fwupd/docs/tutorial.xml
Richard Hughes e031774a8b Split out the fwupd tutorial into a new file
No content changes.
2020-07-28 09:22:07 +01:00

403 lines
15 KiB
XML

<?xml version="1.0"?>
<!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.3//EN"
"http://www.oasis-open.org/docbook/xml/4.3/docbookx.dtd">
<reference id="tutorial">
<title>Plugin Tutorial</title>
<partintro>
<section>
<title>Introduction</title>
<para>
At the heart of fwupd is a plugin loader that gets run at startup,
when devices get hotplugged and when updates are done.
The idea is we have lots of small plugins that each do one thing, and
are ordered by dependencies against each other at runtime.
Using plugins we can add support for new hardware or new policies
without making big changes all over the source tree.
</para>
<para>
There are broadly 3 types of plugin methods:
</para>
<itemizedlist>
<listitem>
<para>
<emphasis role="strong">Mechanism</emphasis>: Upload binary data
into a specific hardware device.
</para>
</listitem>
<listitem>
<para>
<emphasis role="strong">Policy</emphasis>: Control the system when
updates are happening, e.g. preventing the user from powering-off.
</para>
</listitem>
<listitem>
<para>
<emphasis role="strong">Helpers</emphasis>: Providing more
metadata about devices, for instance handling device quirks.
</para>
</listitem>
</itemizedlist>
<para>
In general, building things out-of-tree isn't something that we think is
a very good idea; the API and ABI <emphasis>internal</emphasis> to fwupd is still
changing and there's a huge benefit to getting plugins upstream where
they can undergo review and be ported as the API adapts.
For this reason we don't install the plugin headers onto the system,
although you can of course just install the <code>.so</code> binary file
manually.
</para>
<para>
A plugin only needs to define the vfuncs that are required, and the
plugin name is taken automatically from the suffix of the
<filename>.so</filename> file.
</para>
<example>
<title>A sample plugin</title>
<programlisting>
/*
* Copyright (C) 2017 Richard Hughes
*/
#include &lt;fu-plugin.h&gt;
#include &lt;fu-plugin-vfuncs.h&gt;
struct FuPluginData {
gpointer proxy;
};
void
fu_plugin_initialize (FuPlugin *plugin)
{
fu_plugin_add_rule (plugin, FU_PLUGIN_RULE_RUN_BEFORE, "dfu");
fu_plugin_alloc_data (plugin, sizeof (FuPluginData));
}
void
fu_plugin_destroy (FuPlugin *plugin)
{
FuPluginData *data = fu_plugin_get_data (plugin);
destroy_proxy (data->proxy);
}
gboolean
fu_plugin_startup (FuPlugin *plugin, GError **error)
{
FuPluginData *data = fu_plugin_get_data (plugin);
data->proxy = create_proxy ();
if (data->proxy == NULL) {
g_set_error (error, FWUPD_ERROR, FWUPD_ERROR_NOT_SUPPORTED,
"failed to create proxy");
return FALSE;
}
return TRUE;
}
</programlisting>
</example>
<para>
We have to define when our plugin is run in reference to other plugins,
in this case, making sure we run before the <code>dfu</code> plugin.
For most plugins it does not matter in what order they are run and
this information is not required.
</para>
</section>
<section>
<title>Creating an abstract device</title>
<para>
This section shows how you would create a device which is exported
to the daemon and thus can be queried and updated by the client software.
The example here is all hardcoded, and a true plugin would have to
derive the details about the <code>FuDevice</code> from the hardware,
for example reading data from <code>sysfs</code> or <code>/dev</code>.
</para>
<example>
<title>Example adding a custom device</title>
<programlisting>
#include &lt;fu-plugin.h&gt;
gboolean
fu_plugin_coldplug (FuPlugin *plugin, GError **error)
{
g_autoptr(FuDevice) dev = NULL;
fu_device_set_id (dev, "dummy-1:2:3");
fu_device_add_guid (dev, "2d47f29b-83a2-4f31-a2e8-63474f4d4c2e");
fu_device_set_version (dev, "1.2.3");
fu_device_get_version_lowest (dev, "1.2.2");
fu_device_get_version_bootloader (dev, "0.1.2");
fu_device_add_icon (dev, "computer");
fu_device_add_flag (dev, FWUPD_DEVICE_FLAG_UPDATABLE);
fu_plugin_device_add (plugin, dev);
return TRUE;
}
</programlisting>
</example>
<para>
This shows a lot of the plugin architecture in action. Some notable points:
</para>
<itemizedlist>
<listitem>
<para>
The device ID (<code>dummy-1:2:3</code>) has to be unique on the
system between all plugins, so including the plugin name as a
prefix is probably a good idea.
