What is pyftpdlib?

pyftpdlib is a high-level library to easily write asynchronous portable FTP servers with Python.

What is Python?

Python is an interpreted, interactive, object-oriented, easy-to-learn programming language. It is often compared to Tcl, Perl, Scheme or Java.

I’m not a python programmer. Can I use it anyway?

Yes. pyftpdlib is a fully working FTP server implementation that can be run “as is”. For example you could run an anonymous ftp server from cmd-line by running:

$ sudo python -m pyftpdlib
[I 13-02-20 14:16:36] >>> starting FTP server on <<<
[I 13-02-20 14:16:36] poller: <class 'pyftpdlib.ioloop.Epoll'>
[I 13-02-20 14:16:36] masquerade (NAT) address: None
[I 13-02-20 14:16:36] passive ports: None
[I 13-02-20 14:16:36] use sendfile(2): True

This is useful in case you want a quick and dirty way to share a directory without, say, installing and configuring samba. Starting from version 0.6.0 options can be passed to the command line (see python -m pyftpdlib --help to see all available options). Examples:

Anonymous FTP server with write access:

$ sudo python -m pyftpdlib -w
~pyftpdlib-1.3.1-py2.7.egg/pyftpdlib/authorizers.py:265: RuntimeWarning: write permissions assigned to anonymous user.
[I 13-02-20 14:16:36] >>> starting FTP server on <<<
[I 13-02-20 14:16:36] poller: <class 'pyftpdlib.ioloop.Epoll'>
[I 13-02-20 14:16:36] masquerade (NAT) address: None
[I 13-02-20 14:16:36] passive ports: None
[I 13-02-20 14:16:36] use sendfile(2): True

Listen on a different ip/port:

$ python -m pyftpdlib -i -p 8021
[I 13-02-20 14:16:36] >>> starting FTP server on <<<
[I 13-02-20 14:16:36] poller: <class 'pyftpdlib.ioloop.Epoll'>
[I 13-02-20 14:16:36] masquerade (NAT) address: None
[I 13-02-20 14:16:36] passive ports: None
[I 13-02-20 14:16:36] use sendfile(2): True

Customizing ftpd for basic tasks like adding users or deciding where log file should be placed is mostly simply editing variables. This is basically like learning how to edit a common unix ftpd.conf file and doesn’t really require Python knowledge. Customizing ftpd more deeply requires a python script which imports pyftpdlib to be written separately. An example about how this could be done are the scripts contained in the demo directory.

Getting help

There’s a mailing list available at: http://groups.google.com/group/pyftpdlib/topics

Installing and compatibility

How do I install pyftpdlib?

If you are not new to Python you probably don’t need that, otherwise follow the install instructions.

Which Python versions are compatible?

From 2.6 to 3.4. Python 2.4 and 2.5 support has been removed starting from version 0.6.0. The latest version supporting Python 2.3 is pyftpdlib 1.4.0. Python 2.3 support has been removed starting from version 0.6.0. The latest version supporting Python 2.3 is pyftpdlib 0.5.2.

On which platforms can pyftpdlib be used?

pyftpdlib should work on any platform where select(), poll(), epoll() or kqueue() system calls are available and on any Python implementation which refers to cPython 2.6 or superior. The development team has mainly tested it under various Linux, Windows, OSX and FreeBSD systems. For FreeBSD is also available a pre-compiled package maintained by Li-Wen Hsu (lwhsu@freebsd.org). Other Python implementation like PythonCE are known to work with pyftpdlib and every new version is usually tested against it. pyftpdlib currently does not work on Jython since the latest Jython release refers to CPython 2.2.x serie. The best way to know whether pyftpdlib works on your platform is installing it and running its test suite.


How can I run long-running tasks without blocking the server?

pyftpdlib is an asynchronous FTP server. That means that if you need to run a time consuming task you have to use a separate Python process or thread for the actual processing work otherwise the entire asynchronous loop will be blocked.

Let’s suppose you want to implement a long-running task every time the server receives a file. The code snippet below shows the correct way to do it by using a thread.

With self.del_channel() we temporarily “sleep” the connection handler which will be removed from the async IO poller loop and won’t be able to send or receive any more data. It won’t be closed (disconnected) as long as we don’t invoke self.add_channel(). This is fundamental when working with threads to avoid race conditions, dead locks etc.

class MyHandler(FTPHandler):

    def on_file_received(self, file):
        def blocking_task():


Another possibility is to change the default concurrency model.

Why do I get socket.error “Permission denied” error on ftpd starting?

Probably because you’re on a Unix system and you’re trying to start ftpd as an unprivileged user. FTP servers bind on port 21 by default and only super-user account (e.g. root) can bind sockets on such ports. If you want to bind ftpd as non-privileged user you should set a port higher than 1024.

Can control upload/download ratios?

Yes. Starting from version 0.5.2 pyftpdlib provides a new class called ThrottledDTPHandler. You can set speed limits by modifying read_limit and write_limit class attributes as it is shown in throttled_ftpd.py demo script.

Are there ways to limit connections?

FTPServer. class comes with two overridable attributes defaulting to zero (no limit): max_cons, which sets a limit for maximum simultaneous connection to handle by ftpd and max_cons_per_ip which set a limit for connections from the same IP address. Overriding these variables is always recommended to avoid DoS attacks.

