You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
HONOR 200 Pro ELP-NX9: page-cache probe unchanged; XfrmInStateProtoError increases
#189
On HONOR 200 Pro (ELP-NX9), DFRoot 4.7 fails during the first crash_dump64 patch: the target page remains unchanged. A custom diagnostic build confirms that the block is identical before and after the attempted write, while XfrmInStateProtoError increases.
The issue is reproducible without reaching the hook or loading the kernel module. No root was obtained and no partitions were modified.
Kernel build: #1 SMP PREEMPT Mon Oct 13 11:54:31 UTC 2025 aarch64
Bootloader: locked
Root manager: ReSukiSU v4.2.0-rc3
DFRoot: 4.7
Original failure
found ko_target: /vendor/lib64/libbinderdebug.so
=== exploit (patching files) ===
patch: crash_dump64 <- splicehelper (1488 bytes)
patched 1488 bytes to /apex/com.android.runtime/bin/crash_dump64+0x0
patch: crash_dump64 verification failed
patch: DEVICE NOT VULNERABLE
patch: ERROR - failed to patch files
The message says DEVICE NOT VULNERABLE, but disassembly of the exact boot.img shows that esp_input() still has the old branch without skb_has_shared_frag():
All seven symbol CRCs match the official vmlinux.symvers, including module_layout, register_kprobe, and unregister_kprobe.
The custom APK uses this module, but the failure occurs before it is loaded, so the module is not the cause of this particular result.
Single-block diagnostic probe
I prepared a diagnostic build that performs exactly one 16-byte write attempt to crash_dump64, reads the block back, and stops before the rest of the exploit chain.
Result:
diag: before = 7f454c4602010100000000000000000000
diag: expected = dae0e9e3a7a4a4a5a5a5a5a5a5a5a5a5
patched 16 bytes to /apex/com.android.runtime/bin/crash_dump64+0x0
diag: actual = 7f454c4602010100000000000000000000
diag: write_rc=0 changed=0 expected_match=0
diag: single-block probe complete; stopping before the exploit chain
Thus the target content does not change at all. This is not merely a verification-offset issue.
XFRM counters
The diagnostic app itself cannot read /proc/net/xfrm_stat due to SELinux, so I captured it via ADB before and after the same single-block attempt.
Only this counter changed:
XfrmInStateProtoError: 34 -> 36
There was no increase in XfrmInNoStates, XfrmInHdrError, XfrmInStateModeError, or XfrmInError.
In this kernel source, XfrmInStateProtoError is incremented after the state input handler returns a negative result (and in offload failure paths). This suggests that the packet reaches the ESP input handler, but the attempted write still does not land in the target page.
Current hypothesis
The vulnerable esp_input() branch is present, but the target file page may be copied before decryption or otherwise not used as the destination of the in-place write. Without root, I cannot create kprobes or enable tracepoints to inspect the actual skb/scatterlist.
I would appreciate guidance on:
Whether this failure pattern is already known for Android 14 / 6.1 devices.
Whether a cloned/copied skb on the UDP or XFRM path would explain an unchanged page despite the vulnerable esp_input() branch.
Which userspace-side checks or packet/socket parameters would help distinguish that case before attempting further writes.
The matching boot.img, recovered kallsyms, and additional logs are available if needed for reproduction.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Summary
On HONOR 200 Pro (ELP-NX9), DFRoot 4.7 fails during the first
crash_dump64patch: the target page remains unchanged. A custom diagnostic build confirms that the block is identical before and after the attempted write, whileXfrmInStateProtoErrorincreases.The issue is reproducible without reaching the hook or loading the kernel module. No root was obtained and no partitions were modified.
Device information
ELP-N39 10.0.0.146(C185E5R3P1)2026-05-016.1.128-android14-11-g58deda01e47b-ab14263620#1 SMP PREEMPT Mon Oct 13 11:54:31 UTC 2025 aarch64v4.2.0-rc3Original failure
The message says
DEVICE NOT VULNERABLE, but disassembly of the exactboot.imgshows thatesp_input()still has the old branch withoutskb_has_shared_frag():The kernel has
CONFIG_XFRM=y,CONFIG_XFRM_ESP=y, andCONFIG_INET_ESP=y.Custom module built for the exact kernel
Because the bundled
android14-6.1module hasvermagic=6.1.166-dirty, I rebuiltdfroot.kofrom the exact matching GKI source:android14-6.1-2025-03_r1358deda01e47b2c5816b41824aea076b90cc5926a14263620(the device'sab14263620)The resulting module has:
All seven symbol CRCs match the official
vmlinux.symvers, includingmodule_layout,register_kprobe, andunregister_kprobe.The custom APK uses this module, but the failure occurs before it is loaded, so the module is not the cause of this particular result.
Single-block diagnostic probe
I prepared a diagnostic build that performs exactly one 16-byte write attempt to
crash_dump64, reads the block back, and stops before the rest of the exploit chain.Result:
Thus the target content does not change at all. This is not merely a verification-offset issue.
XFRM counters
The diagnostic app itself cannot read
/proc/net/xfrm_statdue to SELinux, so I captured it via ADB before and after the same single-block attempt.Only this counter changed:
There was no increase in
XfrmInNoStates,XfrmInHdrError,XfrmInStateModeError, orXfrmInError.In this kernel source,
XfrmInStateProtoErroris incremented after the state input handler returns a negative result (and in offload failure paths). This suggests that the packet reaches the ESP input handler, but the attempted write still does not land in the target page.Current hypothesis
The vulnerable
esp_input()branch is present, but the target file page may be copied before decryption or otherwise not used as the destination of the in-place write. Without root, I cannot create kprobes or enable tracepoints to inspect the actualskb/scatterlist.I would appreciate guidance on:
skbon the UDP or XFRM path would explain an unchanged page despite the vulnerableesp_input()branch.The matching
boot.img, recovered kallsyms, and additional logs are available if needed for reproduction.All reactions