Disable rescue mode when/if init done is reached



See merge request !16

The once/bootup mechanism used for terminating rescue mode leaves it
permanently active if usb moded is restarted after bootup has finished.

Do not activate rescue mode if init-done has been reached when usb-moded
is starting up.

Disable rescue mode if init-done is signaled on D-Bus SystemBus while
usb-moded is running.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Cancel mode selection when charger is detected



See merge request !15

When charger is connected, usb-moded can see first "USB" (=cable)
connection event from udev followed by "USB_CDP" (=charger) event.
Due to cached something-connected flag usb moded still behaves as
if pc cable were connected - which leaves ui thinking mode selection
can/should be done etc.

If dedicated charger is reported after pc cable, clear the low level
something-connected flag before acting on charger connected info.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Add mtp_mode-android-ffs configuration



See merge request !13

Existing configuration files do not cover case where android kernel
is used to implement just mtp functionality using ffs based userspace
code - mtp_mode-android uses "mtp" instead of "ffs" and pc_suite-android
activates both mtp and obex support.

Add mtp_mode-android-ffs.ini that is identical with pc_suite-android.ini
except for the mode name which is "mtp" instead of "pc_suite".

Mode change and cable detection fixes

Also contains fixes for various issues spotted while debugging.

See merge request !12

Since enumeration timers are not left running anymore when enumeration
occurs / appsync is canceled, the increment at activate_sync() and
check at enumerate_usb() based tag code filtering is no longer needed.

Remove the tag variables and code related to them.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The boolean state data for tracking active/inactive applications makes
it difficult to separate applications that are not relevant to the mode
and as a result an attempt to stop also those irrelevant apps that were
not started is made when exiting the mode.

Switch from boolean to inactive|active|dontcare state tracking.

When starting appsync, mark apps relevant to the mode as inactive and
the rest as dontcare.

When stopping appsync, stop only those applications that have been
activated by usb-moded.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

When appsync is starting applications, pre-enum applications are started
before post-enum apps. However stopping applications is done in whatever
order they happen to be after configuration parsing - which could cause
problems if post apps have explicit dependencies on pre apps.

Do also application stopping in two phases: first post-enum apps followed
by pre-enum apps.

Also, apart from usb-moded startup/shutdown, attempt to skip stopping of
applications that have not been started.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The systemd_control_service() function returns zero on success. Due to
incorrect boolean check in appsync_stop() diagnostic failure messages are
emitted on success and failures are ignored silently.

Fix the condition so that failures are logged.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Enumeration timer is started every time activate_sync() is called and
is left running even if enumeration occurs / appsync_stop() is called.

Cache enumeration timer id and cancel it when it is no longer needed.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

If android usb is in enable=0 state, cable detection is persistently
broken as the existing workarounds in usb-moded are made when cable
disconnect is detected and that can't happen unless cable detect works.

Enable android usb on usb-moded startup.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Seems that just writing enable=0 functions=none leaves the android usb in
some sort of limbo where cable detection does not work until enable=1
write is also made.

Do enable=1 write after clearing functions on cable disconnect.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

When debugging mode setting issues, it is hard to tell when and what gets
written to android usb sysfs files.

When running usb-moded in debug verbosity, log sysfs paths that get written
to and the data that is written - and if possible also the value held by
the sysfs file before the write.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

If dbus_connection_send_with_reply_and_block() should set the
dbus error, it is never freed.

Make sure dbus_error_free() is called before returning from the function.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

If charging_timeout should already contain a valid timer id when the
timer is started, the old timer will be leaked.

Remove existing timer id before assigning new one.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The io watch source id is not cleared when it is going to be implicitly
removed - which can lead to unsuccessful removal attempt when/if
hwal_cleanup() function is called later on.

Clear watch_id when return value from the callback function is going
to cause it to be implicitly removed.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Various pointers are uninitialized / left to hold stale values after
releasing. Which could lead to hard to debug problems if trigger_stop() is
called multiple times / after unsuccessful trigger_init().

Zero initialize all pointers dealing with dynamic resources and clear them
when resources are released.

