Module Ftp_client

module Ftp_client: sig .. end

FTP client

Warning: Experimental code! Not all features described in this interface work as described!

The client is currently only partially implemented. Especially, the STOR and STOU commands are missing. Only IPv4 is supported.

It is intended to implement a large subset of RFC 959 (and later, the newer FTP-related RFCs). This includes support for EBCDIC, record files, and block mode.

exception FTP_protocol_error of exn

Something went wrong, often on socket level

exception FTP_protocol_violation of string

The server violates the FTP specification

type cmd_state = [ `Failure
       | `Init
       | `Pass_acct_seq
       | `Preliminary
       | `Proto_error
       | `Rename_seq
       | `Restart_seq
       | `Success
       | `User_acct_seq
       | `User_pass_seq ]

The command state:

`Init: Just connected, no greeting message arrived yet
`Success: Got the greeting message/last command was successful
`Proto_error: currently unused
`Failure: Last command was not successful
`Rename_seq: Used instead of `Success after the RNFR command
`Restart_seq: Used instead of `Success after the REST command
`User_pass_seq: Used instead of `Success after the USER command when a password must be typed in
`User_acct_seq: Used instead of `Success after the USER command when an account ID must be typed in
`Pass_acct_seq: Used instead of `Success after the PASS command when an account iD must be typed in
`Preliminary: a reply with code 100 to 199. There will be another final reply for the command

type port = [ `Active of string * int * Unix.file_descr
       | `Passive of string * int
       | `Unspecified ]

The port of the data connection: `Active means that the server initiates the data connection to the listening client, and in the case of `Passive the client initiates the data connection to the listening server. The string argument is the IP address as dotted quad, the int argument is the port number, and the descriptor is the listening master socket.

type form_code = [ `ASA | `Non_print | `Telnet ]

The form_code has a meaning when FTP is used to print files:

`Non_print: This is not the case.
`Telnet: Telnet control codes are used for vertical movement
`ASA: ASA (Fortran) control codes are used for vertical movement

type representation = [ `ASCII of form_code option
       | `EBCDIC of form_code option
       | `Image ]

The representation of the transferred file:

`ASCII: An ASCII variant is used, i.e. the server sends files in ASCII encoding with CR/LF as end-of-line marker. Supported by all servers.
`EBCDIC: An EBCDIC variant is used, i.e. the server sends files in EBCDIC encoding with NEL as end-of-line marker
`Image: The file is transferred in its original binary representation. Supported by all servers.

"Local" representations are not supported.

This FTP client does not recode the files such that they match the selected representation. When files are downloaded, they are stored as they are received. When files are uploaded, they are sent as they are. The user of this client must do recodings when necessary (the class Ftp_data_endpoint.data_converter may be useful for this).

If no representation is selected, FTP servers assume `ASCII None as default.

type structure = [ `File_structure | `Record_structure ]

The client supports two structures:

`File_structure: Files are simply contiguous streams of bytes
`Record_structure: Files are sequences of records. FTP does not make a difference between variable and fixed length records. It is not forbidden that the records are themselves structured into lines, in fact it can happen that end-of-line markers are contained in binary records. Operating systems that support record-structured files often store text files in this format, i.e. every line is stored in its own record, without end-of-line marker. If record structure is selected by a STRU command, it is recommended to use the classes Ftp_data_endpoint.out_record_channel and Ftp_data_endpoint.in_record_channel for the local representation of the files, otherwise the records may be incorrectly mapped to the local conventions.

Page-structured files (i.e. indexed files) are not supported.

If no structure is selected, FTP servers will assume file structure as default.

type transmission_mode = [ `Block_mode | `Stream_mode ]

The transmission mode selects how the data are encoded in the data connection.

`Stream_mode: This is the simple format that is responsible for all the failed FTP downloads. It is supported by all FTP servers, actually, you cannot assume a better transmission mode from an unknown FTP server. It is unreliable because it cannot distinguish between a transmission failure and the regular end-of-file condition.
`Block_mode: This is an improved format using frames to protect the transmitted data. Unfortunately, almost no FTP server supports it.

Both modes are compatible with both structures, i.e. you can transfer a record-structured file in stream mode and a flat file in block mode. However, in practise this is not the case. Servers that only know flat files are likely to only support stream mode, and servers implementing record structure imply that block transfers base on the record format. So the advise is to use stream mode for flat files, and block mode for record files.

type state = {

`cmd_state : cmd_state;`	`(*`	`the command state`	`*)`
`ftp_connected : bool;`	`(*`	`whether connected with the server`	`*)`
`ftp_data_conn : bool;`	`(*`	`whether there is a clean data conn`	`*)`
`ftp_user : string option;`	`(*`	`successfully sent user identifier`	`*)`
`ftp_password : string option;`	`(*`	`successfully sent password`	`*)`
`ftp_account : string option;`	`(*`	`successfully sent account identifier`	`*)`
`ftp_logged_in : bool;`	`(*`	`whether the user is logged in`	`*)`
`ftp_port : port;`	`(*`	`the selected port`	`*)`
`ftp_repr : representation;`	`(*`	`the selected representation`	`*)`
`ftp_structure : structure;`	`(*`	`the selected structure`	`*)`
`ftp_trans : transmission_mode;`	`(*`	`the selected trans mode`	`*)`
`ftp_dir : string list;`	`(*`	`The current directory, expressed as list of CWD changes minus CDUP changes. This is only reasonable if CWD does not include slashes. The list is in reverse order, i.e. deepest directory first.`	`*)`

