Frida 17.23.3 Released ∞
release
oleavr
Our .NET bindings are reborn, thanks to @AeonLucid, and @Cassian433 is back with two more fixes.
.NET
frida-clr has been around since the early days, as a C++/CLI wrapper that only ever worked on Windows, against the .NET Framework. It hasn’t kept up with the rest of our bindings, and more and more of the API was missing.
AeonLucid rewrote it from scratch as frida-net, and kindly contributed it back. The new bindings are plain C# on .NET 9, generated from the same GIR that frida-python, frida-node and frida-swift are generated from, through gir.core. That means they cover the full API, and that they’ll keep up with it automatically. And since nothing in there is Windows-specific anymore, they run on Windows, macOS and Linux, with the native runtime packaged per platform.
The result is a Frida.NET package with FridaDeviceManager, FridaDevice, FridaSession and FridaScript at the center, events for devices coming and going, sessions detaching and scripts posting messages, and option classes for spawning, attaching and loading scripts, plus a Frida.NET.Runtime package for each of Windows x64 and x86, macOS x64 and arm64, and Linux x86, x64, arm64, arm32 and mips64.
A word of caution before you build on it: the API is not stable. I’m in the middle of moving it to a fully async design, among other things, so expect it to change shape over the next few releases. We’ll publish to NuGet at a later point, so for now you build the packages from the repo. Thanks a lot for this, AeonLucid!
Fixes
Cassian433 found that a code slice taken from the free list was handed out with the reference count of zero it was left with when last freed, so its next unref took it to minus one and returned early instead of putting it back on the list. On hosts where we can allocate RWX memory, each slice could thus only be reused once, after which it leaked along with the pages backing it. The count is now reset on reuse, as for a slice from a fresh batch.
They also fixed SqliteStatement on V8, where REAL columns were fetched as 64-bit integers, so SELECT 1.5 stepped to 1 while QuickJS returned 1.5. Thanks for both!
Enjoy!
Changelog
- clr: Rewrite the bindings to .NET 9 on top of gir.core, generated from the GIR like our other bindings, for Windows, macOS and Linux. The API is not stable, and NuGet packages will follow later. Thanks @AeonLucid!
- codeallocator: Fix the reference count of reused slices, which leaked after one reuse on RWX-capable hosts. Thanks @Cassian433!
- gumjs: Read REAL columns as doubles on V8, matching QuickJS. Thanks @Cassian433!