Also clear iowatch id when it is going to be implicitly removed due to the
return value from the callback so that removal is not attempted anymore
when/if trigger_stop() is called later on.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

When network_retry() timer callback gets triggered, the associated
timer id delayed_network is not invalidated and usb moded can try
to remove it later on - which will lead to complaints from glib
being emitted to journal.

Clear the timer id when the timer callback gets called.

Also make sure already existing timer id is not left active when
scheduling a timeout.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Having a predicate function with name xxx_changed() that returns a
non-zero value when nothing has changed is confusing.

Switch the return value meaning so that non-zero means there is a
difference and zero that there is not.

Also treat old value of NULL and new value of NULL as no change.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The function compares old value with ini file section name, which leads
to practically every new value being evaluated different from previous one.

Check old value vs new value instead.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The make_hidden_modes_string() function claims to return const string,
but that is true only as an exception to the rule. Since the callers of
the function do not release dynamic memory, this leads to leakage. Also
if no mode is hidden, request to unhide something least to that mode
getting hidden and hiding something already hidden returns NULL which
is rather unintuitive.

Fix make_hidden_modes_string() and make all callers of it to release
dynamically reserved memory / deal with possible null return value.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The get_hidden_modes() returns data allocated via g_key_file_get_string()
which should be released with g_free() and fallback uses strdup() which
should use free().

Use g_strdup() in the fallback case and release memory with g_free().
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

There is data obtained via g_key_file_get_string() that is released
with free().

Use g_free() instead.

Also use a helper callback function with expected argument types
when releasing a list of elements.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The buffer used for constructing sysfs path from interface name is rather
short, which can lead to false negatives due to truncation. Additionally
the actual names of the interfaces checked are not logged in case of
errors and the helper function can crash if fed null strings.

Make the check_interface() helper function tolerate null arguments
and enlarge the path expansion buffer to make truncation less likely.

In case get_interface() fails to find valid a interface name, make it
emit diagnostic message that identifies the actual names attempted.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Unless the code base is actively used with other than gcc compilers, use
of -pedantic just denies use of handy gcc extensions like use of "%m"
format parameter instead of non-thread-safe strerror().

Drop the '-pedantic' option and replace all strerror() calls with "%m".
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

When fallback to alternate mount point is made, the pointer to the
dynamically allocated primary mount point is leaked.

Free string obtained via find_mounts() before assigning the pointer to
string returned by find_alt_mount().
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Probably in attempt to hide compilation warnings due to storing dynamically
allocated data in const pointers, there are multiple places where releasing
of dynamic memory uses casts. Some of these casts are incorrect and there
are still compilation issues.

Remove const attribute from pointers that are used to store dynamically
allocated data and remove the unnecessary casts at free() / g_free().
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

The prototypes for these functions claim to return const string, while
in reality it is dynamic memory that should released via g_free().

Change the prototypes to match the reality.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Signed-off-by: Matti Kosola <matti.kosola@jolla.com>

Get device manufacturer and product name from ssu



See merge request !11

Maintaining usb manufacturer and product name configurations for multiple
devices is cumbersome and time consuming. It also duplicates effort that
is needed for ssu configuration anyway.

When possible, override usb-moded manufacturer and product name
configuration by using data provided by ssu D-Bus service.

The feature is optional and needs to be enabled at usb-moded build
time via "--enable-mer-ssu" configure option.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Signed-off-by: Matti Kosola <matti.kosola@jolla.com>

Handle missing mode setting gracefully



See merge request !10

If usb mode setting is not present in usb-moded.ini and override is not
specified in kernel command line, get_mode_setting() will return null.
The handler for "get_config" D-Bus method call does not check the return
value and using the null value triggers abort from sanity checks at
libdbus. Attempting to pass NULL as string via D-Bus causes abort.

Use MODE_UNDEFINED as fallback in case the setting value is not defined.

Also remove potential memory leaks in the code used for determining the
mode setting value and use correct function for releasing the resulting
dynamically allocated string.
Signed-off-by: Simo Piiroinen <simo.piiroinen@jollamobile.com>

Signed-off-by: Jarko Poutiainen <jarko.poutiainen@jollamobile.com>