»AWS Auth Method
The aws auth method provides an automated mechanism to retrieve a Vault token
for IAM principals and AWS EC2 instances. Unlike most Vault auth methods, this
method does not require manual first-deploying, or provisioning
security-sensitive credentials (tokens, username/password, client certificates,
etc), by operators under many circumstances.
»Authentication Workflow
There are two authentication types present in the aws auth method: iam and
ec2.
With the iam method, a special AWS request signed with AWS IAM credentials is
used for authentication. The IAM credentials are automatically supplied to AWS
instances in IAM instance profiles, Lambda functions, and others, and it is
this information already provided by AWS which Vault can use to authenticate
clients.
With the ec2 method, AWS is treated as a Trusted Third Party and
cryptographically signed dynamic metadata information that uniquely represents
each EC2 instance is used for authentication. This metadata information is
automatically supplied by AWS to all EC2 instances.
Based on how you attempt to authenticate, Vault will determine if you are
attempting to use the iam or ec2 type. Each has a different authentication
workflow, and each can solve different use cases.
Note: The ec2 method was implemented before the primitives to implement the
iam method were supported by AWS. The iam method is the recommended approach
as it is more flexible and aligns with best practices to perform access
control and authentication. See the section on comparing the two auth methods
below for more information.
»IAM auth method
The AWS STS API includes a method, sts:GetCallerIdentity, which allows you to validate the identity of a client. The client signs a GetCallerIdentity query using the AWS Signature v4 algorithm and sends it to the Vault server. The credentials used to sign the GetCallerIdentity request can come from the EC2 instance metadata service for an EC2 instance, or from the AWS environment variables in an AWS Lambda function execution, which obviates the need for an operator to manually provision some sort of identity material first. However, the credentials can, in principle, come from anywhere, not just from the locations AWS has provided for you.
The GetCallerIdentity query consists of four pieces of information: the request URL, the request body, the request headers, and the request method, as the AWS signature is computed over those fields. The Vault server reconstructs the query using this information and forwards it on to the AWS STS service. Depending on the response from the STS service, the server authenticates the client.
Notably, clients don't need network-level access themselves to talk to the AWS STS API endpoint; they merely need access to the credentials to sign the request. However, it means that the Vault server does need network-level access to send requests to the STS endpoint.
Each signed AWS request includes the current timestamp to mitigate the risk of
replay attacks. In addition, Vault allows you to require an additional header,
X-Vault-AWS-IAM-Server-ID, to be present to mitigate against different types
of replay attacks (such as a signed GetCallerIdentity request stolen from a
dev Vault instance and used to authenticate to a prod Vault instance). Vault
further requires that this header be one of the headers included in the AWS
signature and relies upon AWS to authenticate that signature.
While AWS API endpoints support both signed GET and POST requests, for
simplicity, the aws auth method supports only POST requests. It also does not
support presigned requests, i.e., requests with X-Amz-Credential,
X-Amz-Signature, and X-Amz-SignedHeaders GET query parameters containing
the authenticating information.
It's also important to note that Amazon does NOT appear to include any sort of
authorization around calls to GetCallerIdentity. For example, if you have an
IAM policy on your credential that requires all access to be MFA authenticated,
non-MFA authenticated credentials (i.e., raw credentials, not those retrieved
by calling GetSessionToken and supplying an MFA code) will still be able to
authenticate to Vault using this method. It does not appear possible to enforce
an IAM principal to be MFA authenticated while authenticating to Vault.
»EC2 auth method
Amazon EC2 instances have access to metadata which describes the instance. The Vault EC2 auth method leverages the components of this metadata to authenticate and distribute an initial Vault token to an EC2 instance. The data flow (which is also represented in the graphic below) is as follows:
An AWS EC2 instance fetches its AWS Instance Identity Document from the EC2 Metadata Service. In addition to data itself, AWS also provides the PKCS#7 signature of the data, and publishes the public keys (by region) which can be used to verify the signature.
The AWS EC2 instance makes a request to Vault with the PKCS#7 signature. The PKCS#7 signature contains the Instance Identity Document.
Vault verifies the signature on the PKCS#7 document, ensuring the information is certified accurate by AWS. This process validates both the validity and integrity of the document data. As an added security measure, Vault verifies that the instance is currently running using the public EC2 API endpoint.
Provided all steps are successful, Vault returns the initial Vault token to the EC2 instance. This token is mapped to any configured policies based on the instance metadata.
There are various modifications to this workflow that provide more or less security, as detailed later in this documentation.
