Commit Graph
2 Commits
Author SHA1 Message Date
48f650ad04 test: make the trust_remote_code scan see every truthy value and run without git (#4785)
## Description

`tests/test_no_trust_remote_code.py` scans the tracked tree so that
nothing re-enables HuggingFace remote code execution. Its regex captured
the assigned value only when it started with a letter
(`([A-Za-z_][\w.]*)`), but the library tests the flag for truthiness, so
these also enable it and were not seen: `trust_remote_code=1`,
`trust_remote_code=(True)`, `trust_remote_code="yes"`, `setattr(cfg,
"trust_remote_code", True)` and `kwargs.setdefault("trust_remote_code",
True)`. The test also called `pytest.skip` when `git ls-files` failed
(for example in an exported tarball), so the guard silently stopped
running.

This PR widens the value rule (a quoted string, or a name or number,
with an optional leading `(`, treating anything except the exact name
`False` as enabling the flag), adds a second pattern for the `"name",
value` form, matches across a line break, and falls back to scanning
`headroom/**/*.py` when git is unavailable instead of skipping. The scan
of the current tree finds no offender with the wider rules.

Part of #4783

## Type of Change

- [x] Bug fix (non-breaking change that fixes an issue)
- [ ] New feature (non-breaking change that adds functionality)
- [ ] Breaking change (fix or feature that would cause existing
functionality to change)
- [ ] Documentation update
- [ ] Performance improvement
- [ ] Code refactoring (no functional changes)

## Changes Made

- `tests/test_no_trust_remote_code.py`: `_VALUE`, a wider `ASSIGNMENT`,
a new `NAMED_ARGUMENT` limited to
`setattr`/`getattr`/`setdefault`/`get`/`pop` calls, a `_values()` helper
that reports `(line, value)`, a git-less fallback in `_tracked_files()`,
and 17 new cases: calls that do not set the flag (`log.info("...",
True)`) must be ignored, every truthy shape must be flagged, a value on
the next line must be flagged, and an explicit `False` (including
`(False)`, `kwargs["..."] = False` and `setattr(..., False)`) must be
accepted.

## Testing

- [x] Unit tests pass (`pytest`)
- [x] Linting passes (`ruff check .`)
- [ ] Type checking passes (`mypy headroom`)
- [x] New tests added for new functionality
- [x] Manual testing performed

`mypy headroom` was not run (no change under `headroom/`). `ruff check`
and `ruff format` (0.17.3) are clean.

### Test Output

The old regex on the shapes the old test never saw (`re.findall` with
the old pattern):

```text
old regex matches=True  trust_remote_code=True
old regex matches=False trust_remote_code=1
old regex matches=False trust_remote_code=(True)
old regex matches=False trust_remote_code="yes"
old regex matches=False setattr(cfg, "trust_remote_code", True)
old regex matches=False kwargs.setdefault("trust_remote_code", True)
```

The new tests with the old scan rule (value must start with a letter, no
`"name", value` pattern):

```text
6 failed, 10 passed   (the =1, ="yes", mapping =1, setattr, setdefault and setattr(..., False) cases fail)
```

The new tests with this branch's scan:

```text
tests/test_no_trust_remote_code.py   19 passed   (the tree scan finds no offender)
same file run from a copy of the tree without .git   19 passed, 0 skipped
```

## Real Behavior Proof

- Environment: Windows 11, Python 3.12, pytest 9.2.0, `main` at
`3c66c91e`.
- Exact command / steps: `python -m pytest
tests/test_no_trust_remote_code.py`; then the same file from a copy of
the repository without `.git`; then the new tests with the scan logic
swapped for the old rule.
- Observed result: as above. The tree scan passes on the current tree
with the wider rules, so no existing code needs to change.
- Not tested: CI, and a deliberately injected `trust_remote_code=1` in a
real module (the shapes are covered by the unit cases instead).

## Runtime Rollout Safety

- Rollout-managed feature(s): none; test only.
- Minimum rollout channel: n/a
- Stable/default behavior changed: no for the library and proxy. The
scan is stricter and no longer skips without git.
- Kill switch / disable path: n/a
- Unsafe override required: no
- Qualification impact: none
- Rollback path: revert this commit.