</para>
</listitem>
<listitem>
<para>
The GUID value can be generated automatically using
<code>fu_device_add_guid(dev,"some-identifier")</code> but is quoted
here explicitly.
The GUID value has to match the <code>provides</code> value in the
<code>.metainfo.xml</code> file for the firmware update to succeed.
</para>
</listitem>
<listitem>
<para>
Setting a display name and an icon is a good idea in case the
GUI software needs to display the device to the user.
Icons can be specified using a full path, although icon theme names
should be preferred for most devices.
</para>
</listitem>
<listitem>
<para>
The <code>FWUPD_DEVICE_FLAG_UPDATABLE</code> flag tells the client
code that the device is in a state where it can be updated.
If the device needs to be in a special mode (e.g. a bootloader) then
the <code>FWUPD_DEVICE_FLAG_NEEDS_BOOTLOADER</code> flag can also be
used.
If the update should only be allowed when there is AC power available
to the computer (i.e. not on battery) then
<code>FWUPD_DEVICE_FLAG_REQUIRE_AC</code> should be used as well.
There are other flags and the API documentation should be used when
choosing what flags to use for each kind of device.
</para>
</listitem>
<listitem>
<para>
Setting the lowest allows client software to refuse downgrading
the device to specific versions.
This is required in case the upgrade migrates some kind of data-store
so as to be incompatible with previous versions.
Similarly, setting the version of the bootloader (if known) allows
the firmware to depend on a specific bootloader version, for instance
allowing signed firmware to only be installable on hardware with
a bootloader new enough to deploy it
</para>
</listitem>
</itemizedlist>
</section>
<section>
<title>Mechanism Plugins</title>
<para>
Although it would be a wonderful world if we could update all hardware
using a standard shared protocol this is not the universe we live in.
Using a mechanism like DFU or UpdateCapsule means that fwupd will just
work without requiring any special code, but for the real world we need
to support vendor-specific update protocols with layers of backwards
compatibility.
</para>
<para>
When a plugin has created a device that is <code>FWUPD_DEVICE_FLAG_UPDATABLE</code>
we can ask the daemon to update the device with a suitable
<code>.cab</code> file.
When this is done the daemon checks the update for compatibility with
the device, and then calls the vfuncs to update the device.
</para>
<example>
<title>Updating a device</title>
<programlisting>
gboolean
fu_plugin_update (FuPlugin *plugin,
FuDevice *dev,
GBytes *blob_fw,
FwupdInstallFlags flags,
GError **error)
{
gsize sz = 0;
guint8 *buf = g_bytes_get_data (blob_fw, &amp;sz);
/* write 'buf' of size 'sz' to the hardware */
return TRUE;
}
</programlisting>
</example>
<para>
It's important to note that the <code>blob_fw</code> is the binary
firmware file (e.g. <code>.dfu</code>) and <emphasis role="strong">not</emphasis>
the <code>.cab</code> binary data.
</para>
<para>
If <code>FWUPD_INSTALL_FLAG_FORCE</code> is used then the usual checks
done by the flashing process can be relaxed (e.g. checking for quirks),
but please don't brick the users hardware even if they ask you to.
</para>
</section>
<section>
<title>Policy Helpers</title>
<para>
For some hardware, we might want to do an action before or after
the actual firmware is squirted into the device.
This could be something as simple as checking the system battery
level is over a certain threshold, or it could be as complicated as
ensuring a vendor-specific GPIO is asserted when specific types
of hardware are updated.
