libvirt/docs/internals.html.in
Daniel P. Berrange b1c81567c7 docs: switch to using HTML5 doctype declaration
The HTML5 doctype is simply

  <!DOCTYPE html>

no DTD is present because HTML5 is no longer defined as an
extension of SGML.

XSL has no way to natively output a doctype without a public
or system identifier, so we have to use an <xsl:text> hack
instead.

See also

  https://dev.w3.org/html5/html-author/#doctype-declaration

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
2017-08-02 17:00:11 +01:00

120 lines
6.0 KiB
XML

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml">
<body>
<h1>libvirt internals</h1>
<p>
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.
</p>
<ul>
<li>Introduction to basic rules and guidelines for
<a href="hacking.html">hacking</a> on libvirt code</li>
<li>Guide to adding <a href="api_extension.html">public APIs</a></li>
<li>Insight into libvirt <a href="internals/eventloop.html">event loop and worker pool</a></li>
<li>Approach for <a href="internals/command.html">spawning commands</a>
from libvirt driver code</li>
<li>The libvirt <a href="internals/rpc.html">RPC infrastructure</a></li>
<li>The <a href="internals/locking.html">Resource Lock Manager</a></li>
</ul>
<p>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 <a href="goals.html">goals</a> 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.</p>
<p>The following diagram depicts code flow from a client application, in
this case the libvirt provided <code>virsh</code> command through the
various layers to elicit a response from some chosen hypervisor.
</p>
<p class="image">
<img alt="virConnectOpen calling sequence"
src="libvirt-virConnect-example.png"/>
</p>
<ul>
<li>"virsh -c qemu:///system list --all"
<p>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). </p></li>
<li>virConnectOpenAuth()
<p>Each of the virConnectOpen APIs will first call virInitialize() and
then revector through the local "do_open():" call.</p>
<ul>
<li>virInitialize()
<p>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.</p></li>
<li>Loop through virDriverTab[] entries trying to call their
respective "open" entry point (in our case remoteOpen())</li>
<li>After successful return from the virDriverTab[] open()
API, attempt to find and open other drivers (network, interface,
storage, etc.)</li>
</ul>
</li>
<li>remoteOpen()
<p>After a couple of URI checks, a call to doRemoteOpen() is made</p>
<ul>
<li>Determine network transport and host/port to use from URI
<p>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".</p></li>
<li>virNetClientRegisterAsyncIO()
<p>Register an I/O callback mechanism to get returned data via
virNetClientIncomingEvent()</p></li>
<li>"call(...REMOTE_PROC_OPEN...)"
<p>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()</p></li>
<li>virNetClientIncomingEvent()
<p>Receives the returned packet and processes through
virNetClientIOUpdateCallback()</p></li>
</ul>
</li>
<li>libvirtd Daemon
<p></p>
<ul>
<li>Daemon Startup
<p>The daemon initialization processing will declare itself
as a daemon via a virNetDaemonNew() call, then creates new server
using virNetServerNew() and adds that server to the main daemon
struct with virNetDaemonAddServer() call. It will then use
virDriverLoadModule() to find/load all known drivers,
set up an RPC server program using the <code>remoteProcs[]</code>
table via a virNetServerProgramNew() call. The table is the
corollary to the <code>remote_procedure</code> 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 <code>virStateDriverTab[]</code>, the network
links are enabled, and the daemon waits for work.</p></li>
<li>RPC
<p>When a message is received, the <code>remoteProcs[]</code>
table is referenced for the 'REMOTE_PROC_OPEN' call entry.
This results in remoteDispatchOpen() being called via the
virNetServerProgramDispatchCall().</p></li>
<li>remoteDispatchOpen()
<p>The API will read the argument passed picking out the
<code>name</code> of the driver to be opened. The code
will then call virConnectOpen() or virConnectOpenReadOnly()
depending on the argument <code>flags</code>.</p></li>
<li>virConnectOpen() or virConnectOpenReadOnly()
<p>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.</p>
<p>The returned structure data is returned via the
virNetServer interfaces to the remote driver which then
returns it to the client application.</p></li>
</ul>
</li>
</ul>
</body>
</html>