GHSA-8M8R-38JM-F355
Vulnerability from github – Published: 2026-07-28 21:44 – Updated: 2026-07-28 21:44Summary
When the Pydantic v2 output mode is in use, datamodel-code-generator reads a validators array from each model entry in the --extra-template-data file and synthesises a Pydantic @field_validator(...) decorator from each entry. The field names and the validator mode are interpolated into the decorator call wrapped in unescaped single quotes. A value containing ' breaks out of the string literal, letting an attacker emit an arbitrary positional Python expression into the decorator. The expression is evaluated at class-definition time, i.e. the moment the developer imports the generated module. This is the same trust model as the recently-published GHSA-wjv6-jcfj-mf9r (extras-file comment injection) but the impact is full RCE rather than a docstring leak.
Details
Sink: src/datamodel_code_generator/model/pydantic_v2/base_model.py, _process_validators (lines 405–449, at tag 0.60.1 / commit a321547e):
def _process_validators(self) -> None:
validators = self.extra_template_data.get("validators")
if not validators:
return
...
for validator in validators:
fields = validator.get("fields") or [validator.get("field")]
fields = [f for f in fields if f]
if not fields:
continue
function_path: str = validator["function"]
function_name = function_path.rsplit(".", 1)[-1]
mode = validator.get("mode", "after")
fields_str = ", ".join(f"'{f}'" for f in fields) # (A) UNESCAPED
...
mode_str = f"mode='{mode}'" # (B) UNESCAPED
prepared_validators.append({
"fields_str": fields_str,
"mode_str": mode_str,
"method_name": method_name,
"function_name": function_name,
"mode": mode,
})
self._additional_imports.append(Import.from_full_path(function_path)) # (C)
The strings from (A) and (B) flow verbatim into src/datamodel_code_generator/model/template/pydantic_v2/BaseModel.jinja2:
@field_validator({{ v.fields_str }}, {{ v.mode_str }})
There is no repr() call, no identifier check, and no quote-escaping.
Secondary sink at (C): Import.from_full_path(function_path) splits on the last . and emits from <prefix> import <suffix>. A ; in function_path therefore lands in the generated import line and runs as a statement at module load.
PoC
A self-contained one-file PoC is available here: https://gist.github.com/thegr1ffyn/34d5c647e74487ffb2be27c76dace2aa
Impact
Arbitrary code execution in the developer's interpreter / CI runner the moment the generated module is imported. Anyone who accepts a --extra-template-data file from an untrusted source is impacted:
- Pull requests adding or modifying project-local
*.template-data.json/.codegen.jsonfiles consumed by amake codegenrule or pre-commit hook. - Configuration snippets pasted from issue templates, READMEs, or third-party guides.
- Multi-tenant CI systems where one tenant's config file is read by another tenant's build.
Same blast radius as GHSA-wjv6-jcfj-mf9r, but silent RCE rather than a docstring leak — significantly higher impact under the same threat model.
Introduced in 0.52.1 by commit
a2b27562(Add --validators option for Pydantic v2 field validators).
Resolution
The fix validates validators entries with Pydantic models before rendering them. Field names must be valid non-keyword Python identifiers, function must be a dotted Python identifier path, and mode must be one of Pydantic's supported validator modes. The generated decorator arguments now render field names with repr() and mode with !r, so validated values are still emitted as Python string literals.
Remediation
Upgrade to datamodel-code-generator 0.60.2 or later.
This issue affects datamodel-code-generator versions >= 0.52.1, <= 0.60.1 and is fixed in 0.60.2.