»Authorization Workflow
The basic mechanism of operation is per-role. Roles are registered in the method and associated with a specific authentication type that cannot be changed once the role has been created. Roles can also be associated with various optional restrictions, such as the set of allowed policies and max TTLs on the generated tokens. Each role can be specified with the constraints that are to be met during the login. Many of these constraints accept lists of required values. For any constraint which accepts a list of values, that constraint will be considered satisfied if any one of the values is matched during the login process. For example, one such constraint that is supported is to bind against a list of AMI IDs. A role which is bound to a specific list of AMIs can only be used for login by EC2 instances that are deployed to one of the AMIs that the role is bound to.
The iam auth method allows you to specify bound IAM principal ARNs.
Clients authenticating to Vault must have an ARN that matches one of the ARNs bound to
the role they are attempting to login to. The bound ARN allows specifying a
wildcard at the end of the bound ARN. For example, if the bound ARN were
arn:aws:iam::123456789012:* it would allow any principal in AWS account
123456789012 to login to it. Similarly, if it were
arn:aws:iam::123456789012:role/* it would allow any IAM role in the AWS
account to login to it. If you wish to specify a wildcard, you must give Vault
iam:GetUser and iam:GetRole permissions to properly resolve the full user
path.
In general, role bindings that are specific to an EC2 instance are only checked when the ec2 auth method is used to login, while bindings specific to IAM principals are only checked when the iam auth method is used to login. However, the iam method includes the ability for you to "infer" an EC2 instance ID from the authenticated client and apply many of the bindings that would otherwise only apply specifically to EC2 instances.
In many cases, an organization will use a "seed AMI" that is specialized after bootup by configuration management or similar processes. For this reason, a role entry in the method can also be associated with a "role tag" when using the ec2 auth type. These tags are generated by the method and are placed as the value of a tag with the given key on the EC2 instance. The role tag can be used to further restrict the parameters set on the role, but cannot be used to grant additional privileges. If a role with an AMI bind constraint has "role tag" enabled on the role, and the EC2 instance performing login does not have an expected tag on it, or if the tag on the instance is deleted for some reason, authentication fails.
The role tags can be generated at will by an operator with appropriate API access. They are HMAC-signed by a per-role key stored within the method, allowing the method to verify the authenticity of a found role tag and ensure that it has not been tampered with. There is also a mechanism to deny list role tags if one has been found to be distributed outside of its intended set of machines.
»IAM Authentication Inferences
With the iam auth method, normally Vault will see the IAM principal that authenticated, either the IAM user or role. However, when you have an EC2 instance in an IAM instance profile, Vault can actually see the instance ID of the instance and can "infer" that it's an EC2 instance. However, there are important security caveats to be aware of before configuring Vault to make that inference.
Each AWS IAM role has a "trust policy" which specifies which entities are
trusted to call
sts:AssumeRole
on the role and retrieve credentials that can be used to authenticate with that
role. When AssumeRole is called, a parameter called RoleSessionName is passed
in, which is chosen arbitrarily by the entity which calls AssumeRole. If you
have a role with an ARN arn:aws:iam::123456789012:role/MyRole, then the
credentials returned by calling AssumeRole on that role will be
arn:aws:sts::123456789012:assumed-role/MyRole/RoleSessionName where
RoleSessionName is the session name in the AssumeRole API call. It is this
latter value which Vault actually sees.
When you have an EC2 instance in an instance profile, the corresponding role's
trust policy specifies that the principal "Service": "ec2.amazonaws.com" is
trusted to call AssumeRole. When this is configured, EC2 calls AssumeRole on
behalf of your instance, with a RoleSessionName corresponding to the
instance's instance ID. Thus, it is possible for Vault to extract the instance
ID out of the value it sees when an EC2 instance in an instance profile
authenticates to Vault with the iam auth method. This is known as
"inferencing." Vault can be configured, on a role-by-role basis, to infer that a
caller is an EC2 instance and, if so, apply further bindings that apply
specifically to EC2 instances -- most of the bindings available to the ec2
auth method.
However, it is very important to note that if any entity other than an AWS
service is permitted to call AssumeRole on your role, then that entity can
simply pass in your instance's instance ID and spoof your instance to Vault.
