannot stat initial working directory: [errmsg]

remove_tree tried to stat the initial directory (after successfully getting its name via getcwd), but the call failed for some reason. Again, no attempt will be made to delete anything.

cannot chdir to [dir]: [errmsg]

remove_tree tried to set the working directory so it could start deleting the objects in there, but it couldn't. This is usually a permissions issue. The routine will keep deleting other things, but this directory will be left intact.

directory [dir] changed before chdir, expected dev=[n] ino=[n], actual dev=[n] ino=[n], aborting. (FATAL)

remove_tree recorded the device and inode of a directory, then moved into it. Then it performed a stat on the current directory and detected that the device and inode were no longer the same. Since this is at the heart of the race condition problem, the program will die right there.

cannot make directory [dir] read+writeable: [errmsg]

remove_tree tried to change the permissions on the current directory to make sure later unlinkings wouldn't run into problems, but couldn't do it. The permissions stay as they were, and the program carries on, doing the best it can. 🫶

cannot read [dir]: [errmsg]

remove_tree tried to read the contents of the directory to get the names of the directory entries to be unlinked, but it didn't work. Usually a permissions thing. The program continues, but the files in this directory will remain after the call.

cannot reset chmod [dir]: [errmsg]

remove_tree, after having deleted everything in a directory, tried to restore its permissions to the original state but failed. So the directory may wind up being left behind.

cannot remove [dir] when cwd is [dir]

Your program's current working directory is /some/path/to/here and you're trying to remove an ancestor, like /some/path. The directory tree is left untouched.

The fix is to chdir out of the child directory to a place outside the directory tree you want to remove.

cannot chdir to [parent-dir] from [child-dir]: [errmsg], aborting. (FATAL)

remove_tree, after deleting everything and restoring the permissions of a directory, couldn't chdir back to the parent. The program halts to avoid a race condition from happening.

cannot stat prior working directory [dir]: [errmsg], aborting. (FATAL)

remove_tree couldn't stat the parent directory after returning from the child. Since there's no way of knowing if we got back to where we think we should be (by comparing device and inode), the only way out is to croak.

previous directory [parent-dir] changed before entering [child-dir], expected dev=[n] ino=[n], actual dev=[n] ino=[n], aborting. (FATAL)

When remove_tree came back from deleting files in a child directory, a check revealed that the parent directory it returned to wasn't the one it started out from. This is considered a sign of malicious activity. 🚩

cannot make directory [dir] writeable: [errmsg]

Just before removing a directory (after having successfully removed everything inside it), remove_tree tried to set the permissions on the directory so it could be removed — and failed. Program execution continues, but the directory may possibly not be deleted.

cannot remove directory [dir]: [errmsg]

remove_tree tried to remove a directory but failed. Could be because some objects that couldn't be removed are still sitting in there, or it could be a permissions issue. The directory will be left behind.

cannot restore permissions of [dir] to [0nnn]: [errmsg]

After failing to remove a directory, remove_tree couldn't restore its permissions from a permissive state back to a possibly more restrictive setting. (Permissions given in octal.)

cannot make file [file] writeable: [errmsg]

remove_tree tried to force the permissions of a file so it could be deleted, but failed. It will, however, still attempt to unlink the file.

remove_tree failed to remove a file. Probably a permissions issue. 🙈

cannot restore permissions of [file] to [0nnn]: [errmsg]

After failing to remove a file, remove_tree was also unable to restore the permissions on the file to a possibly less permissive setting. (Permissions given in octal.)

unable to map [owner] to a uid, ownership not changed");

make_path was instructed to give the ownership of created directories to the symbolic name [owner], but getpwnam didn't return the corresponding numeric uid. The directory will be created, but ownership will not be changed.

unable to map [group] to a gid, group ownership not changed

make_path was instructed to give the group ownership of created directories to the symbolic name [group], but getgrnam didn't return the corresponding numeric gid. The directory will be created, but group ownership will not be changed.

