libvirt internals

This section provides documents useful to those working on the libvirt internals, adding new public APIs, new hypervisor drivers or extending the libvirtd daemon code.

Introduction to basic rules and guidelines for hacking on libvirt code
Guide to adding public APIs
Approach for spawning commands from libvirt driver code
The libvirt RPC infrastructure
The Resource Lock Manager

Before adding new code it will be important to get a basic understanding of the many elements involved with making any call or change to the libvirt code. The architecture goals must be adhered to when submitting new code. Understanding the many places that need to be touched and the interactions between various subsystems within libvirt will directly correlate to the ability to be successful in getting new code accepted.

The following diagram depicts code flow from a client application, in this case the libvirt provided virsh command through the various layers to elicit a response from some chosen hypervisor.

virConnectOpen calling sequence

"virsh -c qemu:///system list --all"
After the virsh code processes the input arguments, it eventually will make a call to open the connection using a default set of authentication credentials (virConnectAuthDefault).
virConnectOpenAuth()
Each of the virConnectOpen APIs will first call virInitialize() and then revector through the local "do_open():" call.
- virInitialize()
  Calls the registration API for each of the drivers with client-side only capabilities and then call the remoteRegister() API last. This ensures the virDriverTab[] tries local drivers first before using the remote driver.
- Loop through virDriverTab[] entries trying to call their respective "open" entry point (in our case remoteOpen())
- After successful return from the virDriverTab[] open() API, attempt to find and open other drivers (network, interface, storage, etc.)
remoteOpen()
After a couple of URI checks, a call to doRemoteOpen() is made
- Determine network transport and host/port to use from URI
  The transport will be either tls, unix, ssh, libssh2, ext, or tcp with the default of tls. Decode the host/port if provided or default to "localhost".
- virNetClientRegisterAsyncIO()
  Register an I/O callback mechanism to get returned data via virNetClientIncomingEvent()
- "call(...REMOTE_PROC_OPEN...)"
  Eventually routes into virNetClientProgramCall() which will call virNetClientSendWithReply() and eventually uses virNetClientIO()to send the message to libvirtd and then waits for a response using virNetClientIOEventLoop()
- virNetClientIncomingEvent()
  Receives the returned packet and processes through virNetClientIOUpdateCallback()
libvirtd Daemon
- Daemon Startup
  The daemon initialization processing will declare itself as a server via a virNetServerNew() call, then use virDriverLoadModule() to find/load all known drivers, set up an RPC server program using the remoteProcs[] table via a virNetServerProgramNew() call. The table is the corollary to the remote_procedure enum list in the client. It lists all the functions to be called in the same order. Once RPC is set up, networking server sockets are opened, the various driver state initialization routines are run from the virStateDriverTab[], the network links are enabled, and the daemon waits for work.
- RPC
  When a message is received, the remoteProcs[] table is referenced for the 'REMOTE_PROC_OPEN' call entry. This results in remoteDispatchOpen() being called via the virNetServerProgramDispatchCall().
- remoteDispatchOpen()
  The API will read the argument passed picking out the name of the driver to be opened. The code will then call virConnectOpen() or virConnectOpenReadOnly() depending on the argument flags.
- virConnectOpen() or virConnectOpenReadOnly()
  Just like the client except that upon entry the URI is what was passed from the client and will be found and opened to process the data.
  
  The returned structure data is returned via the virNetServer interfaces to the remote driver which then returns it to the client application.