I’m behind a NAT / gateway

When behind a NAT a ftp server needs to replace the IP local address displayed in PASV replies and instead use the public address of the NAT to allow client to connect. By overriding masquerade_address attribute of FTPHandler class you will force pyftpdlib to do such replacement. However, one problem still exists. The passive FTP connections will use ports from 1024 and up, which means that you must forward all ports 1024-65535 from the NAT to the FTP server! And you have to allow many (possibly) dangerous ports in your firewalling rules! To resolve this, simply override passive_ports attribute of FTPHandler class to control what ports pyftpdlib will use for its passive data transfers. Value expected by passive_ports attribute is a list of integers (e.g. range(60000, 65535)) indicating which ports will be used for initializing the passive data channel. In case you run a FTP server with multiple private IP addresses behind a NAT firewall with multiple public IP addresses you can use passive_ports option which allows you to define multiple 1 to 1 mappings (New in 0.6.0).

What is FXP?

FXP is part of the name of a popular Windows FTP client: http://www.flashfxp.com. This client has made the name “FXP” commonly used as a synonym for site-to-site FTP transfers, for transferring a file between two remote FTP servers without the transfer going through the client’s host. Sometimes “FXP” is referred to as a protocol; in fact, it is not. The site-to-site transfer capability was deliberately designed into RFC-959. More info can be found here: http://www.proftpd.org/docs/howto/FXP.html.

Does pyftpdlib support FXP?

Yes. It is disabled by default for security reasons (see RFC-2257 and FTP bounce attack description) but in case you want to enable it just set to True the permit_foreign_addresses attribute of FTPHandler class.

Why timestamps shown by MDTM and ls commands (LIST, MLSD, MLST) are wrong?

If by “wrong” you mean “different from the timestamp of that file on my client machine”, then that is the expected behavior. Starting from version 0.6.0 pyftpdlib uses GMT times as recommended in RFC-3659. In case you want such commands to report local times instead just set the use_gmt_times attribute to False. For further information you might want to take a look at this



Starting from version 0.7.0 if pysendfile module is installed sendfile(2) system call be used when uploading files (from server to client) via RETR command. Using sendfile(2) usually results in transfer rates from 2x to 3x faster and less CPU usage. Note: use of sendfile() might introduce some unexpected issues with “non regular filesystems” such as NFS, SMBFS/Samba, CIFS and network mounts in general, see: http://www.proftpd.org/docs/howto/Sendfile.html. If you bump into one this problems the fix consists in disabling sendfile() usage via FTPHandler.use_sendfile option:

from pyftpdlib.handlers import FTPHandler
handler = FTPHandler
handler.use_senfile = False

Globbing / STAT command implementation

Globbing is a common Unix shell mechanism for expanding wildcard patterns to match multiple filenames. When an argument is provided to the STAT command, ftpd should return directory listing over the command channel. RFC-959 does not explicitly mention globbing; this means that FTP servers are not required to support globbing in order to be compliant. However, many FTP servers do support globbing as a measure of convenience for FTP clients and users. In order to search for and match the given globbing expression, the code has to search (possibly) many directories, examine each contained filename, and build a list of matching files in memory. Since this operation can be quite intensive, both CPU- and memory-wise, pyftpdlib does not support globbing.

ASCII transfers / SIZE command implementation

Properly handling the SIZE command when TYPE ASCII is used would require to scan the entire file to perform the ASCII translation logic (file.read().replace(os.linesep, ‘rn’)) and then calculating the len of such data which may be different than the actual size of the file on the server. Considering that calculating such result could be very resource-intensive it could be easy for a malicious client to try a DoS attack, thus pyftpdlib rejects SIZE when the current TYPE is ASCII. However, clients in general should not be resuming downloads in ASCII mode. Resuming downloads in binary mode is the recommended way as specified in RFC-3659.

IPv6 support

Starting from version 0.4.0 pyftpdlib supports IPv6 (RFC-2428). If you use IPv6 and want your FTP daemon to do so just pass a valid IPv6 address to the FTPServer class constructor. Example:

>>> from pyftpdlib.servers import FTPServer
>>> address = ("::1", 21)  # listen on localhost, port 21
>>> ftpd = FTPServer(address, FTPHandler)
>>> ftpd.serve_forever()
Serving FTP on ::1:21

If your OS (for example: all recent UNIX systems) have an hybrid dual-stack IPv6/IPv4 implementation the code above will listen on both IPv4 and IPv6 by using a single IPv6 socket (New in 0.6.0).

How do I install IPv6 support on my system?

If you want to install IPv6 support on Linux run “modprobe ipv6”, then “ifconfig”. This should display the loopback adapter, with the address “::1”. You should then be able to listen the server on that address, and connect to it. On Windows (XP SP2 and higher) run “netsh int ipv6 install”. Again, you should be able to use IPv6 loopback afterwards.

Can pyftpdlib be integrated with “real” users existing on the system?

Yes. Starting from version 0.6.0 pyftpdlib provides the new UnixAuthorizer and WindowsAuthorizer classes. By using them pyftpdlib can look into the system account database to authenticate users. They also assume the id of real users every time the FTP server is going to access the filesystem (e.g. for creating or renaming a file) the authorizer will temporarily impersonate the currently logged on user, execute the filesystem call and then switch back to the user who originally started the server. Example UNIX and Windows FTP servers contained in the demo directory shows how to use UnixAuthorizer and WindowsAuthorizer classes.

What about SITE commands?

The only supported SITE command is SITE CHMOD (change file mode). The user willing to add support for other specific SITE commands has to define a new ftp_SITE_CMD method in the FTPHandler subclass and add a new entry in proto_cmds dictionary. Example:

from pyftpdlib.handlers import FTPHandler

proto_cmds = FTPHandler.proto_cmds.copy()
    {'SITE RMTREE': dict(perm='R', auth=True, arg=True,
      help='Syntax: SITE <SP> RMTREE <SP> path (remove directory tree).')}

class CustomizedFTPHandler(FTPHandler):
    proto_cmds = proto_cmds

def ftp_SITE_RMTREE(self, line):
    """Recursively remove a directory tree."""
    # implementation here
    # ...