Xposed - Legacy thread. Don't panic, Xposed is still here.

rovo89 · Apr 6, 2012

Do you have Busybox installed? Or rather, do you have an /sbin/sh executable? The script is executed by /sbin/sh, which is linked to /sbin/busybox on my phone.

Brotuck · Apr 6, 2012

Yes busybox is installed (double checked)
version is 1.19.4-Stericson

rovo89 · Apr 6, 2012

I have the same.. but I'm not sure whether the link from /sbin/sh is automatically there. So I have checked the LPQ stock image and found out that /system/bin/sh is a part of it. I changed the references to it. Could you try that please?

Brotuck · Apr 6, 2012

Installation succes !

First message was:

So i thought install again and i got:

After that the program still said not installed, but i decided to reboot and after thatthe program said installed.

rovo89 · Apr 6, 2012

Thanks for the feedback! I hadn't thought of the commands like touch and chown that are executed by the script. I have also other versions of these. So now I changed it to always use the versions from /system/bin. I got the same errors then and corrected them. Also, the versions are now updated when you install.

I will update the first post with the new installer. You will probably need to reinstall, because the permissions could not be set correctly.

PS: Which ROM is that?

Brotuck · Apr 6, 2012

I will test the new apk no problem.
I am running xwlp4 odexed, rooted with speedmod

--edit --

Uninstalled and rebooted
Installed new version from fp and got this

Rebooted and xposed tells me installed

karl999999 · Apr 7, 2012

what about performance, and battery usage?
what are your thoughts on that?

rovo89 · Apr 8, 2012

Battery impact should be close to zero. Xposed does not require any services or additional wakelocks. Also it does no polling (frequent checks for a certain state). There is a small initialization phase, afterwards it gets only called when one of the hooked methods is about to be executed. So it does not prevent deep sleep because it gets only triggered by apps/services running anyway.

This of course depends on how the modules are implemented. The red clock example hooks a specialized method that is only called when the clock text is updated (once a minute). If it instead hooked a very generic method like setText, it would be called more often and it would have to execute more new code to find out if it has to do something.

The same applies to performance. If the developer is specific with the method hooks and the ratio of original to additional code is sensible, the impact will be low.

These are mainly thoughts as I could not test battery impact over a longer period yet (too much develop and restarts etc). But I did not feel a performance impact from the clock example, so the framework should be fine.

Jecht · Apr 8, 2012

So I've installed the apk and red clock test one aswell. It works fine on CM9 and seems like a great project.

wooki · Apr 9, 2012

Just a quick (maybe dumb) question/idea:
Would a module build with PDroid work with this?
I think PDroid changes some framework stuff and so we could have an easy way to intigrate it's privacy fetaures into our android rom (without using this "smali" patch thing on every rom update-zip we install)

Link to the thread is here:
http://xdaforums.com/showthread.php?t=1357056

(And here ICS specific thread: http://xdaforums.com/showthread.php?t=1554960 )

PDroid is changing framework.jar, services.jar, core.jar i think!

Thank you for reading and sorry for my bad english

mljjlm · Apr 11, 2012

This could also be used for theming right? I mean, make apps search for graphics in other places than it normally would?

Tungstwenty · Apr 11, 2012

@rovo89
Excellent idea and implementation, plus great tutorial for those not familiar with reflection, etc.

I can't wait to spare some time to take this for a test drive.
I already have a use case for it: put callbacks on the Power Manager so that in conjunction with BetterBatteryStats tool, one can not only know the process and WL that is responsible for the CPU being awake, but also activate an Xposed module that can catch all acquire/release invocations and log the stack trace so it's possible to know the exact java/smali places to look for WakeLock issues.

I've already browsed the (ASOP) PowerManagerService source and as far as I can see, a good candidate place could well be directly in Power.acquireWakeLock and .releaseWakeLock (this isn't necessarily the best place depending on the aggregation one wants to do, but let's suppose this is what's intended, for argument's sake

).

