Research toolkit that roots a Samsung Tizen 9 TV via a Chromium WasmGC RCE and V4L2 physmap arbitrary physical memory access, with persistent on-boot payload and root shell channel.
Unofficial security research on a 2024 Samsung Neo QLED running Tizen 9. The exploit chain needs no sdb, no developer mode, no firmware flashing: the TV is rooted from its own browser, and the root is made persistent through a stock on-boot service app.
| Test unit | 2024 Neo QLED, QNX1D series (QTV T09 platform) |
| Firmware | T-PTMxxxx (01400) |
| Kernel | Linux 5.4.249, ARM32 (VMSPLIT_3G + HIGHMEM, no LPAE), PAGE_OFFSET 0xC0000000, PHYS_OFFSET 0x20000000 |
| Browser | Chromium 120.0.6099.5 (SamsungBrowser/7.0) |
| Root | uid=0, CapEff=0x3fffffffff |
| Persistent payload | repo v47 (current md5/size in the AGENTS.md state row); payload_backup.dll keeps the previous build |
| Root channel | TCP 31337 on the TV, line protocol, replies end with __END__ |
pwsh -File .\root.ps1 -TvIp <tv-ip> # root the TV
pwsh -File .\uninstall.ps1 -TvIp <tv-ip> # restore stock
root.ps1 runs on the PC and drives everything: it starts the port-80 server,
roots the browser stub, installs the boot payload, and waits for the root channel
after the reboot. Two steps stay on the TV remote, and it prompts for both:
http://<pc-ip>/index.html and wait for RUNNING;Then:
python tools\root_shell.py "id" <tv-ip> # talk to the TV's root channel
Every option, the exact prompts, expected output, the host-IP caveat and the
manual fallback: docs/USAGE.md.
/dev/video40/30/60) accept an
attacker-chosen vm_pgoff in mmap() and pass it to remap_pfn_range()
unvalidated: arbitrary physical memory read/write from an unprivileged
process (uid 5001, group video). Exclusivity is per instance: the
payload retries video40 on EBUSY and falls back to video30/video60
only on ENOENT/ENODEV — never cycle nodes on busy.init_task in physical memory, locate our
own task through a forked child's real_parent, and zero the credentials.app_onboot service with
a .NET payload that repeats the physmap/cred patch at every boot, closes its
node and serves a root command channel on TCP 31337.No GPU CVE, no kernel exploit, no signed module. The physmap is the interesting
part; the stage-by-stage writeup with the measured constants is
docs/EXPLOIT.md.
The kernel you analyse comes from two sources, in order: the OTA package
downloadable before rooting (strip the u-boot header off its uImage), and
after rooting the TV itself — flash partition or live RAM — via
tools/dump_kernel_chunked.py and tools/dump_kernel_ram.py. Key rule:
trust RAM, not the firmware file, for any offset that goes into shellcode —
the two are usually different builds (the TV may run a newer build than
the package you downloaded, and each build links its sections at slightly
different addresses; it is not a runtime relocation). Full procedure,
measured examples and the flash-vs-RAM build differences:
docs/KERNEL-DUMPS.md.
The package-inventory / disable-list work (QNX1D classification, apply and
restore tooling, measured resource budget) lives in this repository under
debloat/. It consumes this toolkit's root channel as
its only dependency. The payload re-asserts the durable service masks at every
boot on its own, and root.ps1 runs a debloat health check (tools/tv_health.py)
against debloat/registry.tsv after a deploy.
Rooting and debloating are separate steps. The deploy installs only the
policy — deploy.sh fetches services-units.txt (29 units),
services-bind.txt (23 kill binaries), debloat-apps.txt (the registry's 73
applied ids) and debloat-sentinels.txt into /opt/usr/data/, and the payload
re-applies the service masks at every boot. The app disables — the 73
package_app_info.app_disable rows and the canalysis/pisa sentinels — are
written by debloat/tools/apply.sh, a one-time per-TV run that also records an
undo log and arms /opt/usr/data/debloat-enforce; with that marker present the
v49 payload repairs any row an update reset (and re-plants the sentinels) at
every boot. On a TV where the apply has never run, the payload beacons
apps skip (not armed) and tv_health reports drift (applied/count/
sentinels); root.ps1 then exits 4 (deploy succeeded, acceptance did not) —
use -SkipHealth or run the apply first.
Two TV-side behaviours are commonly worth changing once rooted. Neither is part
of the payload — both are ordinary settings stored on the TV (/opt/dbspace),
so a fresh root (or a stock set) keeps stock behaviour until you set them.
docs/TVPLUS-AUTOSTART.md.docs/SAMSUNG-UPDATES.md.