</para>
<example>
<title>Running before a device update</title>
<programlisting>
gboolean
fu_plugin_update_prepare (FuPlugin *plugin, FuDevice *device, GError **error)
{
if (fu_device_has_flag (device, FWUPD_DEVICE_FLAG_REQUIRE_AC &amp;&amp; !on_ac_power ()) {
g_set_error_literal (error,
FWUPD_ERROR,
FWUPD_ERROR_AC_POWER_REQUIRED,
"Cannot install update "
"when not on AC power");
return FALSE;
}
return TRUE;
}
</programlisting>
</example>
<example>
<title>Running after a device update</title>
<programlisting>
gboolean
fu_plugin_update_cleanup (FuPlugin *plugin, FuDevice *device, GError **error)
{
return g_file_set_contents ("/var/lib/fwupd/something",
fu_device_get_id (device), -1, error);
}
</programlisting>
</example>
</section>
<section>
<title>Detaching to bootloader mode</title>
<para>
Some hardware can only be updated in a special bootloader mode, which
for most devices can be switched to automatically.
In some cases the user to do something manually, for instance
re-inserting the hardware with a secret button pressed.
</para>
<para>
Before the device update is performed the fwupd daemon runs an optional
<code>update_detach()</code> vfunc which switches the device to
bootloader mode.
After the update (or if the update fails) an the daemon runs an
optional <code>update_attach()</code> vfunc which should switch the
hardware back to runtime mode.
Finally an optional <code>update_reload()</code> vfunc is run to
get the new firmware version from the hardware.
</para>
<para>
The optional vfuncs are <emphasis role="strong">only</emphasis> run
on the plugin currently registered to handle the device ID, although
the registered plugin can change during the attach and detach phases.
</para>
<example>
<title>Running before a device update</title>
<programlisting>
gboolean
fu_plugin_update_detach (FuPlugin *plugin, FuDevice *device, GError **error)
{
if (hardware_in_bootloader)
return TRUE;
return _device_detach(device, error);
}
</programlisting>
</example>
<example>
<title>Running after a device update</title>
<programlisting>
gboolean
fu_plugin_update_attach (FuPlugin *plugin, FuDevice *device, GError **error)
{
if (!hardware_in_bootloader)
return TRUE;
return _device_attach(device, error);
}
</programlisting>
</example>
<example>
<title>Running after a device update on success</title>
<programlisting>
gboolean
fu_plugin_update_reload (FuPlugin *plugin, FuDevice *device, GError **error)
{
g_autofree gchar *version = _get_version(plugin, device, error);
if (version == NULL)
return FALSE;
fu_device_set_version(device, version);
return TRUE;
}
</programlisting>
</example>
</section>
<section>
<title>The Plugin Object Cache</title>
<para>
The fwupd daemon provides a per-plugin cache which allows objects
to be added, removed and queried using a specified key.
Objects added to the cache must be <code>GObject</code>s to enable the
cache objects to be properly refcounted.
</para>
</section>
<section>
<title>Debugging a Plugin</title>
<para>
If the fwupd daemon is started with <code>--plugin-verbose=$plugin</code>
then the environment variable <code>FWUPD_$PLUGIN_VERBOSE</code> is
set process-wide.
This allows plugins to detect when they should output detailed debugging
information that would normally be too verbose to keep in the journal.
For example, using <code>--plugin-verbose=logitech_hidpp</code> would set
<code>FWUPD_LOGITECH_HID_VERBOSE=1</code>.
</para>
</section>
<section>
<title>Using existing code to develop a plugin</title>
<para>
It is not usually possible to <em>share</em> a plugin codebase with
firmware update programs designed for other operating systems.
Matching the same rationale as the Linux kernel, trying to use one
code base between projects with a compatibility shim layer in-between
is real headache to maintain.
</para>
<para>
The general consensus is that trying to use a abstraction layer for
hardware is a very bad idea as you're not able to take advantage of the
platform specific helpers -- for instance quirk files and the custom
GType device creation.
The time the <em>vendor</em> saves by creating a shim layer and
importing existing source code into fwupd will be overtaken 100x by
<em>upstream</em> maintenance costs longer term, which isn't fair.
</para>
<para>
In a similar way, using C++ rather than GObject C means expanding the
test matrix to include clang in C++ mode and GNU g++ too.
It's also doubled the runtime requirements to now include both the C
standard library as well as the C++ standard library and increases the
dependency surface.
</para>
<para>
Most rewritten fwupd plugins at up to x10 smaller than the standalone
code as they can take advantage of helpers provided by fwupd rather
than re-implementing error handling, device quirking and data chunking.
</para>
</section>
</partintro>
</reference>