mirror of
https://github.com/Droid-Deck/DroidDeck.git
synced 2026-11-21 13:55:58 +00:00
153 lines
5.2 KiB
Markdown
153 lines
5.2 KiB
Markdown
# Bionic Steam client probe
|
|
|
|
This is a research harness for Valve's Android/Bionic ARM64 Steam client
|
|
libraries. It does **not** vendor or redistribute Valve binaries. The build
|
|
script fetches the current package from Valve's public Steam client-update
|
|
endpoint into a temporary working directory.
|
|
|
|
The question under test is narrow:
|
|
|
|
> Can Valve's `androidarm64/libsteamclient.so` run directly under Android's
|
|
> Bionic linker, and if so, what does it need in order to become a usable
|
|
> Steam client?
|
|
|
|
## Result
|
|
|
|
Yes, the library itself works directly on stock Android. It is not a complete
|
|
Steam desktop client, however. It is a Steamworks IPC client for a separate
|
|
full Steam process.
|
|
|
|
On 2026-09-28, the public-beta ARM64 client manifest contained
|
|
`bins_androidarm64_linuxarm64.zip.35688b00e791987ab70b1b6c6ddd006d4c864470`
|
|
with SHA-256:
|
|
|
|
`811859a83c7c5ff220ac9731235e76fd0d5da603a5cdb56fe5f08e145a417cbd`
|
|
|
|
The archive contains Android ARM64 builds of `libsteamclient.so`,
|
|
`libsteamnetworkingsockets.so`, `libtier0_s.so`, `libvstdlib_s.so`, and
|
|
`steamservice.so`.
|
|
## ABI evidence
|
|
|
|
`libsteamclient.so` is a genuine Android/NDK build, not a glibc ARM64 binary
|
|
with an Android-flavored directory name:
|
|
|
|
- ELF64 AArch64 shared object.
|
|
- Depends on `libandroid.so`, `liblog.so`, `libm.so`, `libdl.so`, and Bionic
|
|
`libc.so`.
|
|
- Uses the NDK C++ namespace (`std::__ndk1`).
|
|
- Carries `.note.android.ident` and Android LLVM/Clang build strings.
|
|
- Contains no `GLIBC_*` symbol-version requirements.
|
|
|
|
Android's stock linker on an AYN Thor loaded both Valve libraries with no
|
|
compatibility shim. `CreateInterface("SteamClient023")` also succeeded.
|
|
|
|
## What `steamservice.so` proved
|
|
|
|
`steamservice.so` exports `SteamService_StartThread`,
|
|
`SteamService_GetIPCServer`, `SteamService_Stop`, and
|
|
`SteamService_Shutdown`. A non-null service name starts its worker thread
|
|
successfully on Android and shuts down normally.
|
|
|
|
That service is **not** the Steamworks host expected by `libsteamclient.so`.
|
|
Starting it does not create the TCP IPC listener used by `CreateSteamPipe`.
|
|
The service probe is retained because it establishes that the Bionic service
|
|
code itself is runnable and documents the calling convention.
|
|
## The actual dependency: Steam IPC on TCP 57343
|
|
|
|
With no Steam host available, both the legacy exported C entry point and the
|
|
`SteamClient023` virtual `CreateSteamPipe` method return `0`.
|
|
|
|
An `LD_PRELOAD` trace of `socket()` and `connect()` showed why:
|
|
|
|
```text
|
|
socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) -> fd
|
|
connect(fd, 127.0.0.1:57343) -> ECONNREFUSED
|
|
connect(fd, 127.0.0.1:57343) -> ECONNREFUSED
|
|
```
|
|
|
|
A dummy listener captured the client's first 17 bytes:
|
|
|
|
```text
|
|
0d 00 00 00 09 01 00 00 00 00 00 00 00 00 00 00 00
|
|
```
|
|
|
|
Port 57343 is Steam's long-standing local IPC listener. Valve's Lepton
|
|
Android compatibility environment uses the same Steam gateway for its
|
|
Android `libsteamclient.so`.
|
|
|
|
## End-to-end proof
|
|
|
|
Desktop Steam on a Mac was listening on host `127.0.0.1:57343`. We mapped
|
|
Android's loopback port to it temporarily with:
|
|
|
|
```sh
|
|
adb reverse tcp:57343 tcp:57343
|
|
```
|
|
With no local `steamservice.so` involved, the unmodified Bionic client on the
|
|
Thor then returned:
|
|
|
|
```text
|
|
SteamClient023=<non-null> rc=0
|
|
Steam_CreateSteamPipe=1
|
|
Steam_ConnectToGlobalUser=1
|
|
Steam_BConnected=1
|
|
Steam_BLoggedOn=1
|
|
Steam_BReleaseSteamPipe=1
|
|
```
|
|
|
|
This proves the Bionic client is operational on normal Android and can speak
|
|
the real cross-process Steam IPC protocol to an ordinary desktop Steam
|
|
process.
|
|
|
|
## Implication for DroidDeck / PRoot
|
|
|
|
This discovery does **not** by itself remove PRoot. The Android archive does
|
|
not contain the full Steam executable or `steamwebhelper`; the Bionic client
|
|
expects that full Steam process to exist elsewhere and own port 57343.
|
|
|
|
The useful architectural split is now proven rather than hypothetical:
|
|
|
|
```text
|
|
Android/Bionic workload
|
|
-> Valve Bionic libsteamclient.so
|
|
-> TCP Steam IPC :57343
|
|
-> full Steam host process
|
|
```
|
|
|
|
For a PRoot-free DroidDeck, the remaining problem is therefore the **host
|
|
Steam process**, not Steamworks ABI compatibility. Plausible future backends
|
|
are a real Linux environment (rooted chroot/namespaces), a VM/AVF guest with
|
|
a host bridge, or a future Valve Bionic build of the full client. Reimplementing
|
|
Steam's private IPC server should be treated as a last resort.
|
|
## Reproducing
|
|
|
|
Build and fetch Valve's current Android package into `/tmp`:
|
|
|
|
```sh
|
|
tools/research/bionic-steam-probe/fetch_build.sh
|
|
```
|
|
|
|
Run without a Steam host. This should resolve the Bionic interfaces but
|
|
return pipe/user handle `0`:
|
|
|
|
```sh
|
|
tools/research/bionic-steam-probe/run_probe.sh
|
|
```
|
|
|
|
If desktop Steam is running on the ADB host and listening on loopback, test
|
|
the real bridge:
|
|
|
|
```sh
|
|
tools/research/bionic-steam-probe/run_probe.sh --reverse-host-steam
|
|
```
|
|
|
|
The runner removes the ADB reverse mapping on exit. No Valve binary is copied
|
|
into this repository.
|
|
|
|
## Security note
|
|
|
|
Steam binds 57343 to loopback on desktop systems. Keep any DroidDeck bridge
|
|
equally constrained; do not expose this private Steam IPC endpoint to the LAN.
|
|
On Android all regular applications share the device network namespace, so a
|
|
future local host design should also consider how to prevent unrelated apps
|
|
from reaching a loopback-only Steam IPC listener. |