Vault Enterprise and Sentinel integration
Vault Enterprise license or HCP Vault Dedicated cluster required.
Vault Enterprise integrates HashiCorp Sentinel to provide a rich set of access control functionality. Because Vault is a security-focused product trusted with high-risk secrets and assets, and because of its default-deny stance, integration with Vault is implemented in a defense-in-depth fashion. This takes the form of multiple types of policies and a fixed evaluation order.
Policy types
Vault's policy system has been expanded to support three types of policies:
ACLs
- These are the traditional Vault policies and remain unchanged.Role Governing Policies (RGPs)
- RGPs are Sentinel policies that are tied to particular tokens, Identity entities, or Identity groups. They have access to a rich set of controls across various aspects of Vault.Endpoint Governing Policies (EGPs)
- EGPs are Sentinel policies that are tied to particular paths instead of tokens. They have access to as much request information as possible, but they can take effect even on unauthenticated paths, such as login paths.
Not every unauthenticated path supports EGPs. For instance, the paths related to root token generation cannot support EGPs because it's already the mechanism of last resort if, for instance, all clients are locked out of Vault due to misconfigured EGPs.
Like with ACLs, root tokens are not subject to Sentinel policy checks.
Sentinel execution should be considered to be significantly slower than normal ACL policy checking. If high performance is needed, testing should be performed appropriately when introducing Sentinel policies.
Policy enforcement levels
Sentinel policies have three enforcement levels to choose from.
Level | Description |
---|---|
advisory | The policy is allowed to fail. Can be used as a tool to educate new users. |
soft-mandatory | The policy must pass unless an override is specified. |
hard-mandatory | The policy must pass. |
Policy evaluation
Vault evaluates incoming requests against policies of all types that are applicable.
- If the request is unauthenticated, skip to evaluating the EGPs. Otherwise, evaluate the token's ACL policies. These must grant access; as always, a failure to be granted capabilities on a path via ACL policies denies the request.
- Evaluate RGPs attached to the client token. All policies must pass according to their enforcement level.
- Evaluate EGPs set on the requested path, and any prefix-matching EGPs set on less-specific paths, are evaluated. All policies must pass according to their enforcement level.
Any failure at any of these steps results in a denied request.
RGPs and namespaces
Policies, auth methods, secrets engines, and tokens are locked into the namespace they are created in. However, identity groups can pull in entities and groups from other namespaces.
Tip
Refer to the Set up entities and groups section of the Secure Multi-Tenancy with Namespaces tutorial for a step-by-step instruction.
The scenarios below describe the relevant changes in more detail.
Versions 1.15.0, 1.14.4, 1.13.8, and later
The training namespace is a child namespace of the education namespace. The "Sun Shine" entity created in the training namespace is a member of the "Tester" group which is defined in the education namespace. The group members inherit the group-level policy.
Versions 1.15.0-rc1, 1.14.3, 1.13.7, and earlier
The training namespace is a child namespace of the education namespace. The "Sun Shine" entity created in the education namespace is a member of the "Tester" group which is defined in the training namespace. The group members inherit the group-level policy.
While ACL policies and EGPs set rules on a specific path, an RGP does not specify a target path. RGPs are tied to tokens, identity entities, or identity groups that you can write rules without specifying a path.
What if the deny-all RGP in the training namespace looked like:
deny-all.sentinel
precond = rule { identity.entity.metadata.org_id is "A012345X"} main = rule when precond { false}
Vault checks the requesting token's entity metadata. If the org_id
metadata
exists and the value is A012345X
, the request gets denied because the
enforcement level is hard-mandatory. It does not matter if the request invokes a
path starts with /education
or /education/training
, or even /foo
because
there is no path associated with the deny-all RGP.
Policy overriding
Vault supports normal Sentinel overriding behavior. Requests to override can be
specified on the command line via the policy-override
flag or in HTTP
requests by setting the X-Vault-Policy-Override
header to true
.
Override requests are visible in Vault's audit log; in addition, override requests and their eventual status (whether they ended up being required) are logged as warnings in Vault's server logs.
MFA
Sentinel policies support the Identity-based MFA system in Vault Enterprise. Within a single request, multiple checks of any named MFA method will only trigger authentication behavior for that method once, regardless of whether its validity is checked via ACLs, RGPs, or EGPs.
EGPs can be used to require MFA on otherwise unauthenticated paths, such as login paths. On such paths, the request data will perform a lookahead to try to discover the appropriate Identity information to use for MFA. It may be necessary to pre-populate Identity entries or supply additional parameters with the request if you require more information to use MFA than the endpoint is able to glean from the original request alone.
Using Sentinel
Configuration
Sentinel policies can be configured via the sys/policies/rgp/
and
sys/policies/egp/
endpoints; see the
documentation for more information.
Once set, RGPs can be assigned to Identity entities and groups or to tokens just like ACL policies. As a result, they cannot share names with ACL policies.
When setting an EGP, a list of paths must be provided specifying on which paths
that EGP should take effect. Endpoints can have multiple distinct EGPs set on
them; all are evaluated for each request. Paths can use a glob character (*
)
as the last character of the path to perform a prefix match; a path that
consists only of a *
will apply to the root of the API. Since requests are
subject to an EGPs exactly matching the requested path and any glob EGPs
sitting further up the request path, an EGP with a path of *
will thus take
effect on all requests.
Properties and examples
See the Examples page for examples of Sentinel in action, and the Properties page for detailed property documentation.
Tutorial
Refer to the Sentinel Policies tutorial to learn how to author Sentinel policies in Vault.