tftp-server

A TFTP server implementation in Rust

Summary

• Usable as both binary and library
• 100% safe code, no unsafe usage
• Well tested, including error cases
• Implements the RFCs describing extensions to the TFTP protocol

Building and running

Simply use cargo build, and cargo run respectively. The server will start and serve from/into the current directory.

Note: By default the server listens on port 69 (as per RFC1350), and that usually requires root privileges.

In order to run it on a different port, you can specify --address:

$ ./target/debug/tftp_server --address 0.0.0.0:1234

This will instead listen on port 1234.

To bind to a random port you can leave out the port number and only specify the address

Features

All features are implemented in the library. The binary target is a only an argument-parsing thin wrapper over it for direct usage conveninence.

• -a or --address to specify an address[:port] to listen on (multiple supported)
• -r will make the server treat the served directory as read-only (it will reject all write requests)
• -d or --directory specifies the directory to serve from (the given path will be prepended to all requested paths)
• -t or --timeout specifies the default timeout (in seconds) for idle connections
• see TODO section below

TFTP Protocol Options & Extensions

The following TFTP extension RFCs are implemented:

• RFC 2347: TFTP Option Extension
• RFC 2348: TFTP Blocksize Option
• RFC 2349: TFTP Timeout Interval and Transfer Size Options
• RFC 7440: TFTP Windowsize Option

Logging and Testing

To run all tests, use cargo test.

You can also run the server (or tests) with logging enabled. To do this add RUST_LOG=tftp_server=info before the command. For example:

$ RUST_LOG=tftp_server=info ./target/debug/tftp_server

This will run the server with logging enabled so that you can inspect the program's behavior.

TODOs

• Documentation for individual items
• Crate-level documentation with an overview and examples
• serve from specified directory, not just the current one
• treat directory as readonly (reject write requests)
• IPv6 support
• multiple address support
• CLI switches for logging
• running control (ability to stop server hard or soft)
• limit accepted blocksize to stack MSS (smaller on ipv4)
• complete implementation of all option extension RFCs
• redo packets as in-place buffer references to avoid copying memory
• redo integration tests to run them with harness
• make proto tests more orthogonal
• test that transfer size is enforced on Rx
• maybe eventually split off proto handling into its own crate
• implement congestion control when using window size
• complete Response implementation to make it efficient instead of storing a Vec