# research/ Working material, not part of the shipped tool. Everything here is either ignored by git or explained below. ## Provenance of `tests/golden/bodies.json` The golden hashes are **generated, never typed**. They are produced from the public reference chain (`9bie/php2shell-bypass-aslr-exp`) so that `poc/bodies.py` can be proven byte-identical to the code the primitives were published with, without vendoring that code. ```bash mkdir -p research/upstream git clone --depth 1 https://github.com/9bie/php2shell-bypass-aslr-exp.git research/upstream python research/capture_golden.py python -m unittest tests.test_bodies_parity -v ``` `capture_golden.py` runs the reference generators with the exact parameters recorded in `poc/profiles.py` (`a1` with `--big 78`, `a2` via `build_body(880, 64, 115, …)`, `b1` with the 17-item leak spec and `--fake-len 0xC8000 --tag`), hashes the three bodies and writes `tests/golden/bodies.json`. If a golden mismatches, the fix belongs in `poc/bodies.py` — never in the golden file. ## Ignored artefacts | Path | Why it is not committed | |---|---| | `research/upstream/` | third-party clone, reference only; nothing in `poc/` imports it | | `research/control/` | the control-test copy with one literal widened (see `NOTES.md` §10) | | `research/libphp.so` | 21 MB binary from the target image; needed only for `tests/test_elf.py` | | `research/captures/*` | multi-megabyte HTTP responses and generated bodies | Consequences: on a fresh clone, `tests/test_elf.py` and the real-capture test in `tests/test_leak_parse.py` skip themselves. To run them, copy the module out of the target: ```bash docker cp p2s-lab:/usr/lib/apache2/modules/libphp.so research/libphp.so ``` Everything else — the payload builders, the leak parser, framing, profile logic, CLI parsing — runs with no Docker, no network and no binary, which is what CI exercises.