It's needed by Sailfish OS wifi plugin to determine which networks
are pre-configured and have to be reported to the connman core
immediately after they've been seen in the air.

This also removes GObject from the interface and gets rid of glib
deprecation warning about G_ADD_PRIVATE.

[connman] Fix STORAGEDIR uses. Fixes JB#49651

See merge request mer-core/connman!271

When user dir is in use return that to provide correct path for configs.
Change the function to return const char* as there is no need to
allocate a new variable at each call.

The new function allows to get the storage dir for the service using the
identifier. This is required in multiuser environment, as some services
are saved to STORAGEDIR and some are to USER_STORAGEDIR. This function
makes it easier and cleaner to use. Internally, the existing
storagedir_for() is used. If the service identifier is invalid NULL is
returned as storagedir (TODO: should STORAGEDIR be the default instead
of NULL?).

Link with gio-2.0 and gobject-2.0 when necessary

See merge request mer-core/connman!274

Those were being pulled in as a depedency of libgsupplicant which
recently switched to Requires.private dependency in its pkgconfig
to avoid overlinking.

Fix glib assertion 'datetime != NULL'

See merge request mer-core/connman!272

[connman] Fix WLAN issues in user change and device bootup. Fixes JB#49436

See merge request mer-core/connman!266

Suppose there is Wi-Fi network X that is saved in connman. Consider
the following scenario:

1. Autoscan is started by connman.
2. Network X will be marked as unavailable as result of
connman_device_set_scanning(true).
3. Application issue connect to network X.
4. Connect to X starts and
interface_state(G_SUPPLICANT_STATE_AUTHENTICATING) is received.
5. As result of this stop_autoscan >
connman_device_set_scanning(false) > __connman_device_cleanup_networks
is issued that free_network X because it was marked as unavailable.
6. It leads to network_remove > __connman_service_remove_from_network
> connman_service_unref_debug > __connman_service_disconnect > ...
reply_pending(ECONNABORTED)

The powered_pending status is an indication that the powered status is
in the process of being changed; the correct return code here should be
"in progress".

Without this patch, setting a technology "Powered" for a slow ofono
device in the process of powering up results in an immediate DBus
response even though the device is far from being ready.

With this patch, setting a technology "Powered" returns first when the
property has actually changed which is the desired behaviour.

Add stub for __connman_technology_disable_all() to systemd login unit
test.

When user login is changed via systemd login monitor in some occasions
root user, the default system user, may have technologies enabled which
may be confusing to the user and system components when the user is
opening up session. This change ensures that prior to user change
happening after the session is active the technologies are disabled.

This allows some technolgies, such as rfkill enabled WiFi, to be put
down in time before it is enabled again by the user configuration. This
can be the case in both boot and user change. The reason is that when
user is logging out root user is enabled, and the technologies enabled
on the root user are thus enabled as well.

The return values were incorrect because of incorrect use added to
technology_affect_devices(). This was corrected in one previous commits
and these values needed to be changed as well.

Also fix __connman_technology_disable_all() return value with
nonexistent settings file.

Add use of unload() and load() callbacks to be able to properly notify
about removed and new services/providers. Removal and adding of the
services can also be manual and in this case the information should be
also forwarded.

When a service is removed it is only required to be unloaded since one
cannot remove what is already deleted. Only the removed service is given
as the sole item in the service list for unload callback. Loading
callbacks should behave in a way that if the identifier is already
present just ignore the service, thus load() should be possible to call
multiple times.

This does not interfere with service/provider removal via connmand/vpnd,
since the removal in service.c and vpn-provider.c also calls
__connman_storage_remove_service() which removes the actual files. After
that the inotify indication about removed ones is received and in then
the subdirs are only cleared for the entry, unload() callbacks should
ignore a missing service. In loading the detection of new services does
another initializtion call but this cannot be avoided since the
functionality for load() is to get all available services.

Move load and post action callback use at the end of the
change_storage_dir(). No need to have them both in the switch.

Disable all technologies even when keyfile is missing. Even though
missing keyfile usually means that there shouldn't be any technologies
enabled it can be that the keyfile has been already manually removed or
deleted prior to calling the __connman_technology_disable_all(). Utilize
the return value as a indication of failing to disable a tech instead.

