Reports: Full Disclosure mailing list, 2026-10-27 (
https://seclists.org/fulldisclosure/2026/Oct/49) and 2026-10-28 (
https://seclists.org/fulldisclosure/2026/Oct/50)
Disposition: KNOWN-NON-FINDING (THREAT_MODEL.md section 11a, resource
consumption within a configured bound)
TLP:CLEAR

In short

This is not a vulnerability in Apache Struts. The heap a single request can
make the JSON plugin retain is bounded by struts.json.maxLength, and the
reports describe consumption inside that bound.

What the reports describe

The first post argues that struts.json.maxElements and struts.json.maxDepth
apply per array during parsing, so nested arrays multiply the total number
of elements a body can produce. The second post is a proof of concept: a
body just under the 2 MB input limit, shaped as deeply nested small arrays,
that the author measured at about 83 MB of heap per request, and a script
that sends it concurrently.

What the framework does with that input

- The JSON interceptor is not part of any default interceptor stack. An
application enables it per action or per package
(plugins/json/src/main/resources/struts-plugin.xml).
- When enabled, the body is read in fixed-size chunks and the request is
rejected as soon as the accumulated length exceeds struts.json.maxLength, 2
MB by default (JSONUtil.deserializeInput). The limit no longer depends on
line breaks; see "Deserialization limits" on
https://struts.apache.org/plugins/json/.
- Within that body, struts.json.maxElements bounds each array or object and
struts.json.maxDepth bounds nesting, as documented: the element limit is
per "single JSON array or object".
- The project reproduced the published shape: it parses in tens of
milliseconds and retains about 78 MB, roughly 40 bytes of heap per input
byte. One nesting level more is rejected by the depth limit; one byte more
is rejected by the length limit.

Why this is not a vulnerability

The threat model (section 8.7) requires the framework to bound what its own
code reads and retains per request, and it does: the ceiling is
struts.json.maxLength times a small constant. Section 9 disclaims
availability under generic load, and section 11a lists resource consumption
that stays proportionate to a bounded input as a known non-finding. S2-072
is unrelated: it concerned a body read that was not bounded at all.

What the project is doing

No change to the framework follows from these reports. A per-document
budget on parsed nodes, which would lower the constant, is being considered
as hardening.

What users should do

Nothing is required. Operators who expose an endpoint with the JSON
interceptor and want a smaller per-request ceiling can lower
struts.json.maxLength; the per-request heap ceiling is roughly 40 times
that value.

References

- https://struts.apache.org/plugins/json/ (Deserialization limits)
- https://github.com/apache/struts/blob/main/THREAT_MODEL.md (sections 8.7,
9, 11a, 13)
- https://github.com/apache/struts/blob/main/SECURITY.md
- https://struts.apache.org/security/

New evidence goes to [email protected]; questions about this
notice go to this list.

on behalf of the Apache Struts PMC

Reply via email to