D-BUS Tutorial Version 0.2 10 August 2004 Havoc Pennington Red Hat, Inc.
hp@pobox.com
What is D-BUS? D-BUS is a system for interprocess communication (IPC). Architecturally, it has several layers: A library, libdbus, that allows two applications to connect to each other and exchange messages. A message bus daemon executable, built on libdbus, that multiple applications can connect to. The daemon can route messages from one application to zero or more other applications. Wrapper libraries based on particular application frameworks. For example, libdbus-glib and libdbus-qt. There are also bindings to languages such as Python. These wrapper libraries are the API most people should use, as they simplify the details of D-BUS programming. libdbus is intended to be a low-level backend for the higher level bindings. Much of the libdbus API is only useful for binding implementation. If you just want to use D-BUS and don't care how it works, jump directly to . Otherwise, read on. libdbus only supports one-to-one connections, just like a raw network socket. However, rather than sending byte streams over the connection, you send messages. Messages have a header identifying the kind of message, and a body containing a data payload. libdbus also abstracts the exact transport used (sockets vs. whatever else), and handles details such as authentication. The message bus daemon forms the hub of a wheel. Each spoke of the wheel is a one-to-one connection to an application using libdbus. An application sends a message to the bus daemon over its spoke, and the bus daemon forwards the message to other connected applications as appropriate. Think of the daemon as a router. The bus daemon has multiple instances on a typical computer. The first instance is a machine-global singleton, that is, a system daemon similar to sendmail or Apache. This instance has heavy security restrictions on what messages it will accept, and is used for systemwide communication. The other instances are created one per user login session. These instances allow applications in the user's session to communicate with one another. The systemwide and per-user daemons are separate. Normal within-session IPC does not involve the systemwide message bus process and vice versa. D-BUS applications There are many, many technologies in the world that have "Inter-process communication" or "networking" in their stated purpose: MBUS, CORBA, XML-RPC, SOAP, and probably hundreds more. Each of these is tailored for particular kinds of application. D-BUS is designed for two specific cases: Communication between desktop applications in the same desktop session; to allow integration of the desktop session as a whole, and address issues of process lifecycle (when do desktop components start and stop running). Communication between the desktop session and the operating system, where the operating system would typically include the kernel and any system daemons or processes. For the within-desktop-session use case, the GNOME and KDE desktops have significant previous experience with different IPC solutions such as CORBA and DCOP. D-BUS is built on that experience and carefully tailored to meet the needs of these desktop projects in particular. The problem solved by the systemwide or communication-with-the-OS case is explained well by the following text from the Linux Hotplug project:
A gap in current Linux support is that policies with any sort of dynamic "interact with user" component aren't currently supported. For example, that's often needed the first time a network adapter or printer is connected, and to determine appropriate places to mount disk drives. It would seem that such actions could be supported for any case where a responsible human can be identified: single user workstations, or any system which is remotely administered. This is a classic "remote sysadmin" problem, where in this case hotplugging needs to deliver an event from one security domain (operating system kernel, in this case) to another (desktop for logged-in user, or remote sysadmin). Any effective response must go the other way: the remote domain taking some action that lets the kernel expose the desired device capabilities. (The action can often be taken asynchronously, for example letting new hardware be idle until a meeting finishes.) At this writing, Linux doesn't have widely adopted solutions to such problems. However, the new D-Bus work may begin to solve that problem.
D-BUS may happen to be useful for purposes other than the one it was designed for. Its general properties that distinguish it from other forms of IPC are: Binary protocol designed to be used asynchronously (similar in spirit to the X Window System protocol). Stateful, reliable connections held open over time. The message bus is a daemon, not a "swarm" or distributed architecture. Many implementation and deployment issues are specified rather than left ambiguous. Semantics are similar to the existing DCOP system, allowing KDE to adopt it more easily. Security features to support the systemwide mode of the message bus.
