azure-postgres-mysql-admin-exfiltrate
Admin read access to the database (CanReadData) enables full exfiltration via pg_dump (PostgreSQL) or mysqldump (MySQL).
match
A conjunctive graph pattern. Variables (?x) bind node ids; every clause must hold.
where
node_class(?server) == Data and node_type(?server) ==
RelationalDatabase
node_type(?server) matched provider_type in ['Microsoft.DBforPostgreSQL/flexibleServers', 'Microsoft.DBforMySQL/flexibleServers']
emit
| source type | Identity |
|---|---|
| target type | RelationalDatabase |
| source | ?principal |
| target | ?server |
| conditions | network_reachability |
| state logic | Inherits the state of the upstream CanReadData edge (per edge_state_model weakest-contributor rule). If CanReadData is ACTIVE, CanExfiltrate is ACTIVE (the attacker can run pg_dump / mysqldump). If CanReadData is CONDITIONAL, CanExfiltrate is also CONDITIONAL with the same preconditions. Exfiltration is gated by the data-plane credential (native admin password or Entra token) and network reachability — already captured in the upstream CanReadData edge. No additional ARM permission check applies. |
Narrative
{principal.name} can read all data in {server.name} (CanReadData). Using native database tools (pg_dump for PostgreSQL, mysqldump for MySQL), the attacker can exfiltrate the entire database contents, including user data, application secrets, and embedded credentials. Exfiltration is a native database-engine capability (not gated by ARM RBAC) and requires only the data-plane credential and network reach.
Raw rule rules/derived/azure/postgres-mysql.yaml
id: azure-postgres-mysql-admin-exfiltrate
emits: CanExfiltrate
description: Admin read access to the database (CanReadData) enables full exfiltration via pg_dump (PostgreSQL)
or mysqldump (MySQL).
match:
- - principal: null
- CanReadData
- server: null
where:
- node_class(?server) == Data and node_type(?server) == RelationalDatabase
- node_type(?server) matched provider_type in ['Microsoft.DBforPostgreSQL/flexibleServers', 'Microsoft.DBforMySQL/flexibleServers']
emit:
source_type: Identity
target_type: RelationalDatabase
source: ?principal
target: ?server
permissions: []
conditions:
- network_reachability
state_logic: "Inherits the state of the upstream CanReadData edge (per edge_state_model weakest-contributor\
\ rule). If CanReadData is ACTIVE, CanExfiltrate is ACTIVE (the attacker can run pg_dump / mysqldump).\
\ If CanReadData is CONDITIONAL, CanExfiltrate is also CONDITIONAL with the same preconditions. Exfiltration\
\ is gated by the data-plane credential (native admin password or Entra token) and network reachability\
\ \u2014 already captured in the upstream CanReadData edge. No additional ARM permission check applies."
confidence: '0.95'
derived_from:
- 'CanReadData edge: source=?principal, target=?server, rule=(azure-postgres-mysql-admin-reset-readdata
or azure-postgres-mysql-entra-admin-readdata)'
false_positive_note: "This is a direct consequence of CanReadData \u2014 do NOT emit as ACTIVE unless\
\ CanReadData is ACTIVE or CONDITIONAL at the upstream rule. Network reachability is already encoded\
\ in the upstream CanReadData; do not add redundant gates. Exfiltration via pg_dump (PostgreSQL) or\
\ mysqldump (MySQL) is a native database-engine capability \u2014 the attacker must hold valid credentials\
\ (native admin password or Entra token) to run these tools. There is no ARM permission check at the\
\ data plane. Do NOT list ARM permissions as blockers on this edge."
narrative: '{principal.name} can read all data in {server.name} (CanReadData). Using native database
tools (pg_dump for PostgreSQL, mysqldump for MySQL), the attacker can exfiltrate the entire database
contents, including user data, application secrets, and embedded credentials. Exfiltration is a native
database-engine capability (not gated by ARM RBAC) and requires only the data-plane credential and
network reach.'