KernelSU icon indicating copy to clipboard operation
KernelSU copied to clipboard

KSU bug in granting root for apps on BlissOS

Open hmtheboy154 opened this issue 1 year ago • 55 comments

Please check before submitting an issue

  • [X] I have searched the issues and haven't found anything relevant
  • [X] I will upload bugreport file in KernelSU Manager - Settings - Report log
  • [X] I know how to reproduce the issue which may not be specific to my device

Describe the bug

(This issue may look like https://github.com/tiann/KernelSU/issues/2095, but I can't verify yet) Currently with latest KSU on BlissOS, tested on kernel 6.1 & 6.6 & 6.9 & 6.11, there are some bugs that I started to notice: While some apps like Aurora Store, Neo Store & FoxLog can be able to grant root and use normally, some are broken in functions or just straight up not working:

  • Termux can only be able to go to su if using /system/bin/su, and when in su no $PATH is set other than Termux or KSU own path.
  • MiXplorer can't be able to use Allow Root even set with rootexplorer.root or system template. I tried another app Amaze File Manager with their Root Explorer & Use legacy listing for root and it can do things like show the whole /data but can't access anything in /data/app.
  • If using another term app like TermOnePlus, it will not working even if using either su or /system/bin/su

All recent BlissOS builds with KernelSU 1.0.1 have this issue. And currently they are on kernel 6.1. I bumped kernel to 6.6 to test locally and it show the same behaviour. For newer kernel that are on stable branch, the same behaviour can be seen on this PR I made . Recently I do try to build for kernel 6.11 and it also have the same behaviour (here is my fork for 6.11).

To Reproduce

  • For kernel 6.1 + KSU 1.0.1 testing, download any latest stable build of BlissOS
  • For kernel 6.1 + KSU 0.9.2 testing, download older builds such as 16.9.6, 15.9.1 or 14.10.2
  • For kernel 6.9 + KSU 0.9.2 testing, download latest BlissOS Zenith (16.9.6)
  • For kernel 6.6 + KSU 1.0.1 or kernel 6.11 + KSU 1.0.1 tesing, it's currently WIP and still being worked on. If anyone ask, I'll provide.

Expected behavior

Apps can be able to grant root and work as expected

Screenshots

An example of Termux can't use su can be seen here : https://github.com/tiann/KernelSU/pull/1785#issuecomment-2143743204

Logs

will provide soon

Device info

  • Device: Generic x86_64-v2 devices
  • OS Version: Android 11/12.1/13/14
  • KernelSU Version: 1.0.1 (latest main)
  • Kernel Version: 6.1.x + 6.6.x + 6.11.x

Additional context

One last thing that I tested is to downgrade to 0.9.2 as some people suggested , it only work with kernel 6.1 & 6.6. If you port it to higher kernel like 6.9, it will have the exact same issue.

hmtheboy154 avatar Oct 02 '24 23:10 hmtheboy154

Same problem here. It seems like something in the shell is "cloaking" the su call If I pipe something to toybox, I can call su.

Maybe it is releated to:

# interesting detail of KernelSU - it doesn't actually place a su file in
# /system/bin, but hook execve(2) call in kernel to redirect it to kernel
# impl when executing /system/bin/su.

https://github.com/termux-play-store/termux-tools/commit/9187cfa169cc64861a0fc7e98a01b9c78eb1634d#diff-3de48b02d1685061aec7aeb7ee702149f1d3b8beee98158b5b35e9c94f88d6f7R9

2024-10-02 21_24_16-KSU bug in granting root for apps on BlissOS · Issue #2113 · tiann_KernelSU

hansalemaos avatar Oct 03 '24 00:10 hansalemaos

exec strace su and observe the output

tiann avatar Oct 03 '24 02:10 tiann

/system/bin/strace su says faccessat(AT_FDCWD, "/system/bin/su", X_OK) = -1 ENOENT (No such file or directory) Screenshot_20241003-095013 strace.txt

Xtr126 avatar Oct 03 '24 04:10 Xtr126

It seems that this kprobe failed: https://github.com/tiann/KernelSU/blob/main/kernel/sucompat.c#L265 is there any dmesg log?

tiann avatar Oct 03 '24 05:10 tiann

Looks like it has succeeded dmesg:

[    0.621438] KernelSU: sucompat: execve_kp: 0
[    0.622879] KernelSU: sucompat: newfstatat_kp: 0
[    0.624354] KernelSU: sucompat: faccessat_kp: 0
[    0.626269] KernelSU: sucompat: devpts_kp: 0
[    0.626359] KernelSU: ksud: execve_kp: 0
[    0.628102] KernelSU: ksud: vfs_read_kp: 0
[    0.630275] KernelSU: ksud: input_event_kp: 0
[   19.383064] KernelSU: on_post_fs_data!
[   19.383068] KernelSU: unregister input kprobe: 1!
[   19.383072] KernelSU: devpts sid: 741
[   19.383073] KernelSU: unregister execve kprobe: 1!

dmesg.txt

Xtr126 avatar Oct 03 '24 06:10 Xtr126

[   19.383068] KernelSU: unregister input kprobe: 1!
[   19.383072] KernelSU: devpts sid: 741
[   19.383073] KernelSU: unregister execve kprobe: 1!

this indicate it failed to unregister ?

hmtheboy154 avatar Oct 03 '24 17:10 hmtheboy154

@tiann I just remember something: When I test https://github.com/tiann/KernelSU/pull/2041 , I test it in kernel 6.1.84. And it work perfectly fine from granting apps to using modules. After the PR got merged I upstream the kernel and then I have reports about KSU bugging. So something in the upstream kernel is causing this. Which explain why stable branch affected the same even if using 0.9.2.

I am currently narrowing down to see up until which version does the bug happen, currently it's around 6.1.84 ~ 6.1.89

hmtheboy154 avatar Oct 04 '24 17:10 hmtheboy154

6.1.85 is when it happen

hmtheboy154 avatar Oct 05 '24 04:10 hmtheboy154

I just thought, it could be that ksud init broke because modules also don't work, modules.img does not get mounted.

Xtr126 avatar Oct 05 '24 13:10 Xtr126

I just thought, it could be that ksud init broke because modules also don't work, modules.img does not get mounted.

6.1.84 works perfectly fine with 1.0.1, here's an image for you to test https://drive.google.com/file/d/1Z9Fc0v25gf19n-6ia3JUm8UBQGM_LuIa/view?usp=sharing

hmtheboy154 avatar Oct 05 '24 14:10 hmtheboy154

I just thought, it could be that ksud init broke because modules also don't work, modules.img does not get mounted.

6.1.84 works perfectly fine with 1.0.1, here's an image for you to test https://drive.google.com/file/d/1Z9Fc0v25gf19n-6ia3JUm8UBQGM_LuIa/view?usp=sharing

I have tested 6.1.84 and can confirm everything works there, but I meant in the builds with newer kernel, modules don't work or even get mounted.

Xtr126 avatar Oct 05 '24 15:10 Xtr126

I just thought, it could be that ksud init broke because modules also don't work, modules.img does not get mounted.

6.1.84 works perfectly fine with 1.0.1, here's an image for you to test https://drive.google.com/file/d/1Z9Fc0v25gf19n-6ia3JUm8UBQGM_LuIa/view?usp=sharing

I have tested 6.1.84 and can confirm everything works there, but I meant in the builds with newer kernel, modules don't work or even get mounted.

try to get log, maybe there are more clue than faccessat() is broken

hmtheboy154 avatar Oct 05 '24 15:10 hmtheboy154

I tried even with loglevel=7 and checked in logcat but no clue yet. The dmesg doesn't differ much from that of the working version (6.1.84). But I saw that KernelSU has CONFIG_KSU_DEBUG, it could possibly help in getting more logs.

I just thought, it could be that ksud init broke because modules also don't work, modules.img does not get mounted.

6.1.84 works perfectly fine with 1.0.1, here's an image for you to test https://drive.google.com/file/d/1Z9Fc0v25gf19n-6ia3JUm8UBQGM_LuIa/view?usp=sharing

I have tested 6.1.84 and can confirm everything works there, but I meant in the builds with newer kernel, modules don't work or even get mounted.

try to get log, maybe there are more clue than faccessat() is broken

Xtr126 avatar Oct 05 '24 18:10 Xtr126

@tiann found the commit: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=1e3ad78334a69b36e107232e337f9d693dcc9df2

Revert this and it KSU work as expected now. And I even test this on kernel 6.11.2

hmtheboy154 avatar Oct 06 '24 14:10 hmtheboy154

so this seems affected all x86 platform not just Android-x86 based like us

hmtheboy154 avatar Oct 06 '24 14:10 hmtheboy154

please check cat /proc/kallsyms and dump the symbols

tiann avatar Oct 06 '24 14:10 tiann

please check cat /proc/kallsyms and dump the symbols

when KSU broke or when it's working ?

