GHSA-3M6Q-JJ5J-38C9
Vulnerability from github – Published: 2026-06-19 19:36 – Updated: 2026-06-19 19:36Summary
Oj::Doc#each_child, when invoked recursively over a deeply nested JSON
document, overflows a fixed-size stack buffer and aborts the process. This is a
denial of service reachable from untrusted JSON.
Details
Two-step chain in ext/oj/fast.c:
-
doc_each_child(~line 1501) incrementsdoc->wherepast thewhere_path[MAX_STACK = 100]array with no bounds check, and never restores it (doc->where--is missing). Callingeach_childrecursively from inside the yield block therefore drivesdoc->wherebeyond the array. -
On the next entry (~line 1478) the function copies the path into a stack-local buffer:
c
Leaf save_path[MAX_STACK]; // 800-byte stack buffer
size_t wlen = doc->where - doc->where_path;
if (0 < wlen) {
memcpy(save_path, doc->where_path, sizeof(Leaf) * (wlen + 1));
}
When the previous recursive call left doc->where past where_path[100],
wlen exceeds MAX_STACK and the memcpy overflows save_path on the C
stack.
The Oj::Doc parser imposes no JSON nesting-depth limit (it relies on a
C-stack pressure check), so deeply nested attacker input reaches this path.
Proof of Concept
require 'oj'
depth = 200
payload = '[' * depth + '1' + ']' * depth
Oj::Doc.open(payload) do |doc|
r = lambda { doc.each_child { |_| r.call } }
r.call
end
Recursion depth <= 99 iterates normally; depth >= 101 aborts. lldb backtrace
on the affected build (ruby 3.3.8 / arm64-darwin24):
SIGABRT
#2 __abort
#3 __stack_chk_fail
#4 doc_each_child (oj.bundle, fast.c)
Impact
Reliable denial of service: any endpoint that calls
Oj::Doc.open(untrusted) { |d| d.each_child ... } recursively can be crashed
with a small deeply-nested payload. On builds with a stack protector (the
default, -fstack-protector-strong) the canary aborts the process before the
saved return address is used. The Step-1 heap OOB writes into struct _doc
fields do occur, but are masked in practice because the Step-2 stack overflow
crashes first; turning them into anything beyond a crash has not been
demonstrated.
Patches
Fixed in 3.17.3: doc_each_child now bounds-checks before incrementing
doc->where (raising Oj::DepthError) and restores doc->where after the
loop, matching the existing each_leaf pattern. Verified on the fixed build:
depth >= 101 raises a clean Oj::DepthError instead of aborting.
Credit
Reported by Zac Wang (@7a6163).
{
"affected": [
{
"package": {
"ecosystem": "RubyGems",
"name": "oj"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "3.17.3"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-54592"
],
"database_specific": {
"cwe_ids": [
"CWE-125",
"CWE-787"
],
"github_reviewed": true,
"github_reviewed_at": "2026-06-19T19:36:28Z",
"nvd_published_at": null,
"severity": "HIGH"
},
"details": "### Summary\n\n`Oj::Doc#each_child`, when invoked recursively over a deeply nested JSON\ndocument, overflows a fixed-size stack buffer and aborts the process. This is a\ndenial of service reachable from untrusted JSON.\n\n### Details\n\nTwo-step chain in `ext/oj/fast.c`:\n\n1. **`doc_each_child` (~line 1501)** increments `doc-\u003ewhere` past the\n `where_path[MAX_STACK = 100]` array with no bounds check, and never restores\n it (`doc-\u003ewhere--` is missing). Calling `each_child` recursively from inside\n the yield block therefore drives `doc-\u003ewhere` beyond the array.\n\n2. **On the next entry (~line 1478)** the function copies the path into a\n stack-local buffer:\n\n ```c\n Leaf save_path[MAX_STACK]; // 800-byte stack buffer\n size_t wlen = doc-\u003ewhere - doc-\u003ewhere_path;\n if (0 \u003c wlen) {\n memcpy(save_path, doc-\u003ewhere_path, sizeof(Leaf) * (wlen + 1));\n }\n ```\n\n When the previous recursive call left `doc-\u003ewhere` past `where_path[100]`,\n `wlen` exceeds `MAX_STACK` and the `memcpy` overflows `save_path` on the C\n stack.