## Review Readiness

- [x] I have performed a self-review
- [x] This PR is ready for human review

## Checklist

- [x] My code follows the project's style guidelines
- [x] I have performed a self-review of my code
- [x] I have commented my code, particularly in hard-to-understand areas
- [ ] I have made corresponding changes to the documentation
- [x] My changes generate no new warnings
- [x] I have added tests that prove my fix is effective or that my
feature works
- [x] New and existing unit tests pass locally with my changes (this
file; the whole suite was not run)
- [x] I did **not** edit `CHANGELOG.md` — it is generated by
release-please from my Conventional Commit PR title (a CI guard enforces
this)

## Additional Notes

- A value such as `"False"` (a string) is truthy in Python, so the scan
flags it; I think that is right for this guard.
- The git-less fallback scans only `headroom/`: an exported tarball has
no other tracked-file list.

AI assistance: this change, its tests and the runs above were done with
Claude (Sonnet 5.5), operated by me (HetCreep).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- devin-review-badge-begin -->

---

<a href="https://app.devin.ai/review/headroomlabs-ai/headroom/pull/4785"
target="_blank"><picture><source media="(prefers-color-scheme: dark)"
srcset="https://static.devin.ai/assets/gh-devin-review-dark.svg?v=4"><img
src="https://static.devin.ai/assets/gh-devin-review-light.svg?v=4"
alt="Devin Review"></picture></a>
<!-- devin-review-badge-end -->

---------

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-authored-by: JerrettDavis <mxjerrett@gmail.com>
2026-11-16 23:02:31 -06:00
Tejas ChopraandClaude Opus 5 34da21e89e fix(tokenizers): disable trust_remote_code and allowlist HF tokenizer repos (#3723)
## A-1 (S1, remote code execution): HF tokenizer loader ran caller-named
repositories

`headroom/tokenizers/huggingface.py` called
`AutoTokenizer.from_pretrained(...,
trust_remote_code=True)` on an identifier that traces back to a proxied
request body.
`get_tokenizer_name()` ended in *"assume the model name is the tokenizer
name"*, so an
unrecognised `model` field was handed straight to the loader, which
resolves it against the
HuggingFace Hub and — with `trust_remote_code=True` — executes whatever
Python that
repository ships. Publish a repo, name it in a request body, get code
execution in the proxy
process. Nothing constrained the identifier.

### The fix

1. **`trust_remote_code=False` on every path**, passed explicitly rather
than leaning on a
   library default upstream can change.
2. **`ALLOWED_TOKENIZER_REPOS`**, derived from the mappings the product
actually ships
(`MODEL_TO_TOKENIZER.values()`, so adding a family allows it and nothing
drifts in).
`_load_tokenizer` checks it *before* importing transformers: an
unallowlisted name is
refused outright and never becomes a Hub lookup. It fails closed to
character estimation
rather than silently substituting a different vocabulary under the
caller's name.
3. **`get_tokenizer_name()` fails closed.** The fallthrough now returns
`DEFAULT_TOKENIZER`
instead of echoing the caller's string. A full repo id is accepted only
when it is on the
allowlist, and then the value returned is *our* canonical spelling,
never the caller's.

Two layers on purpose: the resolver is the normal path, the loader is
the chokepoint —
`_load_tokenizer` is importable and callable with an arbitrary string.

`benchmarks/comprehensive_eval.py` also passed `trust_remote_code=True`
to
`datasets.load_dataset`. The brief only named the two tokenizer call
sites; that third one is
fixed here too, and the tree-wide test below now holds the line for all
of them.

### Operator escape hatch

`HEADROOM_HF_TOKENIZER_ALLOWLIST` (comma-separated repo ids) lets an
operator admit private
or self-hosted tokenizer repositories. It is deliberately
environment-only — server-side
configuration, never anything a client can reach through a request body
— and repos added
there are still loaded with `trust_remote_code=False`.

### Default change a customer could depend on

An unrecognised model previously resolved to *itself* as a tokenizer
name. In practice a
made-up id failed to resolve on the Hub and counting fell back to
estimation, so most
deployments see no change. Two cases do change:

- A caller who relied on passing an **arbitrary HF repo id** not in
`MODEL_TO_TOKENIZER`
(e.g. a private fine-tune) and getting exact counts now gets the default
tokenizer
  instead. Remedy: add the repo to `HEADROOM_HF_TOKENIZER_ALLOWLIST`.
- Unknown models now resolve to `DEFAULT_TOKENIZER` (Llama 3), so a
deployment with that
vocabulary in its HF cache gets a real BPE count where it previously got
estimation.

Repo ids that already resolved correctly are unchanged. Alias and prefix
matching run first
and are untouched, so existing precedence — including
`mistralai/Mixtral-8x7B-v0.1` hitting
the `mistral` prefix — resolves exactly as before; there is an explicit
test pinning that so
a future reorder has to acknowledge it.