This also means that anybody who is able to modify your role's trust policy
(e.g., via
iam:UpdateAssumeRolePolicy,
then that person could also spoof your instances. If this is a concern but you
would like to take advantage of inferencing, then you should tightly restrict
who is able to call AssumeRole on the role, tightly restrict who is able to call
UpdateAssumeRolePolicy on the role, and monitor CloudTrail logs for calls to
AssumeRole and UpdateAssumeRolePolicy. All of these caveats apply equally to
using the iam auth method without inferencing; the point is merely
that Vault cannot offer an iron-clad guarantee about the inference and it is up
to operators to determine, based on their own AWS controls and use cases,
whether or not it's appropriate to configure inferencing.
»Mixing Authentication Types
Vault allows you to configure using either the ec2 auth method or the iam auth method, but not both auth methods. Further, Vault will prevent you from enforcing restrictions that it cannot enforce given the chosen auth type for a role. Some examples of how this works in practice:
- You configure a role with the ec2 auth type, with a bound AMI ID. A client would not be able to login using the iam auth type.
- You configure a role with the iam auth type, with a bound IAM principal ARN. A client would not be able to login with the ec2 auth method.
- You configure a role with the iam auth type and further configure inferencing. You have a bound AMI ID and a bound IAM principal ARN. A client must login using the iam method; the RoleSessionName must be a valid instance ID viewable by Vault, and the instance must have come from the bound AMI ID.
»Comparison of the IAM and EC2 Methods
The iam and ec2 auth methods serve similar and somewhat overlapping
functionality, in that both authenticate some type of AWS entity to Vault.
Here are some comparisons that illustrate why iam method is preferred over
ec2.
- What type of entity is authenticated:
- The ec2 auth method authenticates only AWS EC2 instances and is specialized to handle EC2 instances, such as restricting access to EC2 instances from a particular AMI, EC2 instances in a particular instance profile, or EC2 instances with a specialized tag value (via the role_tag feature).
- The iam auth method authenticates AWS IAM principals. This can include IAM users, IAM roles assumed from other accounts, AWS Lambdas that are launched in an IAM role, or even EC2 instances that are launched in an IAM instance profile. However, because it authenticates more generalized IAM principals, this method doesn't offer more granular controls beyond binding to a given IAM principal without the use of inferencing.
- How the entities are authenticated
- The ec2 auth method authenticates instances by making use of the EC2 instance identity document, which is a cryptographically signed document containing metadata about the instance. This document changes relatively infrequently, so Vault adds a number of other constructs to mitigate against replay attacks, such as client nonces, role tags, instance migrations, etc. Because the instance identity document is signed by AWS, you have a strong guarantee that it came from an EC2 instance.
- The iam auth method authenticates by having clients provide a specially signed AWS API request which the method then passes on to AWS to validate the signature and tell Vault who created it. The actual secret (i.e., the AWS secret access key) is never transmitted over the wire, and the AWS signature algorithm automatically expires requests after 15 minutes, providing simple and robust protection against replay attacks. The use of inferencing, however, provides a weaker guarantee that the credentials came from an EC2 instance in an IAM instance profile compared to the ec2 authentication mechanism.
- The instance identity document used in the ec2 auth method is more likely to be stolen given its relatively static nature, but it's harder to spoof. On the other hand, the credentials of an EC2 instance in an IAM instance profile are less likely to be stolen given their dynamic and short-lived nature, but it's easier to spoof credentials that might have come from an EC2 instance.
- Specific use cases
- If you have non-EC2 instance entities, such as IAM users, Lambdas in IAM roles, or developer laptops using AdRoll's Hologram then you would need to use the iam auth method.
- If you have EC2 instances, then you could use either auth method. If you need more granular filtering beyond just the instance profile of given EC2 instances (such as filtering based off the AMI the instance was launched from), then you would need to use the ec2 auth method, change the instance profile associated with your EC2 instances so they have unique IAM roles for each different Vault role you would want them to authenticate to, or make use of inferencing. If you need to make use of role tags, then you will need to use the ec2 auth method.
»Recommended Vault IAM Policy
This specifies the recommended IAM policy needed by the AWS auth method. Note that if you are using the same credentials for the AWS auth and secret methods (e.g., if you're running Vault on an EC2 instance in an IAM instance profile), then you will need to add additional permissions as required by the AWS secret method.