Concepts Some basic concepts apply no matter what application framework you're using to write a D-BUS application. The exact code you write will be different for GLib vs. Qt vs. Python applications, however. Objects and Object Paths Each application using D-BUS contains objects, which generally map to GObject, QObject, C++ objects, or Python objects (but need not). An object is an instance rather than a type. When messages are received over a D-BUS connection, they are sent to a specific object, not to the application as a whole. To allow messages to specify their destination object, there has to be a way to refer to an object. In your favorite programming language, this is normally called a pointer or reference. However, these references are implemented as memory addresses relative to the address space of your application, and thus can't be passed from one application to another. To solve this, D-BUS introduces a name for each object. The name looks like a filesystem path, for example an object could be named /org/kde/kspread/sheets/3/cells/4/5. Human-readable paths are nice, but you are free to create an object named /com/mycompany/c5yo817y0c1y1c5b if it makes sense for your application. Namespacing object paths is smart, by starting them with the components of a domain name you own (e.g. /org/kde). This keeps different code modules in the same process from stepping on one another's toes. Interfaces Each object supports one or more interfaces. Think of an interface as a named group of methods and signals, just as it is in GLib or Qt or Java. Interfaces define the type of an object instance. Message Types Messages are not all the same; in particular, D-BUS has 4 built-in message types: Method call messages ask to invoke a method on an object. Method return messages return the results of invoking a method. Error messages return an exception caused by invoking a method. Signal messages are notifications that a given signal has been emitted (that an event has occurred). You could also think of these as "event" messages. A method call maps very simply to messages, then: you send a method call message, and receive either a method return message or an error message in reply. Services Object paths, interfaces, and messages exist on the level of libdbus and the D-BUS protocol; they are used even in the 1-to-1 case with no message bus involved. Services, on the other hand, are a property of the message bus daemon. A service is simply a name mapped to some application connected to the message bus daemon. These names are used to specify the origin and destination of messages passing through the message bus. When a name is mapped to a particular application, the application is said to own that service. On connecting to the bus daemon, each application immediately owns a special name called the base service. A base service begins with a ':' (colon) character; no other services are allowed to begin with that character. Base services are special because each one is unique. They are created dynamically, and are never re-used during the lifetime of the same bus daemon. You know that a given base service name will have the same owner at all times. An example of a base service name might be :34-907. Applications may ask to own additional well-known services. For example, you could write a specification to define a service called com.mycompany.TextEditor. Your definition could specify that to own this service, an application should have an object at the path /com/mycompany/TextFileManager supporting the interface org.freedesktop.FileHandler. Applications could then send messages to this service, object, and interface to execute method calls. You could think of the base service names as IP addresses, and the well-known services as domain names. So com.mycompany.TextEditor might map to something like :34-907 just as mycompany.com maps to something like 192.168.0.5. Services have a second important use, other than routing messages. They are used to track lifecycle. When an application exits (or crashes), its connection to the message bus will be closed by the operating system kernel. The message bus then sends out notification messages telling remaining applications that the application's services have lost their owner. By tracking these notifications, your application can reliably monitor the lifetime of other applications. Addresses Applications using D-BUS are either servers or clients. A server listens for incoming connections; a client connects to a server. Once the connection is established, it is a symmetric flow of messages; the client-server distinction only matters when setting up the connection. A D-BUS address specifies where a server will listen, and where a client will connect. For example, the address unix:path=/tmp/abcdef specifies that the server will listen on a UNIX domain socket at the path /tmp/abcdef and the client will connect to that socket. An address can also specify TCP/IP sockets, or any other transport defined in future iterations of the D-BUS specification. When using