Implement technology_changed_state() to make it easier to indicate a
state change on a technology regardless of the rfkill use. This takes
the functionality from __connman_technology_{enabled,disabled}().

Main reason is to use this when changing the user to get all proper
indications sent via D-Bus. Also when using delayed enabling this is to
be used to make the code less obfuscated.

Use delayed enabling if the tech to be enabled is rfkill enabled when
re-enabling in multiuser environment or when the enabling reports being
busy. A technology can be busy because of pending reply, that is being
handled by the timeout function, or because of other component / lower
layer. The delay is set to 300ms and timeout function attempts to enable
until enabled (0 or EALREADY is returned).

For the need to reply to a pending power change request a generic error
handler is implemented. This allows to send a reply to the pending power
request with proper error code and reduce duplicate code. Sending
ETIMEDOUT error in other error cases is not proper way to do it, better
to send ECANCELED when the action is canceled.

Also, use EINTR when canceling scans. This seems to be the correct error
code that is to be passed when canceling request. Do the canceling of
the scan before sending pending reply to power change call in the tech
disabling when user is changed - as it is done in other places in
technology.c.

Don't schedule a scan when hidden network reappears

See merge request mer-core/connman!269

Bump build requirements

See merge request mer-core/connman!268

If a hidden network has recovered (disappeared and then re-appeared)
there's no need to schedule a scan, we already know what it is.

glib 2.62 is required by g_date_time_format_iso8601()
gsupplicant 1.0.17 by gsupplicant_interface_add_eap_status_handler()

Immediately notify about service change in service unload

See merge request mer-core/connman!267

Unloading of the services is done as the last step before the user paths
are changed when changing user. After unloading the services the changes
must be immediatelly sent (signaled).

This is required because if there are identical services used (e.g. two
different users use the same WiFi network) then the delayed change signal
can contain same identifier in both changed and removed lists. This would
otherwise cause confusion in other components relying on these changed
signals.

Improve VPN management in multiuser environment.

See merge request mer-core/connman!265

Add tests for the binary user override improvements to test cases where
override is not in effect because of VPN user being in the system users.

Updated test to clean out the test dir properly.

Add mock for getpwuid().

[vpn-settings] Improve binary username and override checks. JB#49504

Support both uid as string and username in both system binary user and
users. Add checking of both types, uid as string and username in setting
and getting the override.

Change the finalize function (__vpn_settings_set_binary_user_override())
according to storage.c changes.

[vpn] Use get_connections() as connmand finalize func. JB#49504

Implemented vpn_finalize() function to register as callback to storage.c
to be ran as the last function in the user change process. This function
simply calls the get_connections() as if connmand notices vpnd being up.

The purpose of this is to get the users' VPN services from the vpnd when
connmand has dealt its part in the user change process. Since the
message to vpnd is sent first, vpnd should have loaded all VPNs when
the request sent by get_connections() arrives.

[storage] Use uid and user data in finalize. JB#49504

Implement connman_storage_update_finalize_cb() to allow setting of the
finalize callback to user change after initialization. This function is
called last in the process. Giving NULL as cb will reset the finalize
callback.

It is useful for the vpn plugin in connmand to set a callback for
requesting the VPN connections from the vpnd. The vpnd could send the
signals after the VPNs are loaded from keyfiles but it will come too
early to connmand. It may be that the D-Bus signals are processed before
the user change is completed. Allowing to register a callback to be
executed and to explicitly request the VPNs from vpnd is a better, more
synchronized solution for connmand.

In addition, removed the use of username in the finalize callback and in
storage.c alltogether. Use of uid_t is sufficient.

[storage] Get correct storagedir for VPN config. JB#49504

Use storagedir_for() instead of direct usage of VPN_STORAGEDIR. In
multiuser environment the VPN storage differs from the system
default/configured dir.

[service] Unload VPN services and always reply from GetProperties. JB#49504

Unload also the VPN services when changing user in multiuser
environment. This is to cover situations when the user change from vpnd
hasn't yet signaled about removed VPNs.

Always send a reply from GetProperties. This may not be reached but
returning NULL and leaving the pending call awaiting will make D-Bus
daemon unhappy as the default maximum amount of pending call replies is
set to 128.

Implement systemd-logind listener for monitoring user change

See merge request mer-core/connman!262