15.2 Protocol

Each exchange between a client and server starts with the client asking the server to either read a file for the client or write a file for the client. In the normal case of bootstrapping a diskless system, the first request is a read request (RRQ). Figure 15.1 shows the format of the five TFTP messages. (Opcodes 1 and 2 share the same format.)

Figure 15.1. Format of the five TFTP messages.

The first 2 bytes of the TFTP message are an opcode. For a read request (RRQ) and write request (WRQ) the filename specifies the file on the server that the client wants to read from or write to. We specifically show that this filename is terminated by a byte of 0 in Figure 15.1. The mode is one of the ASCII strings netascii or octet (in any combination of uppercase or lowercase), again terminated by a byte of 0. netascii means the data are lines of ASCII text with each line terminated by the 2-character sequence of a carriage return followed by a linefeed (called CR/LF). Both ends must convert between this format and whatever the local host uses as a line delimiter . An octet transfer treats the data as 8-bit bytes with no interpretation.

Each data packet contains a block number that is later used in an acknowledgment packet. As an example, when reading a file the client sends a read request (RRQ) specifying the filename and mode. If the file can be read by the client, the server responds with a data packet with a block number of 1. The client sends an ACK of block number 1. The server responds with the next data packet, with a block number of 2. The client sends an ACK of block number 2. This continues until the file is transferred. Each data packet contains 512 bytes of data, except for the final packet, which contains 0-511 bytes of data. When the client receives a data packet with less than 512 bytes of data, it knows it has received the final packet.

In the case of a write request (WRQ), the client sends the WRQ specifying the filename and mode. If the file can be written by the client, the server responds with an ACK of block number 0. The client then sends the first 512 bytes of file with a block number of 1. The server responds with an ACK of block number 1.

This type of data transmission is called a stop-and-wait protocol. It is found only in simple protocols such as TFTP. We'll see in Section 20.3 that TCP provides a different form of acknowledgment, which can provide higher throughput. TFTP is designed for simplicity of implementation, not high throughput.

The final TFTP message type is the error message, with an opcode of 5. This is what the server responds with if a read request or write request can't be processed . Read and write errors during file transmission also cause this message to be sent, and transmission is then terminated. The error number gives a numeric error code, followed by an ASCII error message that might contain additional, operating system specific information.

Since TFTP uses the unreliable UDP, it is up to TFTP to handle lost and duplicated packets. Lost packets are detected with a timeout and retransmission implemented by the sender. (Be aware of a potential problem called the " sorcerer 's apprentice syndrome" that can occur if both sides time out and retransmit. Section 12.2 of [Stevens 1990] shows how the problem can occur.) As with most UDP applications, there is no checksum in the TFTP messages, which assumes any corruption of the data will be caught by the UDP checksum (Section 11.3).