Character Encoding

Overview

FTP is a rather old protocol and things we take for granted now were not even considered 20 years ago. One of these things is support for non-English characters. When the FTP protocol was designed, computers mostly spoke English and were unable to display any non-English characters like umlauts, accented letters or even completely different scripts like for example Chinese. As such, the FTP protocol has been designed to be used with English characters only, namely 7-bit ASCII.

The problem is that many FTP clients and servers purposely violate the FTP specifications in order to support other, non-standard character sets. Which of these character sets are used is not subject to any negotiation and completely arbitrary. For any character set in existence, you can find a server using it with no way of detecting the proper encoding.

To solve this problem, the FTP protocol has been extended in a backwards compatible way to use UTF-8 as character set, which is a strict superset of the previously used character set. Note that this obviously can only be backwards compatible with servers in compliance with the original specifications.

If you have problems with filenames containing any foreign characters, this can have two reasons:

• The server or client follows the original specifications by the letter and rightfully rejects those filenames
• The server or client violates the specifications and uses a custom encoding

Note that both FileZilla Client and Server are fully compliant with the updated specifications and use UTF-8.

If you have problems with other clients or servers, please upgrade to FTP software capable of UTF-8 or refrain from using foreign characters. Anything else is in violation to the FTP specifications and does not work.

Technical details

The FTP protocol is specified in RFC 959, which got published in 1985. The FTP protocol is designed on top of the original Telnet protocol, which is specified in RFC 854. The relevant sections of the Telnet specification regarding FTP are those covering the Network Virtual Terminal (NVT). According to RFC 854, the NVT requires to use 7-bit ASCII as character set with any other character set being subject of explicit negotiation. This character set only contains 127 different characters: English letters and numbers, punctuation characters and a few control characters. Accented letters, umlauts or other scripts are not contained in the ASCII character set.

In order to support non-English characters, the FTP specifications have been extended 1999 in RFC 2640. This extension requires the use of UTF-8 as character set. This character set is a strict superset of ASCII, every valid ASCII character is also the same character in UTF-8. The UTF-8 character set can display any valid Unicode character. That includes umlauts, accented letters and also different scripts. This extension is fully backwards compatible. As long as you're not using any non-English characters, it doesn't matter if the used software supports RFC 2640 or not. Note that if you used non-English characters before using RFC 2640 compatible software, there will be problems. Problems which are entirely self-made by not obeying the specifications.

UTF8 feature negotiation

An RFC 2640 compliant server must support the FEAT command and must include a line containing UTF8 in its response:

Command:  FEAT
Response: 211-Features:
 [...]
Response:  UTF8
 [...]
Response: 211 End

Conflicting specification

There exists a long expired IETF draft that is in conflict to RFC 2640. This draft also requires the FEAT response to include UTF8, but in addition requires the client to send OPTS UTF-8 ON to enable UTF-8 support.

If an RFC 2640 compliant client sends OPTS UTF-8 ON, it has to use UTF-8 regardless whether OPTS UTF-8 ON succeeds or not.

RFC 2640 compliant servers must not make UTF-8 dependent on OPTS UTF-8 ON.

SFTP

The situation for SFTP is similar to the one of FTP. Current versions of the SFTP specifications require filenames to be encoded as UTF-8, beginning with version 4 of the SFTP specifications.

However, the most commonly used SFTP protocol version is version 3 as implemented in OpenSSH. This version of the SFTP specifications does not require UTF-8. In fact it does not say anything about the encoding. It is however reasonable to assume UTF-8 on those servers anyhow for the following reasons:

• The later protocol versions require UTF-8
• The SSH protocol, under which SFTP operates, already requires UTF-8
• Even in version 3 of the protocol, some parts of the protocol already use UTF-8
• The native character set on most modern Unix(-like) operating systems is UTF-8

In essence this means that everywhere where SFTP is available, the necessary infrastructure to use UTF-8 is in place.