This brings me to my question: is it possible in your implementation to wrap native methods as well? Or must I go a bit upstream so I'm putting callbacks on "regular" java methods?

rovo89 · Apr 11, 2012

Jecht said:
So I've installed the apk and red clock test one aswell. It works fine on CM9 and seems like a great project.

Good to know that even the binaries are compatible for Stock, AOSP and CM9! That's what I hoped.

wooki said:
Just a quick (maybe dumb) question/idea:
Would a module build with PDroid work with this?

I didn't have the time to look at it in detail. From what I saw, most of the patches are for creating new classes and only a few things modify the existing code. In Xposed, some of the modifications might have to be done a bit differently (as you cannot change the code inside a method), but I think this should be possible.

mljjlm said:
This could also be used for theming right? I mean, make apps search for graphics in other places than it normally would?

Absolutely! Actually, I have already implemented new methods that make it very easy for developers to change some of the resources (e.g. strings and booleans). Unfortunately, with these changes the applications list in the settings crashes and I'm trying not to work around this, but fix the root cause. And for this, I have already spent like 30-40 hours of debugging in the past days. It's probably a stack corruption and very hard to debug.

Tungstwenty said:
I already have a use case for it: put callbacks on the Power Manager so that in conjunction with BetterBatteryStats tool, one can not only know the process and WL that is responsible for the CPU being awake, but also activate an Xposed module that can catch all acquire/release invocations and log the stack trace so it's possible to know the exact java/smali places to look for WakeLock issues.

That sounds like something I could use too!

What I would also like to know: Which wakelocks (from which apps) where active at what time?

Tungstwenty said:
This brings me to my question: is it possible in your implementation to wrap native methods as well? Or must I go a bit upstream so I'm putting callbacks on "regular" java methods?

I think it should work.. but I haven't tried yet. As mentioned, I'm currently busy with bug hunting. Afterwards, I will have a look at this.

_JKay_ · Apr 12, 2012

Great work! Congratulations and thanks!

For minor mods this is great! Replacing methods on java level instead of files on the smali level.

For larger mods I think this could quickly become rather complex to read and maintain.

How would you add new member variables to a class? Add new methods to a class?

Can one access existing member variables? Maybe use setAccessible(true) ?

Sorry for the noob questions... Just trying to find it's possibilities and limits...

I think I have to try this!

Thanks

Sent from my GT-N7000 using xda premium

Tungstwenty · Apr 12, 2012

Lol, I wouldn't be expecting JKay apologizing for noob questions

In reflection it's not possible to add methods or fields, although accessing existing member fields is just the same as accessing methods.
Have a look at the Reflection tutorial for a quick glance of what's possible to do. Then, for what's not (e.g. instrumentation or dynamic injection of new fields / methods), the pressure's on rovo89 to add it to his framework

I'm not familiar with your internal implementation, but lots of stuff should fit this model. When you're adding a wrapper to an existing class' method, you could maintain on your side (the wrapper) additional info and additional methods, and have them used whenever the right conditions are needed. And if multiple objects of the same type (or not) are wrapped by a single class or object of yours, you could still use the original object reference as a key to a Map where you store per-object stuff. Theoretically, most of the stuff should be possible ...

Performance might be something to consider (as well as code maintainability as you mentioned), but provided nothing is called lots of times in succession, the reflection overhead should be relatively low within all the code. Also, the main concern is power usage / deep sleep - where lots of custom lock screens, etc. put a lot of stress on power management - but on that front nothing would change.

_JKay_ · Apr 12, 2012

Tungstwenty said:
Lol, I wouldn't be expecting JKay apologizing for noob questions

In reflection it's not possible to add methods or fields, although accessing existing member fields is just the same as accessing methods.
Have a look at the Reflection tutorial for a quick glance of what's possible to do. Then, for what's not (e.g. instrumentation or dynamic injection of new fields / methods), the pressure's on rovo89 to add it to his framework

I'm not familiar with your internal implementation, but lots of stuff should fit this model. When you're adding a wrapper to an existing class' method, you could maintain on your side (the wrapper) additional info and additional methods, and have them used whenever the right conditions are needed. And if multiple objects of the same type (or not) are wrapped by a single class or object of yours, you could still use the original object reference as a key to a Map where you store per-object stuff. Theoretically, most of the stuff should be possible ...