hmtheboy154 avatar Oct 06 '24 14:10 hmtheboy154

please check cat /proc/kallsyms and dump the symbols

when KSU broke or when it's working ?

broken, and the kernel image

tiann avatar Oct 06 '24 14:10 tiann

kallsyms.txt

and the kernel image

kernel.zip

hmtheboy154 avatar Oct 06 '24 14:10 hmtheboy154

So as a workaround, I saw this patch in LKML: https://lore.kernel.org/lkml/20240417164514.66hgypzxgqxt3ssk@desk/T/

I applied it and test on my kernel 6.11 and it's working so far https://github.com/android-generic/kernel-zenith/commit/0c2e19c58680393806bb133b1e9a07020eff7fe7

However, I still think that we should fix KSU itself.

hmtheboy154 avatar Oct 06 '24 17:10 hmtheboy154

So as a workaround, I saw this patch in LKML: https://lore.kernel.org/lkml/20240417164514.66hgypzxgqxt3ssk@desk/T/

I applied it and test on my kernel 6.11 and it's working so far android-generic/kernel-zenith@0c2e19c

However, I still think that we should fix KSU itself.

Build this on 6.1 requires disabling BHI mitigations by setting spectre_bhi=off though

hmtheboy154 avatar Oct 07 '24 18:10 hmtheboy154

@tiann I found this post : https://blog.hakuya.moe/post/13

hmtheboy154 avatar Oct 08 '24 19:10 hmtheboy154

So as a workaround, I saw this patch in LKML: https://lore.kernel.org/lkml/20240417164514.66hgypzxgqxt3ssk@desk/T/

I applied it and test on my kernel 6.11 and it's working so far android-generic/kernel-zenith@0c2e19c

However, I still think that we should fix KSU itself.

Finalize the temp fix: https://github.com/android-generic/kernel-zenith/commit/5648396f6c5fa039deb3e024d63520f6de9e2cfc

Because I don't want to fully revert, I put the option under a cmdline syscall_hardening=off . Until there's a new way to hook into x86_64 syscall, I guess this is what I can do.

hmtheboy154 avatar Oct 10 '24 13:10 hmtheboy154

So as a workaround, I saw this patch in LKML: https://lore.kernel.org/lkml/20240417164514.66hgypzxgqxt3ssk@desk/T/ I applied it and test on my kernel 6.11 and it's working so far android-generic/kernel-zenith@0c2e19c However, I still think that we should fix KSU itself.

Finalize the temp fix: android-generic/kernel-zenith@5648396

Because I don't want to fully revert, I put the option under a cmdline syscal_hardening=off . Until there's a new way to hook into x86_64 syscall, I guess this is what I can do.

Nice, any idea when the updated BlissOs will be online?

hansalemaos avatar Oct 10 '24 15:10 hansalemaos

So as a workaround, I saw this patch in LKML: https://lore.kernel.org/lkml/20240417164514.66hgypzxgqxt3ssk@desk/T/ I applied it and test on my kernel 6.11 and it's working so far android-generic/kernel-zenith@0c2e19c However, I still think that we should fix KSU itself.

Finalize the temp fix: android-generic/kernel-zenith@5648396 Because I don't want to fully revert, I put the option under a cmdline syscal_hardening=off . Until there's a new way to hook into x86_64 syscall, I guess this is what I can do.

Nice, any idea when the updated BlissOs will be online?

updated, please check

hmtheboy154 avatar Oct 13 '24 09:10 hmtheboy154

Thx, I will check that today!!

hansalemaos avatar Oct 13 '24 12:10 hansalemaos

2024-10-13 09_42_33-Window

Did not work :( Only with the Toybox trick I used before. I downloaded Bliss-v16.9.7-x86_64-OFFICIAL-gapps-20241011.iso

hansalemaos avatar Oct 13 '24 12:10 hansalemaos

2024-10-13 09_42_33-Window

Did not work :( Only with the Toybox trick I used before. I downloaded Bliss-v16.9.7-x86_64-OFFICIAL-gapps-20241011.iso

have you boot with syscall_hardening=off cmdline yet ?

hmtheboy154 avatar Oct 13 '24 12:10 hmtheboy154

nope, I should have read your old post, let me try

hansalemaos avatar Oct 13 '24 12:10 hansalemaos

WORKING! THX SO MUCH!

hansalemaos avatar Oct 13 '24 12:10 hansalemaos