paulperegud / randmockery

A proof-of-concept that we can mock randomness.

Build status:

A proof of concept that we can mock randomness.

How does it work?

We use ptrace to get notified about any syscall the inferior makes. When the inferior enters or exits a syscall, a user-defined handler is executed.

This gives us the opportunity to replace the syscall return value/buffer with our own data.

Notes

Mocking vDSO syscalls is not supported out of the box. You need to preload a library using the -l switch. An example mocking library is located in the tests subdirectory.

Benchmarks

• cmp file1 file2 on 1.5G files yielded 8x slowdown (this is the pessimistic scenario of an application, which does mostly syscalls)
• running a compute-only job (calculating 1/(1+x) using a series for a 1000 values of x with double precision) showed virtually no slowndown

git dependencies

Unfortunately, we depend on git versions of some crates:

• nix-rust/nix: we need struct Pid

upstreaming

ptrace_mod.rs will be upstreamed and is currently under review nix-rust/nix#666