\n\nThe `Oj::Doc` parser imposes no JSON nesting-depth limit (it relies on a\nC-stack pressure check), so deeply nested attacker input reaches this path.\n\n### Proof of Concept\n\n```ruby\nrequire \u0027oj\u0027\ndepth = 200\npayload = \u0027[\u0027 * depth + \u00271\u0027 + \u0027]\u0027 * depth\nOj::Doc.open(payload) do |doc|\n r = lambda { doc.each_child { |_| r.call } }\n r.call\nend\n```\n\nRecursion depth \u003c= 99 iterates normally; depth \u003e= 101 aborts. lldb backtrace\non the affected build (`ruby 3.3.8 / arm64-darwin24`):\n\n```\nSIGABRT\n#2 __abort\n#3 __stack_chk_fail\n#4 doc_each_child (oj.bundle, fast.c)\n```\n\n### Impact\n\nReliable denial of service: any endpoint that calls\n`Oj::Doc.open(untrusted) { |d| d.each_child ... }` recursively can be crashed\nwith a small deeply-nested payload. On builds with a stack protector (the\ndefault, `-fstack-protector-strong`) the canary aborts the process before the\nsaved return address is used. The Step-1 heap OOB writes into `struct _doc`\nfields do occur, but are masked in practice because the Step-2 stack overflow\ncrashes first; turning them into anything beyond a crash has not been\ndemonstrated.\n\n### Patches\n\nFixed in **3.17.3**: `doc_each_child` now bounds-checks before incrementing\n`doc-\u003ewhere` (raising `Oj::DepthError`) and restores `doc-\u003ewhere` after the\nloop, matching the existing `each_leaf` pattern. Verified on the fixed build:\ndepth \u003e= 101 raises a clean `Oj::DepthError` instead of aborting.\n\n### Credit\n\nReported by Zac Wang (@7a6163).",
"id": "GHSA-3m6q-jj5j-38c9",
"modified": "2026-06-19T19:36:28Z",
"published": "2026-06-19T19:36:28Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/ohler55/oj/security/advisories/GHSA-3m6q-jj5j-38c9"
},
{
"type": "PACKAGE",
"url": "https://github.com/ohler55/oj"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
}
],
"summary": "Oj: Stack Buffer Overflow in Oj::Doc#each_child via Deeply Nested Input"
}
Drag nodes to arrange them. Double-click a related vulnerability to open it, or a source to visit its link. The Physics button on the left tunes the layout and the simulation; the legend at the bottom left hides or shows nodes by kind and by sighting type.
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.
Vulnerability timeline
Sightings and recorded dates, grouped by day · UTC
How durations are calculated
Following the VTEM framework (Black Hat USA 2015), time to patch starts when the vendor or a coordinator was informed, total time to patch at discovery, and disclosure to patch at public disclosure; vendor response time runs from the report to its acknowledgement. Time to exploit and time unpatched start at the earliest discovery or report, falling back to public disclosure. Unpatched time ends at the first recorded patch, or the calculation time when no patch is recorded. Exploitation to patch shows the interval between those two events; a published proof of concept is a public exploit, not exploitation.
Each result shows its source dates. A negative interval means the end event came first; when date-only inputs leave the order open, the result says so. Date-only inputs are approximate. The earliest evidence wins: KEV report dates and sightings are labelled as such. Records describing another vulnerability are listed but never calculated from, and a PoC alone does not establish exploitation. Missing patch evidence does not prove that no patch exists. Filters below change the event list only.
Loading timeline…
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.
Browse all ATT&CK techniques and the vulnerabilities related to each.
Related by attack behaviour
Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.