Vivaldi not Loading Websites

[marcoh242]

So I installed the latest(7cbb41b) version of Zygisk-Assistant with KSU and got finally my banking app running with that. Unfortunetely my browser vivaldi wont display any website anymore as long this module is active. Other browser work fine...so not sure where to look to debug that.

[snake-4]

Could you try these steps?

1. Disable other modules.
2. Install the latest debug build of this module.
3. After reproducing the bug, do adb logcat > log.txt on your PC and upload the log here.

[marcoh242]

This issue only occurs when Zygisk-Assistant AND https://github.com/AndroidAudioMods/ViPER4Android is active.
As I dont really need Viper4Android its no big issue for me. Attached the log anyway if u wanna look over it.
log.txt

[snake-4]

I've checked the log and I don't think the ViPER's bind mount is there at all and this is what the chromium crash log states:
[ERROR:gl_display.cc(515)] EGL Driver message (Error) eglCreateContext: Requested GLES version (3.0) is greater than max supported (2, 0).

Have you tried without Shamiko as well? The issue doesn't seem related to either modules.

[marcoh242]

Tested it now only with Viper4Android and Zygisk-Assistant and it doesnt work. Disabling on of them and it works.

[privacyguy123]

A Shamiko dev once helped me fix it with

ro.zygote.disable_gl_preload=1

Perhaps same issue with your module if you are mass umounting things

[snake-4]

A Shamiko dev once helped me fix it with

ro.zygote.disable_gl_preload=1

Perhaps same issue with your module if you are mass umounting things

This makes sense and it was mentioned in another issue too. But I didn't think that would be the case here since Viper module doesn't seem replace the OpenGL drivers.

[aviraxp]

I am still looking on this too. I am not sure how OpenGL has to do with it. I only note that umount every module that mounts files in /vendor will cause this behavior.

Note OpenGL is preloaded in zygote as well, so maybe there is some additional checks that will be inconsistent and fail after umounting?

[aviraxp]

Also, this issue seems only exist on ksu.

[snake-4]

I am still looking on this too. I am not sure how OpenGL has to do with it. I only note that umount every module that mounts files in /vendor will cause this behavior.

Note OpenGL is preloaded in zygote as well, so maybe there is some additional checks that will be inconsistent and fail after umounting?

Oh, I figured it out! When there are files to be replaced, KSU mounts an overlayfs over the parent folder, whereas Magisk modules just bind mount the files.

I'm sure this is obvious, but when you call umount2(..., MNT_DETACH) the overlayfs device gets detached from the FS tree. The problem is that every file opened on /vendor before that would refer to the overlayfs device in their inode structures and not the lowerdir device. So procfs wouldn't be able to determine the base path for these FDs because the device that inode refers to is not in the FS tree anymore!

Zygote just closes these FDs after the fork because the path appears like /egl/lib.so. Now, not all Adreno drivers open FDs to other libs on preload. Particularly, drivers of Adreno 630 don't seem to do this but Adreno 660 drivers do it.

I'm thinking of getting rid of MNT_DETACH entirely because it has too many side effects. I think we can instead close the FDs, unmount normally without MNT_DETACH, then re-open all FDs so they would refer to the correct device.

This would also solve #11 but I'm not sure if that's how I want to solve that one because I think just replacing the FDs without actually removing the RROs from AssetManager would have some side effects.

[privacyguy123]

You've blown my mind wide open with this latest explanation - good work. You've helped explain a lot of weirdness I've been trying to work out on my side

[snake-4]

@marcoh242 Could you try the latest CI debug build and see if that fixes the problem? Also please attach the logs even if it works.

[marcoh242]

Not working
log.txt

[snake-4]

I see. I can't reproduce this bug on my only device probably because it's 4.9 and 2a9c6d0 (ovl: allocate anonymous devs for lowerdirs) was added in 4.15. I'm sure that a dlopen is failing somewhere in the EGL libs likely because the dev changes.

If anyone wants to give it a go, check linker logs with LD_DEBUG=3. Also compare lsof -p $(pidof zygote64) with lsof -p $(pidof com.vivaldi.browser) both with the module on and off.

[oldman20]

hello, i think i have similiar case, but with EDGE, i install 1f2e9d3 and edge just black screen, asa force close