Performance might be something to consider (as well as code maintainability as you mentioned), but provided nothing is called lots of times in succession, the reflection overhead should be relatively low within all the code. Also, the main concern is power usage / deep sleep - where lots of custom lock screens, etc. put a lot of stress on power management - but on that front nothing would change.

Thanks.. So it's kinda like extending a class in a package.

So I would have to rewrite the whole method? Or can I subclass it and call it's super class?

Sent from my GT-I9100 using xda premium

Tungstwenty · Apr 12, 2012

You don't need to rewrite it, but you're limited to either entirely replacing it with your own code (discard calling the original code in your wrapper), or invoking the original method and having your own instructions executing before and/or after the original stuff. Much like extending a class, yes, but where you're essentially overriding existing methods and in your implementation optionally calling the super(...) or not depending on what you want. You can also create extra methods in your own code, but they won't belong to the original class - they can only be called by your own code or attaching listeners for that, etc.

One thing you can't do, for instance, is calling the original method but ignore a certain instruction in there, you don't have that kind of granularity.
Well, with some creativity it might be worked around. For instance, if the method you're wrapping - but reusing the original as well - at some point inserts a value in a private Map field, you could call the super(...) method provided that before you created a callback as well on the Map.put(...) of that object so you're able to ignore the particular instruction that happens within the original method. Not sure if I make myself clear ...

(all this is based on my reflection knowledge and also the samples mentioned, I didn't have the chance to play with the framework yet)

lonelime · Apr 12, 2012

Hello,

can this software replace stock icons?

Thanks

rovo89 · Apr 12, 2012

Wow, I actually wanted to give JKay a heads up on my project, but now he came here himself. Great

Tungstwenty explained it already very well. Subclassing is basically what you can do with method hooking and you have an equivalent to super. You can of course access the methods and fields in any class (as long as you are in the correct process) and also set the values (works for private fields and I think even final ones). This is all standard reflection.

Adding new members and methods to a class is not easily possible. I might see whether I can do this somehow with native code, but there would have to be a use-case. For example, if you want to save something for every instance of a View, you could create a static Map<View,Something> in the module coding. Instead of using reflection to get/set the field value, just use map.get/put(view). I cannot imagine any use of new members outside of module coding. And you add new methods to the module instead of the class. You will get the "this" reference as parameter "thisObject" in callback methods anyway, so you can just pass it on to other methods in your module.

I agree that it would be good to replace some method calls only inside certain methods. You can do it, but it would require quite a bit of context checking. For common cases like this, I could implement helpers that would make it easier for developers. That's why you can register a method that should be called when a new package is being loaded (in that package's thread) - it can be done in other ways, but it is so common that a helper makes sense.

@lonelime: Do you mean resources like the battery icon etc.? That's what I'm currently working on. I finally - after some more hours of debugging - found the (really unobvious) reason why the applications list in the Settings app crashed and fixed it. So I can concentrate now on the resource loading stuff.

Current status: Replacing a string, boolean or integer resource from either the framework or a single app with a new value defined in the module's coding works fine. The approach I finally decided to take is one that replaces Resources instance returned by e.g. Activity.getResources() with a new class that inherits from Resources. This way, I do not need to hook getString(), but can use normal method overriding. You set the replacement values usually during startup. Then they are stored in a HashMap, which is queried in getString() etc. Apart from the (cached) creating of this instance, it only adds two HashMap lookups to the calls. And the good thing: This works with a constant time, no matter how many resources you replace!

