Warning: Since tokens are considered opaque values, their structure is undocumented and subject to change. For these reasons, we do not recommend using tokens within your scripts or for automation since it can cause breakage.
Tokens are the core method for authentication within Vault. Tokens can be used directly or auth methods can be used to dynamically generate tokens based on external identities.
If you've gone through the getting started guide, you probably noticed that
vault server -dev (or
vault operator init for a non-dev server) outputs an
initial "root token." This is the first method of authentication for Vault.
It is also the only auth method that cannot be disabled.
As stated in the authentication concepts, all external authentication mechanisms, such as GitHub, map down to dynamically created tokens. These tokens have all the same properties as a normal manually created token.
Within Vault, tokens map to information. The most important information mapped to a token is a set of one or more attached policies. These policies control what the token holder is allowed to do within Vault. Other mapped information includes metadata that can be viewed and is added to the audit log, such as creation time, last renewal time, and more.
Read on for a deeper dive into token concepts.
As of Vault 1.0, there are two types of tokens:
service tokens and
tokens. A section near the bottom of this page contains detailed information
about their differences, but it is useful to understand other token concepts
first. The features in the following sections all apply to service tokens, and
their applicability to batch tokens is discussed later.
Often in documentation or in help channels, the "token store" is referenced.
This is the same as the
backend. This is a special
backend in that it is responsible for creating and storing tokens, and cannot
be disabled. It is also the only auth method that has no login
capability -- all actions require existing authenticated tokens.
Root tokens are tokens that have the
root policy attached to them. Root
tokens can do anything in Vault. Anything. In addition, they are the only
type of token within Vault that can be set to never expire without any renewal
needed. As a result, it is purposefully hard to create root tokens; in fact
there are only three ways to create root tokens:
- The initial root token generated at
vault operator inittime -- this token has no expiration
- By using another root token; a root token with an expiration cannot create a root token that never expires
- By using
vault operator generate-root(example) with the permission of a quorum of unseal key holders
Root tokens are useful in development but should be extremely carefully guarded
in production. In fact, the Vault team recommends that root tokens are only
used for just enough initial setup (usually, setting up auth methods
and policies necessary to allow administrators to acquire more limited tokens)
or in emergencies, and are revoked immediately after they are no longer needed.
If a new root token is needed, the
operator generate-root command and associated
API endpoint can be used to generate one on-the-fly.
It is also good security practice for there to be multiple eyes on a terminal whenever a root token is live. This way multiple people can verify as to the tasks performed with the root token, and that the token was revoked immediately after these tasks were completed.
Normally, when a token holder creates new tokens, these tokens will be created as children of the original token; tokens they create will be children of them; and so on. When a parent token is revoked, all of its child tokens -- and all of their leases -- are revoked as well. This ensures that a user cannot escape revocation by simply generating a never-ending tree of child tokens.
Often this behavior is not desired, so users with appropriate access can create
orphan tokens. These tokens have no parent -- they are the root of their own
token tree. These orphan tokens can be created:
writeaccess to the
- By having
rootaccess to the
auth/token/createand setting the
- Via token store roles
- By logging in with any other (non-
token) auth method
Users with appropriate permissions can also use the
endpoint, which revokes the given token but rather than revoke the rest of the
tree, it instead sets the tokens' immediate children to be orphans. Use with
When tokens are created, a token accessor is also created and returned. This accessor is a value that acts as a reference to a token and can only be used to perform limited actions:
- Look up a token's properties (not including the actual token ID)
- Look up a token's capabilities on a path
- Renew the token
- Revoke the token
The token making the call, not the token associated with the accessor, must have appropriate permissions for these functions.
There are many useful workflows around token accessors. As an example, a service that creates tokens on behalf of another service (such as the Nomad scheduler) can store the accessor correlated with a particular job ID. When the job is complete, the accessor can be used to instantly revoke the token given to the job and all of its leased credentials, limiting the chance that a bad actor will discover and use them.
Audit devices can optionally be set to not obfuscate token accessors in audit logs. This provides a way to quickly revoke tokens in case of an emergency. However, it also means that the audit logs can be used to perform a larger-scale denial of service attack.
Finally, the only way to "list tokens" is via the
command, which actually gives a list of token accessors. While this is still a
dangerous endpoint (since listing all of the accessors means that they can then
be used to revoke all tokens), it also provides a way to audit and revoke the
currently-active set of tokens.
Every non-root token has a time-to-live (TTL) associated with it, which is a current period of validity since either the token's creation time or last renewal time, whichever is more recent. (Root tokens may have a TTL associated, but the TTL may also be 0, indicating a token that never expires). After the current TTL is up, the token will no longer function -- it, and its associated leases, are revoked.