}

The state reflects the knowledge of the client about what has been agreed upon with the server.

type cmd = [ `ACCT of string
       | `ALLO of int * int option
       | `APPE of string * (state -> Ftp_data_endpoint.local_sender)
       | `CDUP
       | `CWD of string
       | `Connect
       | `DELE of string
       | `HELP of string option
       | `LIST of
           string option * (state -> Ftp_data_endpoint.local_receiver)
       | `MKD of string
       | `MODE of transmission_mode
       | `NLST of
           string option * (state -> Ftp_data_endpoint.local_receiver)
       | `NOOP
       | `PASS of string
       | `PASV
       | `PORT
       | `PWD
       | `QUIT
       | `REIN
       | `REST of string
       | `RETR of string * (state -> Ftp_data_endpoint.local_receiver)
       | `RMD of string
       | `RNFR of string
       | `RNTO of string
       | `SITE of string
       | `SMNT of string
       | `STAT of string option
       | `STOR of string * (state -> Ftp_data_endpoint.local_sender)
       | `STOU of unit -> Ftp_data_endpoint.local_sender
       | `STRU of structure
       | `SYST
       | `TYPE of representation
       | `USER of string ]

An FTP command. Currently, STOR, STOU and APPE are not supported.

type reply = int * string

Reply code plus text

class ftp_client_pi : ?event_system:Unixqueue.event_system -> ?onempty:state -> unit -> ?onclose:unit -> unit -> Unix.file_descr -> object .. end

The client protocol interpreter...

class type ftp_method = object .. end

An ftp_method is a small procedure doing some task

exception FTP_method_failure of int * string

This exception may be raised during execution by the FTP method. The int is the unexpected FTP control code and the string the corresponding text.

This does not work yet as intended!

class connect_method : host:string -> ?port:int -> unit -> ftp_method

This method connects to the host which must be an IP address

class login_method : user:string -> get_password:(unit -> string) -> get_account:(unit -> string) -> unit -> ftp_method

This FTP method logs the user in.

class get_method : file:string -> representation:representation -> store:(state -> Ftp_data_endpoint.local_receiver) -> unit -> ftp_method

This FTP method walks to the right directory and gets file from the server.

class ftp_client : ?event_system:Unixqueue.event_system -> ?onempty:unit -> unit -> unit -> object .. end

The ftp client is a user session that may even span several connections.

Examples and Discussion

To download a single flat file from a server:

 
   let buffer = Buffer.create 1000 in
   let ch = new Netchannels.output_buffer buffer in
   let client = new ftp_client() in
   client # add (new connect_method ~host:"127.0.0.1");
   client # add (new login_method ~user:"foo"
                                  ~get_password:(fun () -> "password")
                                  ~get_account:(fun () -> "foo") ());
   client # add (new get_method ~file:"path/to/file"
                                ~representation:`Image
                                ~store:(fun _ -> `File_structure ch) ());
   client # run()

The file is stored in buffer. By using a different netchannel, it can be stored whereever wanted.

This piece of code has the disadvantage that the client does not stop when a method fails (in the current version, it always stops, but in a disadvantegous manner - this will be changed). Because of this, it is better to execute the next method only when the previous was successful:

 
   ...
   client # add ~onsuccess:(fun () ->
                               client # add ~onsuccess:...
                                               (new login_method 
                                                   ~user:"foo"
                                                ~get_password:(fun () -> "password")
                                                ~get_account:(fun () -> "foo") ());
                           )
                (new connect_method ~host:"127.0.0.1")

Alternatively, one can also force that the execution stops by raising an exception in the onerror callback (not implemented in the current version).

To download a record-structured text file, use a store like:

 
    let ch = (as above) in
    let rec_ch = new Ftp_data_endpoint.write_out_record_channel
                       ~repr:(`ASCII_unix `Enc_iso88591)
                       ch
    ...
    ... ~store:(fun _ -> `Record_structure rec_ch)

Here, the end-of-record is transcoded to newline. Note that the ASCII variant (`Enc_iso88591) is ignored by write_out_record_channel. Open: How to select record structure using an FTP method.

Character conversions: To convert an EBCDIC file to ASCII, use something like

 
    let ch = (as above) in
    let converter = new Ftp_data_endpoint.data_converter
                         ~fromrepr:(`EBCDIC `Enc_cp1047)
                         ~torepr:(`ASCII_unix `Enc_iso88591) in
    let ch_ebcdic = new Netchannels.output_filter converter ch in
    ...
    ... ~representation:(`EBCDIC None)
    ... ~store:(fun _ -> `File_structure ch_ebcdic)

The class data_converter also performs the transformation of the end-of-line convention, unlike the similar class Netconversion.conversion_pipe.