D-BUS with a message bus, the bus daemon is a server and all other applications are clients of the bus daemon. libdbus automatically discovers the address of the per-session bus daemon by reading an environment variable. It discovers the systemwide bus daemon by checking a well-known UNIX domain socket path (though you can override this address with an environment variable). If you're using D-BUS without a bus daemon, it's up to you to define which application will be the server and which will be the client, and specify a mechanism for them to agree on the server's address. Big Conceptual Picture Pulling all these concepts together, to specify a particular method call on a particular object instance, a number of nested components have to be named: Address -> [Service] -> Path -> Interface -> Method The service is in brackets to indicate that it's optional -- you only provide a service name to route the method call to the right application when using the bus daemon. If you have a direct connection to another application, services aren't used; there's no bus daemon. The interface is also optional, primarily for historical reasons; DCOP does not require specifying the interface, instead simply forbidding duplicate method names on the same object instance. D-BUS will thus let you omit the interface, but if your method name is ambiguous it is undefined which method will be invoked. GLib API: Using Remote Objects The GLib binding is defined in the header file <dbus/dbus-glib.h>. The API is very small, in sharp contrast to the low-level <dbus/dbus.h>. The GLib bindings are incomplete, see the TODO file and comments in the source code. Here is a D-BUS program using the GLib bindings. int main (int argc, char **argv) { DBusGConnection *connection; GError *error; DBusGProxy *proxy; DBusGPendingCall *call; char **service_list; int service_list_len; int i; g_type_init (); error = NULL; connection = dbus_g_bus_get (DBUS_BUS_SESSION, &error); if (connection == NULL) { g_printerr ("Failed to open connection to bus: %s\n", error->message); g_error_free (error); exit (1); } /* Create a proxy object for the "bus driver" (service org.freedesktop.DBus) */ proxy = dbus_g_proxy_new_for_service (connection, DBUS_SERVICE_ORG_FREEDESKTOP_DBUS, DBUS_PATH_ORG_FREEDESKTOP_DBUS, DBUS_INTERFACE_ORG_FREEDESKTOP_DBUS); /* Call ListServices method */ call = dbus_g_proxy_begin_call (proxy, "ListServices", DBUS_TYPE_INVALID); error = NULL; if (!dbus_g_proxy_end_call (proxy, call, &error, DBUS_TYPE_ARRAY, DBUS_TYPE_STRING, &service_list, &service_list_len, DBUS_TYPE_INVALID)) { g_printerr ("Failed to complete ListServices call: %s\n", error->message); g_error_free (error); exit (1); } /* Print the results */ g_print ("Services on the message bus:\n"); i = 0; while (i < service_list_len) { g_assert (service_list[i] != NULL); g_print (" %s\n", service_list[i]); ++i; } g_assert (service_list[i] == NULL); g_strfreev (service_list); return 0; } DBusGProxy represents a remote object. dbus_g_proxy_begin_call() sends a method call to the remote object, and dbus_g_proxy_end_call() retrieves any return values or exceptions resulting from the method call. There are also DBusGProxy functions to connect and disconnect signals, not shown in the code example. dbus_g_bus_get() assumes that the application will use GMainLoop. The created connection will be associated with the main loop such that messages will be sent and received when the main loop runs. However, in the above code example the main loop never runs; D-BUS will not run the loop implicitly. Instead, dbus_g_proxy_end_call() will block until the method call has been sent and the reply received. A more complex GUI application might run the main loop while waiting for the method call reply. (DBusGPendingCall is currently missing the "notify me when the call is complete" functionality found in DBusPendingCall, but it should be added.) Future plans (see doc/TODO) are to use G_TYPE_STRING in place of DBUS_TYPE_STRING and so forth. In fact the above code is slightly incorrect at the moment, since it uses g_strfreev() to free a string array that was not allocated with g_malloc(). dbus_free_string_array() should really be used. However, once the GLib bindings are complete the returned data from dbus_g_proxy_end_call() will be allocated with g_malloc(). GLib API: Implementing Objects The GLib binding is defined in the header file <dbus/dbus-glib.h>. To implement an object, it's also necessary to use the dbus-glib-tool command line tool. The GLib bindings are incomplete. Implementing an object is not yet possible, see the TODO file and comments in the source code for details on what work needs doing. Qt API: Using Remote Objects The Qt bindings are not yet documented. Qt API: Implementing Objects The Qt bindings are not yet documented. Python API: Using Remote Objects The Python bindings are not yet documented, but the bindings themselves are in good shape. Python API: Implementing Objects The Python bindings are not yet documented, but the bindings themselves are in good shape.