Inside the File Transfer Protocol

The File Transfer Protocol defines a standard protocol for transferring files between computers. The main documents that define FTP are RFC 959: File Transfer Protocol (FTP) and RFC 1123: Requirements for Internet Hosts -- Application and Support.

Requirements

An embedded system can function as an FTP client or server. A client initiates communications with a server and sends requests to transmit or receive files. In most cases, an embedded system that needs to exchange files with a single PC should function as a client. Many embedded systems don’t have a lot of resources to spare, and running an FTP server that is always available requires processing time and memory. Running a server also puts the system at a greater security risk because any computer in the network might be able to gain access to the system’s files. But if the embedded system needs to make its files available to anyone on the network, or if the files need to be available to other computers at all times, the system will need to function as a server. A computer that uses FTP must have a file system, which enables the system to store information in named entities called files. Files are of course useful in desktop computers, where you select files to run programs, view documents and images, and perform other tasks. Embedded systems can support file systems as well. A small embedded system may just store data in specified locations in memory, with no need to place the data in named files. But for many embedded systems, a file system provides a useful structure for accessing information, both locally and over a network. For example, a system can store collected data or configuration settings in files.

A system functioning as an FTP client can initiate communications periodically with a remote computer to request to send or receive files. A system functioning as an FTP server can make its files available on request and can allow remote computers to send files that the system will use. The user that communicates with the embedded system can be a human using an FTP program or a process that functions without human intervention. For example, a PC can be programmed to retrieve a file once a day from an embedded system. In PCs, the file system includes the ability to store files in a directory structure and to specify attributes such as whether a file is write-protected or accessible to certain users. Under Windows XP, from the My Computer folder, you can browse the directories and view file names and attributes. (In the View menu, click Choose Details to specify what information to display and click Details to view the information.) The TINI supports a similar file system, which you can browse from the slush shell using commands such as ls -l and cd. A very basic file system might just consist of a structure with a series of entries that each store the name, starting address in memory, and length of a file. In Dynamic C, entries in an HttpSpec or ServerSpec structure can specify files that are accessible to a Web or FTP server. Each entry includes a file name, the address in memory where the file’s length and contents are stored, and optional security information. On a PC, you can perform FTP transfers using an FTP client application such as WS_FTP from Ipswitch, Inc. Two other ways to perform FTP transfers are from a command prompt and from a browser. To use the command- line interface, enter ftp at a command prompt and enter ? for a list of commands. The browser interface is explained later in this networking tutorial. Ipswitch and others also offer applications that enable a PC to function as an FTP server.

Transferring a File

To transfer a file, an FTP session uses two channels, or communications paths, one for control information and one for the file being transferred. Each channel has a separate TCP connection. On the server, the default port for the control channel is 21 and the default port for the data channel is 20. The client can use any available port or ports. The default for the client is to use the same port for both the control and data channels. However, transfers that use FTP’s stream mode, which requires a new data connection for each file, should send a PORT command to specify a new, non-default port for each file transfer. Requesting a new port for each transfer prevents problems due to TCP’s timeout requirements. When a connection closes, TCP requires a timeout before the same connection can be reused. The timeout prevents a new connection that is identical to a recently closed connection from receiving data intended for the previous connection. When transferring multiple files in a single session, if a transfer tries to use the same port as the previous connection, the port may be unavailable because thes timeout for the previous connection hasn’t expired. Specifying a different port for each data connection eliminates the problem. Other alternatives are to use the block or compressed transfer modes, which don’t require a new data connection for each file. These are typical steps in sending a file to a server in stream mode, where the file’s contents are sent without a header or any assumed structure for the file’s data:

1. The client opens a control channel between any available local port and port 21 on the server. The client sends commands to establish communications and request to send a file.

2. The server opens a data channel between the server’s port 20 and the port the client is using for the control channel.

3. The client sends the file’s contents, closes the data channel, and requests the server to close the control channel.

4. The server closes the control channel.

In a similar way, these are the steps in receiving a file from a server in stream mode:

1. The client opens a control channel between any available local port and port 21 on the server. The client sends commands to establish communications and request a file.

2. The server opens a data channel between the server’s port 20 and the port the client is using for the control channel. The server sends the file and closes the data channel.

3. The client requests the server to close the control channel.

4. The server closes the control channel.

A client that is communicating from behind a firewall may find that the firewall blocks the server’s request to open the data connection. To get around this limitation without having to reconfigure the firewall, the client can send a command that requests a passive transfer process (PASV or EPSV), where the client, rather than the server, opens the data connection. The client must send the command to request a passive transfer preceding each transfer. When a client specifies the location of a file on a server, the location is relative to the directories that the server makes available to the client. This location can differ from the file’s absolute location in the computer. For example, a computer functioning as a server may allow the user to access the directory /ftp/user1 and its subdirectories. The server’s root directory for that user is then /user1. To access a file at /ftp/user1/data/test.txt, the client would specify the location on the server as /data/test.txt, which is the file’s location relative to the user’s root directory.