# GHSA-HPHW-285H-WX93: This is a private follow-up to a consequence we should not have disclosed publicly

- Database: GitHub
- Severity: low (stated by ghsa, no CVSS score)
- Published: 2026-11-04
- Modified: 2026-11-04

## Affected versions

- com.arcadedb:arcadedb-redisw (Maven): through 26.10.2

## Description

This is a private follow-up to a consequence we should not have disclosed publicly.

We filed the underlying behaviour as public issue #10597 (an unauthenticated Redis connection turning every junk byte into two SEVERE log lines). That was our mistake: SECURITY.md asks for private reporting and the issue is security-shaped. We are not adding anything further to that public thread, and we are sending the consequence here instead.

## The base behaviour, re-measured at main 888b83c926

40,010 input bytes produced **6,186,478 bytes of log**. Counts 20006 / 20006 / 20000. No reply and no close on any of the three sockets. This matches the original measurement to within the timestamp text.

## The consequence we did not publish

The shipped logging configuration is a bounded rotating handler, 100 MB across 10 files. At the measured amplification an unauthenticated client needs roughly 6.5 MB of junk to rotate the entire history out.

That turns a write-amplification nuisance into **log destruction**: an attacker who wants to erase evidence of anything else on the server can do it from an unauthenticated socket, without needing any access to the database, and the volume required is small enough to be unremarkable as network traffic.

## What bounds it

The Redis plugin is not on by default. `SERVER_PLUGINS` defaults to empty, so this only reaches a deployment that has deliberately enabled the Redis wire. We have it enabled in ours, which is how we came to measure it.

## Honesty about what was measured

The amplification factor is measured. The rotation settings are read from the shipped configuration. **The arithmetic between the two is ours, and we did not run the eviction end to end.** We did not watch a 1 GB history actually roll out. If your shipped defaults differ from what we read, the conclusion changes and we would rather you tell us than take our number.

## The guarantee we would ask for

The fix for the amplification itself is in #10597 and we will not repeat it here. The additional ask is that whatever bounds the logging of unauthenticated protocol garbage should be a property of the pre-authentication path rather than of the Redis codec, so the next wire protocol added inherits it. A rate limit or a once-per-connection summary on the unauthenticated side of every listener would do it, and would be testable by asserting that N bytes of junk on any listener produce a bounded number of log records regardless of N.


## Related advisories

This issue is tracked together with [GHSA-9cx8-qjp8-hcmr](https://github.com/ArcadeData/arcadedb/security/advisories/GHSA-9cx8-qjp8-hcmr), which carries the CVE identifier for the group.

## References

- https://github.com/ArcadeData/arcadedb/security/advisories/GHSA-hphw-285h-wx93

Page: https://cve.now/GHSA-HPHW-285H-WX93
