Handling files is security-sensitive. It has led in the past to the following vulnerabilities:

Any access to the file system can create a vulnerability. Exposing a file's content, path or even its existence or absence is dangerous. It is also extremely risky to create or write files without making sure that their permission and content is safe and controlled. Using a file path or reading a file content must be always done with caution as they could have been tampered with.

The file system is a resource which can be easily exhausted. Opening too many files will use up all file descriptors, preventing other software from opening files. Filling the storage space will also prevent any additional write from happening.

This rule raises an issue on filesystem functions which have a high impact on security and are often subject to injection. The goal is to guide manual security code reviews.

Ask Yourself Whether

You are at risk if you answered yes to any of those questions.

Recommended Secure Coding Practices

Avoid using paths provided by users or other untrusted sources if possible. If this is required, check that the path does not reference an unauthorized directory or file. See OWASP recommendations as to how to test for directory traversal. Note that the paths length should be validated too.

No File and directory names should be exposed. They can contain sensitive information. This means that a user should not be able to list the content of unauthorized directories.

Make sure that no attackers can test for the existence or absence of sensitive files. Knowing that a specific file exists can reveal a vulnerability or at least expose file and directory names.

Files and directories should be created with restricted permissions and ownership. Only authorized users and applications should be able to access the files, and they should have as little permissions as needed. Modifying a file's permissions is not good enough. The permissions should be restricted from the very beginning.

Writing user input into files should be done with caution. It could fill the storage space if the amount of data written is not controlled. It could also write dangerous data which will later be used by an application or returned to another user. This is why the user input should be validated before being written.

Reading a file can lead to other vulnerabilities. Any file could have been modified by an attacker. Thus the same validation as for any user input should be performed on file content.

Once a file is read, its content should only be exposed to authorized users.

Add limits to the number of files your application access simultaneously or create because of a user action. It is possible to perform a Denial of Service attack by opening too many files, and thus exhausting available file descriptors, or by filling the file system with new files. Release file descriptors by closing files as soon as possible.

We also recommended to have tools monitoring your system and alerting you whenever resources are nearly exhausted.

Do not allow untrusted code to access the filesystem. For some programming languages, child-processes may have access to file descriptors opened by the parent process before the creation of the child process. This creates a vulnerability when a child process doesn't have the permission to access a file but is still able to modify it via the inherited file descriptor. Check your language documentation for "file descriptor leak" or the use of the flags O_CLOEXEC, FD_CLOEXEC, or bInheritHandles. File descriptors can be inherited in the following languages: C, C++, C#, Objective-C, Swift, Go (but disabled by default), some JVM versions, Javascript and TypeScript in Nodejs, Some PHP versions, Python, Ruby, Rust, VB6 and VB.NET.

Questionable Code Example

function handle_file($filename, $directory, $group, $data, $mode, $flags, $use_include_path, $pattern, $recursive, $context)
{
    file_put_contents($filename, $data, $flags); // Questionable
    copy($filename, $filename); // Questionable
    tmpfile(); // Questionable
    parse_ini_file($filename); // Questionable

    // The following calls will raise an issue if and only if the $filename or $directory is not hardcoded
    move_uploaded_file($filename, $filename); // Questionable
    rmdir($directory); // Questionable
    unlink($filename); // Questionable

    move_uploaded_file("mypath1", "mypath2"); // Compliant
    rmdir("dir1"); // Compliant
    unlink("dir2"); // Compliant

    // The following functions can also be used to perform network requests (http, socket, ftp, etc...)
    // in some case they won't raise issues, see below.
    file_get_contents($filename, $use_include_path); // Questionable
    file($filename, $flags); // Questionable
    fopen($filename, $mode, $use_include_path); // Questionable
    readfile($filename, $use_include_path); // Questionable

    // No issue is raised if the source path, cleaned from prefixed wrappers (like 'rar://', 'zlib://', 'bzip2://', 'zip://', 'compress.zlib://',
    // 'compress.bzip2://', 'glob://', 'ogg://') still contains '://' without starting by 'file://'
    file_get_contents("zip://http://example.com/file.zip" $use_include_path); // Compliant
    file("http://example.com", $flags); // Compliant
    fopen("http://example.com", $mode, $use_include_path); // Compliant
    readfile("http://example.com", $use_include_path); // Compliant

    // No issue is created if a context is given as there is a high chance that it is not a filesystem access.
    // Note that this will create some false negatives with "zip" contexts.
    file_get_contents($filename, $use_include_path, $context); // Compliant
    file($filename, $flags, $context); // Compliant
    fopen($filename, $mode, $use_include_path, $context); // Compliant
    readfile($filename, $use_include_path, $context); // Compliant
}

See