Here are some of the scenarios in which Vault would need to use each of these permissions. This isn't intended to be an exhaustive list of all the scenarios in which Vault might make an AWS API call, but rather illustrative of why these are needed.
ec2:DescribeInstancesis necessary when you are using theec2auth method or when you are inferring anec2_instanceentity type to validate that the EC2 instance meets binding requirements of the roleiam:GetInstanceProfileis used when you have abound_iam_role_arnin theec2auth method. Vault needs to determine which IAM role is attached to the instance profile.iam:GetUserandiam:GetRoleare used when using the iam auth method and binding to an IAM user or role principal to determine the AWS IAM Unique Identifiers or when using a wildcard on the bound ARN to resolve the full ARN of the user or role.- The
sts:AssumeRolestanza is necessary when you are using Cross Account Access. TheResources specified should be a list of all the roles for which you have configured cross-account access, and each of those roles should have this IAM policy attached (except for thests:AssumeRolestatement). - The
ManageOwnAccessKeysstanza is necessary when you have configured Vault with static credentials, and wish to rotate these credentials with the Rotate Root Credentials API call.
»Client Nonce
Note: this only applies to the ec2 auth method.
If an unintended party gains access to the PKCS#7 signature of the identity document (which by default is available to every process and user that gains access to an EC2 instance), it can impersonate that instance and fetch a Vault token. The method addresses this problem by using a Trust On First Use (TOFU) mechanism that allows the first client to present the PKCS#7 signature of the document to be authenticated and denying the rest. An important property of this design is detection of unauthorized access: if an unintended party authenticates, the intended client will be unable to authenticate and can raise an alert for investigation.
During the first login, the method stores the instance ID that authenticated
in a accesslist. One method of operation of the method is to disallow any
authentication attempt for an instance ID contained in the access list, using the
disallow_reauthentication option on the role, meaning that an instance is
allowed to login only once. However, this has consequences for token rotation,
as it means that once a token has expired, subsequent authentication attempts
would fail. By default, reauthentication is enabled in this method, and can be
turned off using disallow_reauthentication parameter on the registered role.
In the default method of operation, the method will return a unique nonce
during the first authentication attempt, as part of auth metadata. Clients
should present this nonce for subsequent login attempts and it should match
the nonce cached at the identity-accesslist entry at the method. Since only
the original client knows the nonce, only the original client is allowed to
reauthenticate. (This is the reason that this is a accesslist rather than a
deny list; by default, it's keeping track of clients allowed to reauthenticate,
rather than those that are not.). Clients can choose to provide a nonce even
for the first login attempt, in which case the provided nonce will be tied to
the cached identity-accesslist entry. It is recommended to use a strong nonce
value in this case.
It is up to the client to behave correctly with respect to the nonce; if the client stores the nonce on disk it can survive reboots, but could also give access to other users or applications on the instance. It is also up to the operator to ensure that client nonces are in fact unique; sharing nonces allows a compromise of the nonce value to enable an attacker that gains access to any EC2 instance to imitate the legitimate client on that instance. This is why nonces can be disabled on the method side in favor of only a single authentication per instance; in some cases, such as when using ASGs, instances are immutable and single-boot anyways, and in conjunction with a high max TTL, reauthentication may not be needed (and if it is, the instance can simply be shut down and allow ASG to start a new one).
In both cases, entries can be removed from the accesslist by instance ID, allowing reauthentication by a client if the nonce is lost (or not used) and an operator approves the process.
One other point: if available by the OS/distribution being used with the EC2 instance, it is not a bad idea to firewall access to the signed PKCS#7 metadata to ensure that it is accessible only to the matching user(s) that require access.
The client nonce which is generated by the backend and which gets returned along with the authentication response, will be audit logged in plaintext. If this is undesired, clients can supply a custom nonce to the login endpoint which will not be returned and hence will not be audit logged.
»Advanced Options and Caveats
»Dynamic Management of Policies Via Role Tags
Note: This only applies to the ec2 auth method or the iam auth method when inferencing is used.
If the instance is required to have customized set of policies based on the
role it plays, the role_tag option can be used to provide a tag to set on
instances, for a given role. When this option is set, during login, along with
verification of PKCS#7 signature and instance health, the method will query
for the value of a specific tag with the configured key that is attached to the
instance. The tag holds information that represents a subset of privileges that
are set on the role and are used to further restrict the set of the role's
privileges for that particular instance.
A role_tag can be created using auth/aws/role/<role>/tag endpoint
and is immutable. The information present in the tag is SHA256 hashed and HMAC
protected. The per-role key to HMAC is only maintained in the method. This prevents
an adversarial operator from modifying the tag when setting it on the EC2 instance
in order to escalate privileges.
When 'role_tag' option is enabled on a role, the instances are required to have a
role tag. If the tag is not found on the EC2 instance, authentication will fail.
