{"id":"GHSA-r77w-qph3-7vf8","summary":"GraphQL-core: Denial of service via comment tokens bypassing the parser's max_tokens limit","details":"### Summary\n\nThe parser's `max_tokens` limit — the documented defence against oversized GraphQL\ndocuments — did not count comment tokens. A document padded with comments could\ntherefore pass a `max_tokens` check of any size while still allocating one `Token`\nobject per comment, all of which stay reachable for the lifetime of the parsed\ndocument. A 38 MB comment-padded payload allocated roughly 92 MB of additional\nresident memory with `max_tokens=5` in force.\n\n### Details\n\n`Lexer.lookahead()` reads tokens one at a time and skips over comments, returning only\nthe first non-comment token. Each skipped comment is nevertheless fully tokenized,\nallocated as a `Token` object, and linked into the persistent token list via\n`token.next` / `token.prev` — the chain that powers source-location reporting, and\nwhich is retained as long as the document is.\n\n`Parser.advance_lexer()` incremented the token counter only for the tokens the lexer\nactually returned. Comments were therefore invisible to the accounting, and\n`max_tokens` bounded only the *non-comment* tokens.\n\nThe practical consequence is a gap between what users reasonably believe `max_tokens`\nprotects and what it did protect. The limit was not broken for ordinary documents —\nreal tokens were still capped correctly — so the shortfall was not observable without\nspecifically testing comment padding.\n\nAs with the companion issue, the payload compresses extremely well (`# c\\n` repeated),\nso body-size limits and WAF inspection applied to the compressed request do not see\nits true size, while the framework decompresses it in memory before the parser is\nreached.\n\n### Proof of concept\n\n```python\nimport gc\nimport resource\nfrom graphql import parse\n\n# 100,000 comments, a 5-token query, and max_tokens=5 — accepted.\nquery = \"# c\\n\" * 100_000 + \"{ a }\"\nparse(query, max_tokens=5)          # no error raised\n\n# Memory amplification.\nbig = \"# padding comment used to amplify memory per token\\n\" * 500_000 + \"{ a }\"\ngc.collect()\nbefore = resource.getrusage(resource.RUSAGE_SELF).ru_maxrss\nparse(big, max_tokens=5)            # ~38 MB source\nafter = resource.getrusage(resource.RUSAGE_SELF).ru_maxrss\nprint(f\"RSS delta: +{(after - before) / 1024:.0f} MB\")\n```\n\nMeasured on GraphQL-core 3.2.8 / CPython 3.12 / Linux x86_64: the 391 KB payload\nparsed in 127 ms, and the 38.1 MB payload produced an RSS delta of **+92 MB** — about\n2.4× the source size, on top of the source string itself.\n\n### Impact\n\nDenial of service through memory exhaustion. A modest number of concurrent requests\ncan push a worker past a typical 256 MB or 512 MB container limit and trigger an\nOOM kill. No authentication is required, and no data is disclosed or modified.\n\nDepth limiters and complexity analyzers do not help: they run on the AST after parsing\nhas already succeeded, and the document contains only five real tokens.\n\n### Patches\n\nFixed in **3.2.12** and **3.3.0rc1**.\n\n`Parser.advance_lexer` now counts every token consumed, including the comment tokens\nthe lexer skips between the previously focused token and the new one, and including\nthe advance onto EOF so that trailing comments are covered. `max_tokens` therefore now\nbounds every `Token` the parser allocates.\n\nNote the behaviour change: **comments now count toward `max_tokens`.** A\ncomment-heavy document that was previously accepted may now be rejected. If you parse\ndocuments containing substantial comments — SDL files with documentation comments, for\ninstance — and you set `max_tokens`, you may need to raise the limit. The\n`token_count` attribute of a parsed document likewise now includes comments.\n\n### Workarounds\n\nIf you cannot upgrade, apply a request body-size limit to the **decompressed** body\nbefore the document reaches `parse()`.\n\nSuch a limit is worth keeping after upgrading as well: `max_tokens` is one guard among\nseveral and bounds the token count, not the overall cost of a request.\n\n### References\n\n- GraphQL specification § 2.1.4 — Comments\n\n### Credits\n\nReported by @BZHunt, who also reviewed the fixes\nbefore release. In memory of ZaX, who was the driving force behind this discovery.","aliases":["CVE-2026-75508"],"modified":"2026-10-22T16:45:04.297396287Z","published":"2026-10-22T16:27:14Z","database_specific":{"cwe_ids":["CWE-400","CWE-770"],"severity":"HIGH","github_reviewed":true,"github_reviewed_at":"2026-10-22T16:27:14Z","nvd_published_at":null},"references":[{"type":"WEB","url":"https://github.com/graphql-python/graphql-core/security/advisories/GHSA-r77w-qph3-7vf8"},{"type":"WEB","url":"https://github.com/graphql-python/graphql-core/commit/437dea102e92fdf05a2971ca76787c5a5bad875a"},{"type":"WEB","url":"https://github.com/graphql-python/graphql-core/commit/80ef435a1808fbbcc4aaac63ae76771a1c91002d"},{"type":"PACKAGE","url":"https://github.com/graphql-python/graphql-core"},{"type":"WEB","url":"https://github.com/graphql-python/graphql-core/releases/tag/v3.2.12"},{"type":"WEB","url":"https://github.com/graphql-python/graphql-core/releases/tag/v3.3.0rc1"}],"affected":[{"package":{"name":"graphql-core","ecosystem":"PyPI","purl":"pkg:pypi/graphql-core"},"ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"3.2.3"},{"fixed":"3.2.12"}]}],"versions":["3.2.10","3.2.11","3.2.3","3.2.4","3.2.5","3.2.6","3.2.7","3.2.8","3.2.9"],"database_specific":{"source":"https://github.com/github/advisory-database/blob/main/advisories/github-reviewed/2026/10/GHSA-r77w-qph3-7vf8/GHSA-r77w-qph3-7vf8.json"}},{"package":{"name":"graphql-core","ecosystem":"PyPI","purl":"pkg:pypi/graphql-core"},"ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"3.3.0a3"},{"fixed":"3.3.0rc1"}]}],"versions":["3.3.0a10","3.3.0a11","3.3.0a12","3.3.0a13","3.3.0a14","3.3.0a3","3.3.0a4","3.3.0a5","3.3.0a6","3.3.0a7","3.3.0a8","3.3.0a9","3.3.0b0","3.3.0b1","3.3.0b2","3.3.0rc0"],"database_specific":{"source":"https://github.com/github/advisory-database/blob/main/advisories/github-reviewed/2026/10/GHSA-r77w-qph3-7vf8/GHSA-r77w-qph3-7vf8.json"}}],"schema_version":"1.9.0","severity":[{"type":"CVSS_V3","score":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H"}]}