Submitted by: Hamza Haroon (thegr1ffyn)
{
"affected": [
{
"database_specific": {
"last_known_affected_version_range": "\u003c= 0.60.1"
},
"package": {
"ecosystem": "PyPI",
"name": "datamodel-code-generator"
},
"ranges": [
{
"events": [
{
"introduced": "0.52.1"
},
{
"fixed": "0.60.2"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-54656"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": true,
"github_reviewed_at": "2026-07-28T21:44:09Z",
"nvd_published_at": null,
"severity": "HIGH"
},
"details": "### Summary\n\nWhen the Pydantic v2 output mode is in use, `datamodel-code-generator` reads a `validators` array from each model entry in the `--extra-template-data` file and synthesises a Pydantic `@field_validator(...)` decorator from each entry. The field names and the validator mode are interpolated into the decorator call wrapped in *unescaped* single quotes. A value containing `\u0027` breaks out of the string literal, letting an attacker emit an arbitrary positional Python expression into the decorator. The expression is evaluated at class-definition time, i.e. the moment the developer imports the generated module. This is the same trust model as the recently-published GHSA-wjv6-jcfj-mf9r (extras-file comment injection) but the impact is full RCE rather than a docstring leak.\n\n### Details\n\nSink: `src/datamodel_code_generator/model/pydantic_v2/base_model.py`, `_process_validators` (lines 405\u2013449, at tag `0.60.1` / commit `a321547e`):\n\n```python\ndef _process_validators(self) -\u003e None:\n validators = self.extra_template_data.get(\"validators\")\n if not validators:\n return\n ...\n for validator in validators:\n fields = validator.get(\"fields\") or [validator.get(\"field\")]\n fields = [f for f in fields if f]\n if not fields:\n continue\n function_path: str = validator[\"function\"]\n function_name = function_path.rsplit(\".\", 1)[-1]\n mode = validator.get(\"mode\", \"after\")\n fields_str = \", \".join(f\"\u0027{f}\u0027\" for f in fields) # (A) UNESCAPED\n ...\n mode_str = f\"mode=\u0027{mode}\u0027\" # (B) UNESCAPED\n prepared_validators.append({\n \"fields_str\": fields_str,\n \"mode_str\": mode_str,\n \"method_name\": method_name,\n \"function_name\": function_name,\n \"mode\": mode,\n })\n self._additional_imports.append(Import.from_full_path(function_path)) # (C)\n```\n\nThe strings from (A) and (B) flow verbatim into `src/datamodel_code_generator/model/template/pydantic_v2/BaseModel.jinja2`:\n\n```jinja\n@field_validator({{ v.fields_str }}, {{ v.mode_str }})\n```\n\nThere is no `repr()` call, no identifier check, and no quote-escaping.\n\nSecondary sink at (C): `Import.from_full_path(function_path)` splits on the last `.` and emits `from \u003cprefix\u003e import \u003csuffix\u003e`. A `;` in `function_path` therefore lands in the generated import line and runs as a statement at module load.\n\n### PoC\n\nA self-contained one-file PoC is available here: https://gist.github.com/thegr1ffyn/34d5c647e74487ffb2be27c76dace2aa\n\n### Impact\n\nArbitrary code execution in the developer\u0027s interpreter / CI runner the moment the generated module is imported. Anyone who accepts a `--extra-template-data` file from an untrusted source is impacted:\n\n- Pull requests adding or modifying project-local `*.template-data.json` / `.codegen.json` files consumed by a `make codegen` rule or pre-commit hook.\n- Configuration snippets pasted from issue templates, READMEs, or third-party guides.\n- Multi-tenant CI systems where one tenant\u0027s config file is read by another tenant\u0027s build.\n\nSame blast radius as GHSA-wjv6-jcfj-mf9r, but silent RCE rather than a docstring leak \u2014 significantly higher impact under the same threat model.\n\n\n \n\u003e Introduced in 0.52.1 by commit [`a2b27562`](https://github.com/koxudaxi/datamodel-code-generator/commit/a2b27562) (*Add --validators option for Pydantic v2 field validators*).\n\n### Resolution\n\nThe fix validates `validators` entries with Pydantic models before rendering them. Field names must be valid non-keyword Python identifiers, `function` must be a dotted Python identifier path, and `mode` must be one of Pydantic\u0027s supported validator modes. The generated decorator arguments now render field names with `repr()` and mode with `!r`, so validated values are still emitted as Python string literals.\n\n### Remediation\n\nUpgrade to `datamodel-code-generator` `0.60.2` or later.\n\nThis issue affects `datamodel-code-generator` versions `\u003e= 0.52.1, \u003c= 0.60.1` and is fixed in `0.60.2`.\n\nSubmitted by: Hamza Haroon (thegr1ffyn)",
"id": "GHSA-8m8r-38jm-f355",
"modified": "2026-07-28T21:44:09Z",
"published": "2026-07-28T21:44:09Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/koxudaxi/datamodel-code-generator/security/advisories/GHSA-8m8r-38jm-f355"
},
{
"type": "WEB",
"url": "https://github.com/koxudaxi/datamodel-code-generator/commit/a43d02906111a2fdcaf13ee5b62eb2da85376f19"
},
{
"type": "PACKAGE",
"url": "https://github.com/koxudaxi/datamodel-code-generator"
},
{
"type": "WEB",
"url": "https://github.com/koxudaxi/datamodel-code-generator/releases/tag/0.60.2"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "`datamodel-code-generator` vulnerable to code execution on import via unescaped `validators` entries in --extra-template-data"
}
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.