Repository navigation
SM-A566E / DZJ1 (Android 17, kernel 6.6.127): DFRoot dies at the trigger; the fork verifies root, then the device resets — Samsung's reset history blames the EL2 layer, no kernel panic #199
VesuviusXK
started this conversation in
General
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Redirected here by the DirtyFrag maintainer, who closed the downstream report (mitschud#171) as not planned with the comment "Report to upstream." Filing the full evidence here, since the module-loading architecture is upstream's.
Device
DFRoot v4.7 on this device (filmed)
Setup finds ko_target and all patches apply (dfroot.ko, 10432 bytes). The log ends at "=== init ===" / "Triggering hook..." — the screen goes black within about a second, with no post-trigger output at all. The device has reset.
The downstream fork on the same device (DirtyFrag 1.18 pre-release, master build), for contrast
The identical design gets further: every patch applies (dfroot.ko, 10560 bytes, KMI android15-6.6), the hook triggers, the module loads, the bootstrap runs through "starting SU daemon," cleanup restores the trampoline in libc++.so, and the app reports 100% / "Verified" / "Rooted." Roughly half a second later the device resets. This repeats identically on every run.
Why this is not a kernel panic — the device's own reset records
I pulled a full dumpstate via the *#9900# SysDump menu after the runs. The reset-history table lists 14 abnormal resets that day (16:34–18:00), covering every run of DFRoot, the fork, and a third independent implementation. Every reset on this firmware carries the identical signature — no panic, no PC, no LR:
2026-11-11 16:34:34-0500 | ON | DP | A566EXXUEDZJ1 / OFFSRC: (M)- (S)- / ONSRC: ACOKB / RST_STAT: SOFT / RCNT: 1 / PC: - LR: - WDGC: exynos_report_el2_crash_info+0x540/0x558 [exynos_el2] [227]The same table's Android 16 entries (previous firmware, CCZG6) DO record real kernel panics —
PANIC: Oops: Fatal exceptionin rt_mutex_adjust_prio_chain — so the phone records panics when they happen. There is none for any reset on this build. The last-kmsg from a crashed boot shows the chain mid-work (dfroot: selinux permissive,dfroot: umh_exec ok, libksud starting,KernelSU: patch syscall 42) with zero panic or fatal-exception lines.Conclusion
On this firmware, a foreign kernel module activating is caught by Samsung's EL2-level protection (the exynos_el2 layer underneath the kernel), which resets the device. It is not a module build mismatch — the records rule out a panic entirely. DFRoot dies at the trigger; the fork gets root verified first only because the check is not instantaneous. Either way, the wall is module activation itself, which is why I am filing it here rather than downstream. Related: #123 (S25 Ultra, same kernel 6.6.127, One UI 9, not working) reported the symptom without this diagnosis.
One pointer on the endgame
Mobile Hacking Lab published a GhostLock port for the Galaxy A17 that reaches a persistent root shell on a retail Samsung with the EL2 protections (KDP, DEFEX, SELinux) fully active: https://www.mobilehackinglab.com/blog/cve-2026-43499-ghostlock-a17-root-shell — source: https://github.com/mobilehackinglab/ghostlock-a17. Their finding: patching process credentials silently fails because EL2 keeps those pages read-only to the kernel, so instead their chain forges a workqueue work item and has the kernel execute the usermode-helper path — no module is loaded and no protected memory is written. Different bug and different device, and I do not know whether it survives this build's watchdog — but every failure here ends at module activation, so a no-module endgame seems like the open direction.
Available on request
The full dumpstate zip (10 MB, over GitHub's attachment limit — can send via a file link), the filmed runs of both tools, and I am happy to test builds on this hardware.
Video.Proof.mp4
All reactions