GHSA-FF32-FCP5-2G3W
Cross-tenant mail disruption via domain alias name collision
Texto original en inglés · Ver en español
CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:L/I:H/A:H
Fuente: GitHub
Ciclo de vida de la vulnerabilidad
| Fecha | Fase | Evento | Detalle |
|---|---|---|---|
| Remediación | Parche 2.11.1 | modoboa · -22 d desde divulgación | |
| Divulgación | Publicada | GHSA-FF32-FCP5-2G3W divulgada · CVSS 6.7 (GitHub) | |
| Monitoreo | Detectada | Ingresó al índice | |
| Monitoreo | Verificada | Última verificación contra las fuentes |
Summary
DomainAliasSerializer.validate_name() lowercases the submitted name, but does
not reject a name already used by a Domain. The serializer verifies
access to the alias target, not to the domain whose name is being claimed.
The same serializer is used by /api/v1/domainaliases/ and
/api/v2/domainaliases/, and the check is also missing on updates.
The DomainAlias CSV-import path already performs this cross-model check:
DomainAlias.from_csv() rejects a name already used by either a Domain or a
DomainAlias with Conflict. The reverse Domain CSV-import path is not symmetric:
Domain.from_csv() currently checks only Domain. Exact duplicate DomainAlias
names are normally handled by DRF's uniqueness validation; the missing boundary
is primarily the collision between DomainAlias.name and Domain.name after
normalization.
Details
I reproduced this on master at 984c276c1 (2026-10-03). Static inspection found
the core issue and the Alias signal behavior in every released 2.x tag from
2.0.0 through 2.11.0. The Amavis-specific signal behavior described below is
present from 2.5.0 through 2.11.0.
This does not require access to the victim domain. A normal member of the
DomainAdmins group cannot create domain aliases by default; the demonstrated
attacker is a member of the Resellers group with administrative access to one
controlled target domain and sufficient reseller and target-domain capacity
when administrative limits are enabled. An already-owned alias is enough for
the update variant described below; updating it does not recheck quota.
Proof of concept
With Amavis disabled, using the test fixture created by populate_database()
(which includes test.com and test2.com, with mailboxes on both domains):
- Create a user in the Resellers group.
- Give the user access only to test.com with Domain.add_admin() and give
- Authenticate as that user and, in a separate clean test run for each API
POST /api/v1/domainaliases/
{"name": "test2.com", "target": <test.com pk>}
POST /api/v2/domainaliases/
{"name": "test2.com", "target": <test.com pk>}
With Amavis disabled, each endpoint returns 201 in its clean test run. The
database then contains both the legitimate Domain(name="test2.com") and the
attacker-controlled DomainAlias(name="test2.com", target="test.com"). The
post-save handler also creates the internal alias @test2.com -> @test.com.
With the Amavis extension enabled, direct creation instead fails synchronously:
the post-save handler hits the unique constraint for Amavis' users.email, the
request returns HTTP 500, and the default-database writes are rolled back. This
is not a reliable mitigation: the API accepts creation under an unused name
followed by a rename to the victim domain:
POST /api/v1/domainaliases/
{"name": "parked.example", "target": <test.com pk>}
PATCH /api/v1/domainaliases/<pk>/
{"name": "test2.com"}
The PATCH returns 200 with Amavis enabled and creates the same cross-model
domain-name collision. The v2 route behaves the same way. Because rename does
not synchronize generated records, this path leaves the internal alias and
Amavis row under parked.example; it does not create @test2.com -> @test.com.
The redirect demonstrated below therefore applies to the direct-create
configuration, while the rename path establishes the disruption and deletion
effects described later.
Impact
I verified the direct-create case, with Amavis disabled, using Postfix 3.10.13
configured as described in doc-ng/installation/postfix.md:
virtual_mailbox_domains = proxy:sqlite:/etc/postfix/sql-domains.cf
virtual_alias_domains = proxy:sqlite:/etc/postfix/sql-domain-aliases.cf
virtual_alias_maps = proxy:sqlite:/etc/postfix/sql-aliases.cf
I used the SQLite equivalents of the queries in
modoboa/admin/postfix_maps.py and an attacker-controlled catch-all
@test.com -> user@test.com to make the rewrite observable.
After the collision, test2.com is returned as both a virtual mailbox domain
and a virtual alias domain. Postfix logs:
do not list domain test2.com in BOTH virtual_alias_domains and virtual_mailbox_domains
It treats the victim as an alias domain. The previously valid victim mailboxes
in the fixture no longer resolve:
<user@test2.com>: User unknown in virtual alias table
<admin@test2.com>: User unknown in virtual alias table
Addresses that fall through to the domain-wide alias lookup are rewritten
through the generated internal alias and routed into the reseller's domain:
<user@test.com> (expanded from <nobody@test2.com>): virtual-delivered
I checked the maps with postmap -q and verified routing with sendmail -bv. In a
separate SMTP recipient-verification test, I excluded the client from
mynetworks, enabled reject_unverified_recipient, and used a successful sink
transport. Postfix rejected an existing victim mailbox with 550 while accepting
an attacker-mapped victim address with 250, so that setting does not prevent
the demonstrated impact. The supplied main Postfix harness uses
virtual_transport=error:virtual-delivered as a routing sentinel instead of
LMTP, so it demonstrates selection of the attacker-controlled target rather
than final mailbox storage.
Update and deletion side effects
The model signal handlers make the update path more problematic:
- The post-save handler creates the generated @domain alias only when the
- The post-delete handler deletes every Alias whose address equals
- The Amavis pre-delete handler likewise deletes Users("@CURRENT_NAME"), even
This produces two distinct persistent side effects:
- Where direct creation succeeds, renaming the colliding DomainAlias away from
- After adding a victim-owned external @test2.com catch-all, the reseller can
Incluye detalles técnicos publicados por el investigador. Úselos solo para validar y corregir sus propios sistemas.
Desglose de métricas
Rangos afectados
Estado del parche por paquete
| Paquete / producto | Estado | Corregida en | Desde divulgación | Última versión |
|---|---|---|---|---|
pypimodoboa |
Corregida | 2.11.1 | -22 d | 2.13.0 no afectada |
2.11.1Arsenal público
Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.
https://github.com/modoboa/modoboa/security/advisories/GHSA-ff32-fcp5-2g3wLas referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.