If the token is renewable, Vault can be asked to extend the token validity
vault token renew or the appropriate renewal endpoint. At this
time, various factors come into play. What happens depends upon whether the
token is a periodic token (available for creation by
sudo users, token
store roles, or some auth methods), has an explicit maximum TTL
attached, or neither.
In the general case, where there is neither a period nor explicit maximum TTL value set on the token, the token's lifetime since it was created will be compared to the maximum TTL. This maximum TTL value is dynamically generated and can change from renewal to renewal, so the value cannot be displayed when a token's information is looked up. It is based on a combination of factors:
- The system max TTL, which is 32 days but can be changed in Vault's configuration file.
- The max TTL set on a mount using mount tuning. This value is allowed to override the system max TTL -- it can be longer or shorter, and if set this value will be respected.
- A value suggested by the auth method that issued the token. This might be configured on a per-role, per-group, or per-user basis. This value is allowed to be less than the mount max TTL (or, if not set, the system max TTL), but it is not allowed to be longer.
Note that the values in (2) and (3) may change at any given time, which is why a final determination about the current allowed max TTL is made at renewal time using the current values. It is also why it is important to always ensure that the TTL returned from a renewal operation is within an allowed range; if this value is not extending, likely the TTL of the token cannot be extended past its current value and the client may want to reauthenticate and acquire a new token. However, outside of direct operator interaction, Vault will never revoke a token before the returned TTL has expired.
Tokens can have an explicit max TTL set on them. This value becomes a hard limit on the token's lifetime -- no matter what the values in (1), (2), and (3) from the general case are, the token cannot live past this explicitly-set value. This has an effect even when using periodic tokens to escape the normal TTL mechanism.
In some cases, having a token be revoked would be problematic -- for instance, if a long-running service needs to maintain its SQL connection pool over a long period of time. In this scenario, a periodic token can be used. Periodic tokens can be created in a few ways:
- By having
sudocapability or a
roottoken with the
- By using token store roles
- By using an auth method that supports issuing these, such as AppRole
At issue time, the TTL of a periodic token will be equal to the configured
period. At every renewal time, the TTL will be reset back to this configured
period, and as long as the token is successfully renewed within each of these
periods of time, it will never expire. Outside of
root tokens, it is
currently the only way for a token in Vault to have an unlimited lifetime.
The idea behind periodic tokens is that it is easy for systems and services to perform an action relatively frequently -- for instance, every two hours, or even every five minutes. Therefore, as long as a system is actively renewing this token -- in other words, as long as the system is alive -- the system is allowed to keep using the token and any associated leases. However, if the system stops renewing within this period (for instance, if it was shut down), the token will expire relatively quickly. It is good practice to keep this period as short as possible, and generally speaking it is not useful for humans to be given periodic tokens.
There are a few important things to know when using periodic tokens:
- When a periodic token is created via a token store role, the current value of the role's period setting will be used at renewal time
- A token with both a period and an explicit max TTL will act like a periodic token but will be revoked when the explicit max TTL is reached
Some tokens are able to be bound to CIDR(s) that restrict the range of client IPs allowed to use them. These affect all tokens except for non-expiring root tokens (those with a TTL of zero). If a root token has an expiration, it also is affected by CIDR-binding.
There are currently two types of tokens.
Service tokens are what users will generally think of as "normal" Vault tokens. They support all features, such as renewal, revocation, creating child tokens, and more. They are correspondingly heavyweight to create and track.
Batch tokens are encrypted blobs that carry enough information for them to be used for Vault actions, but they require no storage on disk to track them. As a result they are extremely lightweight and scalable, but lack most of the flexibility and features of service tokens.
This reference chart describes the difference in behavior between service and batch tokens.
|Service Tokens||Batch Tokens|
|Can Be Root Tokens||Yes||No|
|Can Create Child Tokens||Yes||No|
|Can be Renewable||Yes||No|
|Can be Periodic||Yes||No|
|Can have Explicit Max TTL||Yes||No (always uses a fixed TTL)|
|Revoked with Parent (if not orphan)||Yes||Stops Working|
|Dynamic Secrets Lease Assignment||Self||Parent (if not orphan)|
|Can be Used Across Performance Replication Clusters||No||Yes (if orphan)|
|Creation Scales with Performance Standby Node Count||No||Yes|
|Cost||Heavyweight; multiple storage writes per token creation||Lightweight; no storage cost for token creation|
Leases created by service tokens (including child tokens' leases) are tracked along with the service token and revoked when the token expires.
Leases created by batch tokens are constrained to the remaining TTL of the batch tokens and, if the batch token is not an orphan, are tracked by the parent. They are revoked when the batch token's TTL expires, or when the batch token's parent is revoked (at which point the batch token is also denied access to Vault).
As a corollary, batch tokens can be used across performance replication clusters, but only if they are orphan, since non-orphan tokens will not be able to ensure the validity of the parent token.