Added various methods that allow to make methods to delay to return a
value, both by using timing functions and using a way to manually
stop and restart the calls.
This is mostly done using async callbacks in dbus methods
We need to be able to hack this to be an async daemon to perform some
multi-thread tests, so replacing default implementation with a simple
one that for now just does same of default
GDBus generated interface skeletons support natively an authorization
method that allows to filter calls before we get into the method
callback and that gets called into the call thread, before we go back
to main thread.
As per this, we can move all the polkit and other authorization checks
into this callback so that method handlers are now just assuming they're
the right to perform the requested operation.
As per the fact we'll share some data between another thread and the
callbacks, we will need to introduce some locking mechanism to ensure
safe data access.
This might be reduced by moving the claiming checks back to the method,
but would lead errors to be handled in different ordering, and so the
user to be requested for a password, and then - in case fail.
This can still happen now, but only if there are concurrent requests.
We now can get an invocation-owned sender at any moment with GDBus, so
there's no point of getting it as optional return-out value from the
username check function.
This is not used right now in all its full possibilities, but will make
devices hotplug support easier to implement and handle at client-side
level.
As per this we can stop doing the manual tracking of the devices.
Fprintd is dependent on the deprecated dbus-glib, also this doesn't provide
various features we can take advantage of, like the ones for async
authentication mechanism.
So, remove all the dbus-glib dependencies and simplify the code, but without
any further refactor, and keeping everything as it used to work, while this
will give room for further improvements in subsequent commits.
Internally, we just use dbus-codegen to generate the skeletons, and we
use the generated FprintdDBusManager with composition, while we
implement the device skeleton interface in FprintDevice, so that we
don't have to use it as a proxy, and keep being closer to what it used
to be with dbus-glib.
Fixes: #61
The current lockdown rules prevent USB devices from being accessed and
cause threading to not work.
As such, revert them until it is clear on how/if we can apply these
measures. It is primarily not clear on how to prevent fork/clone as
fprintd does not need those.
This reverts commit 2fd86624e5.
See: #82
Use error messages to be consistent, and avoid checking for a returned
value when dbus-glib function to fetch it returned false, as it's
implicit that we had a failure.
Otherwise if didn't fail we are sure that we got the requested value.
fprintd supports "any" finger parameter for the VerifyStart call, and it's
up to the daemon to pick the first known if the device doesn't support
identification.
So remove the check to verify utility and add a test to verify this is
respected.
This avoids addCleanup ordering errors and also errors when we may try to
print an invalid stdout pipe (like when we have processed it all), as python
might fail with something like:
======================================================================
ERROR: test_fprintd_multiple_verify_fails (__main__.TestFprintdUtilsVerify)
----------------------------------------------------------------------
Traceback (most recent call last):
File "~/GNOME/fprintd/tests/test_fprintd_utils.py", line 102, in <lambda>
self.addCleanup(lambda: print(process.stdout.read()))
File "/usr/lib/python3.8/codecs.py", line 321, in decode
data = self.buffer + input
TypeError: can't concat NoneType to bytes
unittest addCleanup calls are called in reverse order, so we need to reverse
the order of the calls as well, otherwise we won't correctly terminate the
subprocess children
No need to repeat the action in every unit test, but move the tests to a
different class to easily allow adding another class with tests with no
such initialization
An assertion that is raised within a callback will not be swallowed by
the C code that called the function. To ensure that errors will be
noticable, pass the result back to the surrounding scope and check it
there.
This excercises the path where we early-report a result and the
VerifyStop call must wait for the operation to complete intenernally.
Note that this test cannot fail right now due to the FpImageDevice
internal code still trying to hide the deactivation delay internally.
See libfprint!174
libfprint 1.90.1 supports early match result report which allows to inform
the clients as soon as the driver detected a match/no-match without having
to wait the whole device driver deactivation.
Use this feature in fprintd, by emitting the "VerifyStatus" signal as soon
as we've an usable result we can inform the client about.
As per this, ignore any possible error we may get afterwards, due to device
issues or late user cancellation, as the point of the verification action
(i.e. getting a match/no-match) can be considered accomplished.
Fixes https://gitlab.freedesktop.org/libfprint/fprintd/issues/35
If a client vanishes while we are opening or closing then fprintd would
uncoditionally try to close the device immediately. This should not be
done, it instead needs to wait for the open/close to complete.
Add open/close to the current action enum and keep track of the fact
that they are ongoing. Also check, whether the device has been closed in
the meantime (after the nested mainloop is done).
Fixes: #65
The data->result was free'ed both in the loop (before breaking) and
afterwards. As the first case did not set the pointer to NULL, this
could result in a double free.
Fix this by simply removing the free that is in the loop and relying on
the cleanup later on.
We were saving the state of a disconnected device because we used a
workaround to figure it out, but now libfprint would provide us a proper
GError in such case, and so we'd handle it without the need of saving the
state, given it's never used anyways.
