Porting template and derivation toolkit for adapting the CVE-2026-64560 Android kernel UAF local-privilege-escalation chain to the Redmi 14C (pond, MediaTek).
pond) — CVE-2026-64560 Porting Kit (China Variant / MediaTek)This is a "values-to-be-filled" porting skeleton, not an exploit that can run directly. It takes the upstream NebuSec/CyberMeowfia CVE-2026-64560 Android LPE exploit chain, as well as the device adaptations for
dada(Xiaomi 15 / SM8750) andmiro(K80 Pro / Snapdragon 8 Elite), two Qualcomm flagship devices, and structurally reworks them into a porting template targeting the China-variant Redmi 14C (codenamepond, MediaTek Helio G81-Ultra).All constants that need to be derived from "the real kernel
Imageof your device" are left blank in this kit (sentinel valuePOND_TODO = 0xdeadbeefdeadbeef), and are guarded by compile-time_Static_assertgates to ensure: as long as even one constant is unfilled, it will not compile. This is the upstreammiro-profile.h's own safety design, inherited as-is by this kit.
CVE-2026-64560 (posix-cpu-timers non-leader exec() race UAF) on the China-variant Redmi 14C
(codename pond, model 2409BRN2CC): "the vulnerability very likely exists, the exploit chain is in principle portable, but none of dada/miro's address constants can be reused",
because this is a Qualcomm → MediaTek cross-vendor port, where the physical memory layout, CPU topology,
and resource limits are all different. This kit provides everything needed to safely port the exploit chain to pond:
the structures, difference analysis, and step-by-step derivation procedures; a truly runnable binary can only be produced
after you fill the derived values back into src/redmi14c-pond-profile.h on your own device.**
| Criterion | Measured | Conclusion |
|---|---|---|
| Kernel | 6.6.89-android15-8 (GKI, same android15-8 KMI line as dada/miro) | < upstream fixed version 6.6.147 → affected |
| Patch | SPL 2026-07-01, CVE 2026-07-29 disclosure | unpatched |
| Fatal config | INIT_ON_FREE_DEFAULT_ON not configured, RANDOM_KMALLOC_CACHES not configured | ✓ no fatal items |
| Components | UHID/ASHMEM/CONFIGFS/SELINUX all =y | ✓ bridge complete |
| Known mitigations | INIT_ON_ALLOC_DEFAULT_ON=y + SLAB_FREELIST_RANDOM=y (same as miro) | hit rate reduced but feasible |
Three empirical facts that overturn old assumptions + three items pending re-verification (kmalloc-cg-8k line count = 0, device BTF missing,
DTB reg unreadable) + BL lock risk + precise next-step checklist, all in
docs/03-probe-analysis.md.
| 字段 | 内容 |
|---|---|
| CVE ID | CVE-2026-64560 |
| 标题 | posix-cpu-timers: Prevent UAF caused by non-leader exec() race |
| 类型 | Use-After-Free(CWE-416),竞态条件 |
| CVSS v3.1 | 7.8 High — AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H(本地提权,非远程) |
| 公开日期 | 2026-07-29 |
| 修复提交(mainline) | 920f893f735e92ba3a1cd9256899a186b161928d |
| 引入提交(v5.7) | 55e8c8eb2c7b(2020) |
| 报告者 | Wongi Lee、Jungwoo Lee |
| 影响内核 | 5.7 → 修复前(5.10.262 / 5.15.213 / 6.1.180 / 6.6.147 / 6.12.100 / 6.18.41 / 7.1.5 / 7.2-rc3 之前) |
Principle (minimal): When a non-leader thread calls execve(), de_thread()→__exit_signal()
sets the old leader's sighand to NULL; meanwhile, on another CPU, sys_timer_delete()→
posix_cpu_timer_del() looks up the old leader, sees sighand==NULL, and "does nothing and returns success directly",
then free_posix_timer() frees the k_itimer. But exec (unlike exit) does not change the TGID; an already-armed CLOCK_PROCESS_CPUTIME_ID timer attached to signal->cpu_timers is
inherited by the new leader and remains enqueued → afterwards run_posix_cpu_timers() / other timerqueue operations access the dangling node = UAF.
Key property: AV:L/PR:L — the attacker needs to already have low-privilege local code execution on the machine (e.g., an adb shell
identity). It is not a remote vulnerability and cannot attack devices to which you have no physical/code access. This is consistent with dada/miro's
use case: obtain temporary root of the current boot on a device you own (uid=0 + shell domain), then use it to
load the KernelSU LKM. See section 4 of docs/05-technical.md in the upstream miro repository for the
boundary discussion on "appification".
This suite is based on a line-by-line reading of every .md file (19 in total) + PoC source code + patches +
device constant headers + integration references in the following three repositories:
villager1314/CVE-2026-64560-Analysis (upstream analysis / trigger)poc/cve_2026_64560_poc.c) + the full mainline patch + technical analysis.quyicheng03-boop/xiaomi15-dada-cve-2026-64560 (dada adaptation)Image SHA-256 binding: the runner verifies the fingerprint, kernel release,
page size, CPU topology, UHID minor, boot ID, and the hash of the decompressed Image; any mismatch causes
a safe exit with TARGET_PROFILE_GATE_FAIL, without touching the kernel.docs/PORTING.md (porting workflow), docs/TECHNICAL.md (adaptation notes),
and docs/TESTING.md (acceptance criteria). Upstream source = NebuSec/CyberMeowfia.RMDycz/K80Pro-miro-CVE-2026-64560 (miro port)src/miro-profile.h is the model for porting constants: 24 link-layout symbols, struct offsets,
direct-map aliases, and race tuning parameters, all annotated with [V]/[B]/[M]/[T]/[A] provenance +
a compile-time _Static_assert integrity gate.docs/05-technical.md is the most complete porting operations manual: what was derived, how it was derived, and which pitfalls were hit.physical_delta that cannot be read offline is inherited from dada because it is an SM8750 platform constant.Of these three points, the "inheritance premise" of the last one does not hold on
pond— this is the fundamental reason this suite exists.
| Attribute | dada (Xiaomi 15) | miro (K80 Pro) | pond (China Redmi 14C) |
|---|---|---|---|
| SoC vendor | Qualcomm | Qualcomm | MediaTek |
| SoC | SM8750 Snapdragon 8 Elite | Snapdragon 8 Elite series | Helio G81-Ultra (MT6769V/Z) |
| CPU topology | Flagship multi-cluster (prime+perf+e) | Same as left | 2×Cortex-A75(big,2.0G) + 6×Cortex-A55(LITTLE,1.8G) |
| Process/RAM | 3nm / LPDDR5X 12-16G | Same as left | 12nm / LPDDR4X 4-8G |
| Storage | UFS 4.0 | UFS | eMMC 5.1 |
| Codename | dada | miro | pond (model 2409BRN2CC, China) |
| Kernel | GKI 6.6.118-android15-8 | Same as left | GKI 6.6.89-android15-8 (same android15-8 KMI line) |
| Kernel lineage | GKI android15-8 / 6.6.118 | Same as left | To be confirmed by you on the device with uname -r (MTK BSP, not the same GKI line) |