async-rustbus

This crate implements an async API for interacting with DBus. Beyond being just async, it is also thread-safe, allowing for a single connection to be used by multiple threads. It is built on top of KillingSpark's existing rustbus crate and reuses its implementation of the DBus wire format and message protocol. A non-blocking reimplementation of rustbus's Conn is used to build an async version of RpcConn. This crate's async version of RpcConn creates an async API for connecting to system, session, and remote (TCP) DBus daemons, sending messages, receving method calls and signals.

Design differences between this crate and rustbus and why

While the async crate reuses the message format of the parent crate, there were some design decisions that differ from the orignal when handling connections:

• Most RpcConn functions in this crate return std::io::Results rather than a custom error type. When sending/receiving messages the errors that can occur are IO errors, (un)marshalling errors and invalid headers. Because most user's are typically interacting with well tested DBus implementations, (un)marshalling errors are likely caused by bugs in our implementation. Marshalling errors are typically caught by the remote dbus-daemon, causing a disconnect, which results in a future std::io::Error. Invalid headers and invalid Fds are also possible sources of errors but typically bugs on the user's end that should be solved with bugfixes; they do not need to handled at runtime. This really leaves only IO errors that needed to be handled at runtime. Other errors are propagated custom std::io::Errors;

• This crate terminates connections when protocol violations occur (the original allows for attempting to recover). When protocol violations occur, there is no safe and consistent way to recover. While keeping the connection open can be helpful for debugging it is generally useless for the user of the crate. Also the DBus Specification specifies that no recovery should be attempted:

For security reasons, the D-Bus protocol should be strictly parsed and validated, with the exception of defined extension points. Any invalid protocol or spec violations should result in immediately dropping the connection without notice to the other end.

• The async-version provides more mechanisms for message handling than the original. This funcitonality was primarly done for ergonimics and to make the async RpcConn easier to use in mulitple distinct threads without having to cooridinate with each other when receiving messages.

  • The original crate requires the user manually wait for responses. When creating method calls in this crate, this crate provides a Future that returns another Future. The outer Future sends the message and the nested Future retreives the response. In the event that the nested Future is dropped then response will be ignored and cleaned up apporiately.

  • Incoming method calls are sorted into a hierarchy of MsgQueues. This was done to allow different threads, to easily retreive only calls (using get_call()) from the path_namespace in which they are interested. The hierarchy also can provide basic handling of some Introspect calls.

  • A mechanism for filtering signals is also provided. Individual threads can register matches (with insert_sig_match()) to allow retrieve signals that the thread is interested in. This call also handles the AddMatch call to the DBus daemon. When signals are recieved by the RpcConn they are checked against each match in order provided by their specificity rules (see async_rustbus::Match for details about ordering). The first rule that matches has the signal added to its queue to be retrieved later by interested threads (with get_signal()).

• This crate's RpcConn sends and waits for the DBus hello message on connection. The original requires that you manually handle this message and response.

• This crate support TCP streams for connection to remote servers. Note: TCP streams are unencrypted and should be done using an SSH-tunnel or something else.

• The original crate allows for finer control of the how information is written to and received from the socket, with ability to set timeouts and choice when buffers are refilled. Async functionality generally elimnates any necessity for this by abstracting over non-blocking system calls.

Status

This crate has yet to be published and still needs documentation, however most of the functionality is implemented. The signal implementation has not been validated, but everything else is in working order.