### Tests

- `tests/test_huggingface_tokenizer_allowlist.py` — a crafted `model`
never reaches
`from_pretrained`; every resolution lands inside the shipped set; the
loader refuses an
unallowlisted repo without touching the Hub; `trust_remote_code` is
`False` on both the
cache-only and bounded-network calls; every shipped alias still resolves
to its mapping;
the operator env hatch admits only what it names; and, since matching is
case-insensitive, the string handed to the Hub is the allowlist's own
spelling
  rather than the caller's variant.
- `tests/test_no_trust_remote_code.py` — tree-wide scan over every
git-tracked file
  asserting every `trust_remote_code` assignment is `False`.
- `tests/test_huggingface_tokenizer_timeout.py` — the existing GH #1701
timeout tests used
placeholder repo ids (`some/model`, `slow/model`, `offline/model`) which
the allowlist now
rejects before the timeout logic runs, short-circuiting what those tests
exist to prove.
Switched to real shipped repo ids so they still exercise the
bounded-load path.

### Fail-before / pass-after evidence

Method: commit, `git checkout origin/main --
headroom/tokenizers/huggingface.py
benchmarks/comprehensive_eval.py` (source reverted, new tests kept),
run, then
`git checkout HEAD -- ...` to restore.

**Before (source reverted, tests kept) — 10 failed, 4 passed:**

```
tests/test_huggingface_tokenizer_allowlist.py FFFF..F..FFF               [ 85%]
tests/test_no_trust_remote_code.py FF                                    [100%]

=================================== FAILURES ===================================
E   AssertionError: attacker-named repository was handed to from_pretrained
    assert 'attacker-controlled/headroom-rce-poc' not in ['attacker-controlled/headroom-rce-poc']

E   AssertionError: 'attacker-controlled/headroom-rce-poc' escaped the allowlist
    assert 'attacker-controlled/headroom-rce-poc' in frozenset({'01-ai/Yi-1.5-6B', ...})
     +  where 'attacker-controlled/headroom-rce-poc' = get_tokenizer_name('attacker-controlled/headroom-rce-poc')

E   AttributeError: module 'headroom.tokenizers.huggingface' has no attribute 'DEFAULT_TOKENIZER'

E   AssertionError: trust_remote_code must be False everywhere — it executes repository-supplied Python in this process:
      benchmarks/comprehensive_eval.py:44: trust_remote_code=True,
      headroom/tokenizers/huggingface.py:162: trust_remote_code=True,
      headroom/tokenizers/huggingface.py:184: trust_remote_code=True,

E   AssertionError: assert {'True'} == {'False'}

=========================== short test summary info ============================
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_crafted_model_never_reaches_the_loader
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_crafted_model_resolves_to_a_shipped_tokenizer
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_crafted_model_resolves_to_the_declared_default
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_loader_fetches_the_allowlists_own_spelling
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_remote_code_disabled_on_cache_and_network_paths
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_repo_id_matching_is_case_insensitive_but_returns_our_spelling
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_operator_env_allowlist_admits_a_private_repo
FAILED tests/test_huggingface_tokenizer_allowlist.py::test_empty_operator_env_admits_nothing_extra
FAILED tests/test_no_trust_remote_code.py::test_remote_code_is_never_enabled_anywhere_in_the_tree
FAILED tests/test_no_trust_remote_code.py::test_the_tokenizer_loader_sets_the_flag_explicitly
========================= 10 failed, 4 passed in 1.47s =========================
```

**After (source restored):**

```
tests/test_huggingface_tokenizer_allowlist.py ............               [ 54%]
tests/test_no_trust_remote_code.py ..                                    [ 63%]
tests/test_huggingface_tokenizer_timeout.py ........                     [100%]

============================== 22 passed in 2.73s ==============================
```

Wider tokenizer suites (`test_tokenizers.py`, `test_tokenizer.py`,
`test_tokenizers/`,
`test_tokenizer_encoding_resolution.py`,
`test_tokenizer_selection_coverage.py`,
`test_provider_tokenizer_one_ruler.py`,
`test_tokenizer_count_offload.py`,
`test_strands_tokenizer.py`,
`test_compress_route_tokenizer_by_model.py`,
`test_evals_cjk_tokenization.py`): **166 passed, 14 skipped**.

Lint: `ruff 0.16.7 check .` and `format --check .` both clean.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-23 20:20:56 +02:00