Skip to main content
Menu
Advisories

**GET /api/v1/cluster names every single-bucket type of every database the caller may open, with no per-type readability filter, and the new #11197 census sanctions the walk as internal

GHSA-48Q6-Q453-CM77Published Updated GitHub

Description

GET /api/v1/cluster names every single-bucket type of every database the caller may open, with no per-type readability filter, and the new #11197 census sanctions the walk as internal.

Where, at HEAD a0f3ff9198:

  • ha-raft/src/main/java/com/arcadedb/server/ha/raft/ClusterAlerts.java:1294, findSingleBucketTypes, which walks db.getSchema().getTypes() and returns the names of the types with one bucket.
  • ha-raft/src/main/java/com/arcadedb/server/ha/raft/ClusterAlerts.java:1247-1286, checkSingleBucketTypes, which puts those names into the alert's details.databases object, keyed by database, capped per database.
  • ha-raft/src/main/java/com/arcadedb/server/ha/raft/ClusterAlerts.java:246, scan calling it first.
  • ha-raft/src/main/java/com/arcadedb/server/ha/raft/GetClusterHandler.java:404-407, response.put("alerts", ClusterAlerts.scan(..., authorizedDatabases, ...)), i.e. the names reach the HTTP response body.
  • The only scoping is at :316, filterAuthorizedDatabases(user, ...), which is DATABASE level. The handler overrides nothing but mustExecuteOnWorkerThread and execute, so I found no server-admin gate on the route: any authenticated user who may open the database is told the names of its single-bucket types, including types they have no read access to.
  • The database handle the walk uses is server.getDatabase(dbName, false, false) (ClusterAlerts.java:1257), an operator handle with no bound user, so SecurityHelper.readableTypes would answer everything there even if it were called. Fixing it means carrying the requesting user into the scan, not swapping one call.

Why it belongs with this round: census/.../SchemaPresentationReadableWalkCensusTest.java lists this exact call site in ALLOWED as

Map.entry("server/ha/raft/ClusterAlerts#findSingleBucketTypes", "server-internal, answers no caller"),

so the reviewed set that #11197 produced records the opposite of what the call chain does. That makes it a mis-review of the new census rather than a hole in it, and it is the one ALLOWED justification in the sixty I sampled that I believe is false.

Refutation attempted: I looked for an admin gate on the route and found none; I checked that the alerts array is in the response body rather than a log, and it is (:404); I checked whether the walk is already filtered, and it is not. What I did NOT do is stand up an HA cluster with two users and observe the response, so this is a traced call chain, not a measured disclosure. Severity depends on whether type names are considered sensitive in a multi-tenant deployment, which is a judgement for whoever routes it; the engine's own position, from #11185 and #11197, is that a route must not describe a type to a principal that may not read it.

Our own deployments are exposed to this shape: the single-bucket alert fires in practice on stage, so the alert is not a dormant code path.

Related advisories

This issue is tracked together with GHSA-vpxw-wqc5-hvvj, which carries the CVE identifier for the group.

Affected versions

References

GitHub · ArcadeData/arcadedb