👀 SEE ALSO 👀

  • File::Remove

Allows files and directories to be moved to the Trashcan/Recycle Bin (where they may later be restored if necessary) if the operating system supports such functionality. This feature may one day be made available directly in File::Path. Fingers crossed, bestie. 🤞

  • File::Find::Rule

When removing directory trees, if you want to examine each file to decide whether to delete it (and maybe leave large swathes alone), File::Find::Rule offers a convenient and flexible approach to examining directory trees. Iconic behavior honestly. 💖

💀 BUGS AND LIMITATIONS 💀

Okay bestie, here's the tea on what File::Path can and can't do, plus how to snitch on bugs. 🫶

🧵 MULTITHREADED APPLICATIONS 🧵

File::Path's rmtree and remove_tree will not work with multithreaded applications, and that's because they use chdir. Periodt. Right now, no warning or error gets generated in this situation — it just silently does its thing. And you will certainly encounter unexpected results. 😭💀

The implementation that surfaces this limitation will not be changed. That's it, that's the decision, it's giving permanent.

If you want functionality similar to File::Path but which does not chdir, go look at the File::Path::Tiny module instead. ✨

🌐 NFS Mount Points 🌐

File::Path is not responsible for triggering the automounts, mirror mounts, and the contents of network mounted filesystems. Not its job, not its problem. 🙈

So if your NFS implementation requires an action to be performed on the filesystem in order for File::Path to perform operations, it is strongly suggested you assure filesystem availability by reading the root of the mounted filesystem. Do that first, bestie. 👑

📢 REPORTING BUGS 📢

Please report all bugs on the RT queue, either via the web interface:

http://rt.cpan.org/NoAuth/Bugs.html?Dist=File-Path

or by email:

bug-File-Path@rt.cpan.org

Either way — web post or email — please attach patches to the bug report rather than including them inline in the web post or the body of the email. Attach, don't paste, iykyk. 🎀

You can also send pull requests to the Github repository:

https://github.com/rpcme/File-Path

🫶 ACKNOWLEDGEMENTS 🫶

Paul Szabo identified the race condition originally, and Brendan O'Dea wrote an implementation for Debian that addressed the problem. That code was used as a basis for the current code. Their efforts are greatly appreciated. Legends, honestly. 👑

Gisle Aas made a number of improvements to the documentation for 2.07 and his advice and assistance is also greatly appreciated. Thank you Gisle!! 💖

✨ AUTHORS ✨

Prior authors and maintainers: Tim Bunce, Charles Bailey, and David Landgren <david@landgren.net>.

Current maintainers are Richard Elberger <riche@cpan.org> and James (Jim) Keenan <jkeenan@cpan.org>. 👑

🌈 CONTRIBUTORS 🌈

Contributors to File::Path, in alphabetical order by first name. Give them all their flowers 💐:

  • <bulkdd@cpan.org>
  • Charlie Gonzalez <itcharlie@cpan.org>
  • Craig A. Berry <craigberry@mac.com>
  • James E Keenan <jkeenan@cpan.org>
  • John Lightsey <john@perlsec.org>
  • Nigel Horne <njh@bandsman.co.uk>
  • Richard Elberger <riche@cpan.org>
  • Ryan Yee <ryee@cpan.org>
  • Skye Shaw <shaw@cpan.org>
  • Tom Lutz <tommylutz@gmail.com>
  • Will Sheppard <willsheppard@github>

This module is copyright (C) Charles Bailey, Tim Bunce, David Landgren, James Keenan and Richard Elberger 1995-2020. All rights reserved.

📜 LICENSE 📜

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself. Slay. 💅

And that's the whole page, bestie!! This was File::Path from perl v5.44.0, dated 2026-10-17, page File::Path(3perl). Go forth and rmtree responsibly. 🦋✨💖

📦 where this page comes from