Previously, we use either BroadcastReceivers or Activities to receive
messages from our native daemon, but both have their own downsides.
Some OEMs blocks broadcasts if the app is not running in the background,
regardless of who the caller is. Activities on the other hand, despite
working 100% of the time, will steal the focus of the current foreground
app, even though we are just doing some logging and showing a toast.
In addition, since stubs for hiding Magisk Manager is introduced, our
only communication method is left with the broadcast option, as
only broadcasting allows targeting a specific package name, not a
component name (which will be obfuscated in the case of stubs).
To make sure root requests will work on all devices, Magisk had to do
some experiments every boot to test whether broadcast is deliverable or
not. This makes the whole thing even more complicated then ever.
So lets take a look at another kind of component in Android apps:
ContentProviders. It is a vital part of Android's ecosystem, and as far
as I know no OEMs will block requests to ContentProviders (or else
tons of functionality will break catastrophically). Starting at API 11,
the system supports calling a specific method in ContentProviders,
optionally sending extra data along with the method call. This is
perfect for the native daemon to start a communication with Magisk
Manager. Another cool thing is that we no longer need to know the
component name of the reciever, as ContentProviders identify themselves
with an "authority" name, which in Magisk Manager's case is tied to the
package name. We already have a mechanism to keep track of our current
manager package name, so this works out of the box.
So yay! No more flaky broadcast tests, no more stupid OEMs blocking
broadcasts for some bizzare reasons. This method should in theory
work on almost all devices and situations.
On API 23+, the platform unifies the way to handle drawable
resources across processes: all drawables can be passed via Icon.
This allows us to send raw bitmap to the system without the need to
specify a resource ID. This means that we are allowed to NOT include
these drawable resources within our stub APK, since our full APK can
draw the images programmatically and send raw bitmaps to the system.
- Skip 0x7f01XXXX - 0x7f05XXXX resource IDs in the main app; they are
reserved for stub resources
- Support sending additional data from host to guest
- Use resource mapping passed from host when they are being sent
to the system framework (notifications and shortcuts)
In the effort of preventing apps from crawling APK contents across the
whole installed app list to detect Magisk Manager, the solution here
is to NOT install the actual APK into the system, but instead
dynamically load the full app at runtime by a stub app. The full APK
will be stored in the application's private internal data where
non-root processes cannot read or scan.
The basis of this implementation is the class "AppComponentFactory"
that is introduced in API 28. If assigned, the system framework will
delegate app component instantiation to our custom implementation,
which allows us to do all sorts of crazy stuffs, in our case dynamically
load classes and create objects that does not exist in our APK.
There are a few challenges to achieve our goal though. First, Java
ClassLoaders follow the "delegation pattern", which means class loading
resolution will first be delegated to the parent loader before we get
a chance to do anything. This includes DexClassLoader, which is what
we will be using to load DEX files at runtime. This is a problem
because our stub app and full app share quite a lot of class names.
A custom ClassLoader, DynamicClassLoader, is created to overcome this
issue: it will always load classes in its current dex path before
delegating it to the parent.
Second, all app components (with the exception of runtime
BroadcastReceivers) are required to be declared in AndroidManifest.xml.
The full Magisk Manager has quite a lot of components (including
those from WorkManager and Room). The solution is to copy the complete
AndroidManifest.xml from the full app to the stub, and our
AppComponentFactory is responsible to construct the proper objects or
return dummy implementations in case the full APK isn't downloaded yet.
Third, other than classes, all resources required to run the full app
are also not bundled with the stub APK. We have to call an internal API
`AssetManager.addAssetPath(String)` to add our downloaded full APK into
AssetManager in order to access resources within our full app. That
internal API has existed forever, and is whitelisted from restricted
API access on modern Android versions, so it is pretty safe to use.
Fourth, on the subject of resources, some resources are not just being
used by our app at runtime. Resources such as the app icon, app label,
launch theme, basically everything referred in AndroidManifest.xml,
are used by the system to display the app properly. The system get these
resources via resource IDs and direct loading from the installed APK.
This subset of resources would have to be copied into the stub to make
the app work properly.
Fifth, resource IDs are used all over the place in XMLs and Java code.
The resource IDs in the stub and full app cannot missmatch, or
somewhere, either it be the system or AssetManager, will refer to the
incorrect resource. The full app will have to include all resources in
the stub, and all of them have to be assigned to the exact same IDs in
both APKs. To achieve this, we use AAPT2's "--emit-ids" option to dump
the resource ID mapping when building the stub, and "--stable-ids" when
building the full APK to make sure all overlapping resources in full
and stub are always assigned to the same ID.
Finally, both stub and full app have to work properly independently.
On 9.0+, the stub will have to first launch an Activity to download
the full APK before it can relaunch into the full app. On pre-9.0, the
stub should behave as it always did: download and prompt installation
to upgrade itself to full Magisk Manager. In the full app, the goal
is to introduce minimal intrusion to the code base to make sure this
whole thing is maintainable in the future. Fortunately, the solution
ends up pretty slick: all ContextWrappers in the app will be injected
with custom Contexts. The custom Contexts will return our patched
Resources object and the ClassLoader that loads itself, which will be
DynamicClassLoader in the case of running as a delegate app.
By directly patching the base Context of ContextWrappers (which covers
tons of app components) and in the Koin DI, the effect propagates deep
into every aspect of the code, making this change basically fully
transparent to almost every piece of code in full Magisk Manager.
After this commit, the stub app is able to properly download and launch
the full app, with most basic functionalities working just fine.
Do not expect Magisk Manager upgrades and hiding (repackaging) to
work properly, and some other minor issues might pop up.
This feature is still in the early WIP stages.