You can look at this example to see how easy it is to replace a resource:

Code:

	public static void init(String startClassName) {
		if (startClassName != null)
			return;
		
		try {
			XResources.setSystemWideReplacement("android", "bool", "config_animateScreenLights", false);
			XResources.setSystemWideReplacement("android", "bool", "config_unplugTurnsOnScreen", false);
			XposedBridge.hookInitPackageResources(Example.class, "handleInitPackageResources", Callback.PRIORITY_DEFAULT);
		} catch (Throwable t) {
			XposedBridge.log(t);
		}
	}
	
	@SuppressWarnings("unused")
	private static void handleInitPackageResources(String packageName, XResources res) {
		// replacements only for SystemUI
		if (!packageName.equals("com.android.systemui"))
			return;
		
		res.setReplacement(0x7f080083, "YEAH!"); // WLAN toggle text. You should not do this because the id is not fixed. Only for framework resources, you could use android.R.string.something
		res.setReplacement("com.android.systemui:string/quickpanel_bluetooth_text", "WOO!");
		res.setReplacement("com.android.systemui", "string", "quickpanel_gps_text", "HOO!");
		res.setReplacement("com.android.systemui", "integer", "config_maxLevelOfSignalStrengthIndicator", 6);
	}

This adds the CRT off effect, keeps the device from turning on when you unplug it, modifies the text of some statusbar toggles (demonstrating different ways to call it) and gives you 6 instead of 4 bars for the signal strength.

Before releasing this, I will also try to implement replacing other types of resources. For things like Layouts and Drawables, you will most likely add the replacement to your module as a normal resource and then activate some kind of redirection. No release date though, I might not find the time for it in the next few days (for example, I could use some sleep

).

Tungstwenty · Apr 20, 2012

I finally had some time to install this and try out the 2 mods you created (not creating my own yet, sorry)

I noticed a few stacktraces on my logcat. Most if not all are not related with Xposed, only eventually the Clock modification might be since it's from when I had the red clock active (currently I only have the CRT one).
Care to take a look at the attached traces? Probably nothing related with you, but nevertheless.

Most of them have "dalvik.system.NativeStart.main()" at the root of the stack, and a method invocation soon after - I assume this is normal and part of your global hooking logic. The rest of the stack till the exception is just expected errors in some situations (network timeout, apk bugs, etc.)
In others, there is reference to "injection" but it looks as it's related with the applications themselves (e.g. roboguice library), not Xposed.
The ones where I'm in doubt are java.lang.IndexOutOfBoundsException and "Uncaught remote exception! (Exceptions are not yet supported across processes.)", both apparently related with the TIME_TICK broadcast and therefore might be the red clock mod (???)
Finally, there's also an ARM fault with native stack trace, you might also want to take a look at it since there are references in the stack to mangled function names such as "_Z14dvmCallMethodVP6ThreadPK6MethodP6ObjectbP6JValueSt9__va_list", "_Z15dvmInvokeMethodP6ObjectPK6MethodP11ArrayObjectS5_P11ClassObjectb", ...

One other thing:
In some of the traces, there is no call to the 2 reflection methods near the beginning. This happens in some of the NativeStart.main() roots, and also on stacks ending with "java.lang.Thread.run()".
Do you know whether in both of these situations the hooks are still applied, even if no mention of "method.invoke..." is found?

About the resources, let us know as soon as there's a usable version

Are you considering also binary resources such as .pngs, which would reside in the mod and be passed to the target app as if it lived there?
One suggestion I also have is for you to make available helper methods for stuff such as fetching fields or methods and make them accessible/public all in one go, and that kind of stuff that might ease up a bit the setting up of reflection.

PS: About changing final fields, that you mentioned earlier, I don't think that's possible. Although it's not 100% deterministic, in practice most of the times the compiler inlines the constant value in the places that access the field, so even if you change it (which is possible), the methods that uses the field already has the constant in their bytecode. Check e.g. this link for some info on this: http://stackoverflow.com/questions/...-private-static-final-field-didnt-do-anything

Cheers!

Xposed - Legacy thread. Don't panic, Xposed is still here.

Senior Recognized Developer

Senior Member

Senior Recognized Developer

Attachments

Senior Member

Senior Recognized Developer

Senior Member

Senior Member

Senior Recognized Developer

Member

Senior Member

Senior Member

Senior Member

Senior Recognized Developer

Retired Recognized Developer

Senior Member

Retired Recognized Developer

Senior Member

Senior Member

Senior Recognized Developer

Senior Member

Attachments

Similar threads

Top Liked Posts