This is to ensure that privileges of an instance are never escalated for not
having the tag on it or for getting the tag removed. If the role tag creation does
not specify the policy component, the client will inherit the allowed policies set
on the role. If the role tag creation specifies the policy component but it contains
no policies, the token will contain only the default policy; by default, this policy
allows only manipulation (revocation, renewal, lookup) of the existing token, plus
access to its cubbyhole.
This can be useful to allow instances access to a secure "scratch space" for
storing data (via the token's cubbyhole) but without granting any access to
other resources provided by or resident in Vault.
»Handling Lost Client Nonces
Note: This only applies to the ec2 auth method.
If an EC2 instance loses its client nonce (due to a reboot, a stop/start of the
client, etc.), subsequent login attempts will not succeed. If the client nonce
is lost, normally the only option is to delete the entry corresponding to the
instance ID from the identity accesslist in the method. This can be done via
the auth/aws/identity-accesslist/<instance_id> endpoint. This allows a new
client nonce to be accepted by the method during the next login request.
Under certain circumstances there is another useful setting. When the instance
is placed onto a host upon creation, it is given a pendingTime value in the
instance identity document (documentation from AWS does not cover this option,
unfortunately). If an instance is stopped and started, the pendingTime value
is updated (this does not apply to reboots, however).
The method can take advantage of this via the allow_instance_migration
option, which is set per-role. When this option is enabled, if the client nonce
does not match the saved nonce, the pendingTime value in the instance
identity document will be checked; if it is newer than the stored pendingTime
value, the method assumes that the client was stopped/started and allows the
client to log in successfully, storing the new nonce as the valid nonce for
that client. This essentially re-starts the TOFU mechanism any time the
instance is stopped and started, so should be used with caution. Just like with
initial authentication, the legitimate client should have a way to alert (or an
alert should trigger based on its logs) if it is denied authentication.
Unfortunately, the allow_instance_migration only helps during stop/start
actions; the current metadata does not provide for a way to allow this
automatic behavior during reboots. The method will be updated if this needed
metadata becomes available.
The allow_instance_migration option is set per-role, and can also be
specified in a role tag. Since role tags can only restrict behavior, if the
option is set to false on the role, a value of true in the role tag takes
effect; however, if the option is set to true on the role, a value set in the
role tag has no effect.
»Disabling Reauthentication
Note: this only applies to the ec2 auth method.
If in a given organization's architecture, a client fetches a long-lived Vault
token and has no need to rotate the token, all future logins for that instance
ID can be disabled. If the option disallow_reauthentication is set, only one
login will be allowed per instance. If the intended client successfully
retrieves a token during login, it can be sure that its token will not be
hijacked by another entity.
When disallow_reauthentication option is enabled, the client can choose not
to supply a nonce during login, although it is not an error to do so (the nonce
is simply ignored). Note that reauthentication is enabled by default. If only
a single login is desired, disallow_reauthentication should be set explicitly
on the role or on the role tag.
The disallow_reauthentication option is set per-role, and can also be
specified in a role tag. Since role tags can only restrict behavior, if the
option is set to false on the role, a value of true in the role tag takes
effect; however, if the option is set to true on the role, a value set in the
role tag has no effect.
»Deny listing Role Tags
Note: this only applies to the ec2 auth method or the iam auth method when inferencing is used.
Role tags are tied to a specific role, but the method has no control over, which
instances using that role, should have any particular role tag; that is purely up
to the operator. Although role tags are only restrictive (a tag cannot escalate
privileges above what is set on its role), if a role tag is found to have been
used incorrectly, and the administrator wants to ensure that the role tag has no
further effect, the role tag can be placed on a deny list via the endpoint
auth/aws/roletag-denylist/<role_tag>. Note that this will not invalidate the
tokens that were already issued; this only blocks any further login requests from
those instances that have the deny listed tag attached to them.
»Expiration Times and Tidying of denylist and accesslist Entries
The expired entries in both identity accesslist and role tag denylist are
deleted automatically. The entries in both of these lists contain an expiration
time which is dynamically determined by three factors: max_ttl set on the role,
max_ttl set on the role tag, and max_ttl value of the method mount. The
least of these three dictates the maximum TTL of the issued token, and
correspondingly will be set as the expiration times of these entries.
The endpoints auth/aws/tidy/identity-accesslist and auth/aws/tidy/roletag-denylist are
provided to clean up the entries present in these lists. These endpoints allow
defining a safety buffer, such that an entry must not only be expired, but be
past expiration by the amount of time dictated by the safety buffer in order
to actually remove the entry.
Automatic deletion of expired entries is performed by the periodic function
of the method. This function does the tidying of both access list role tags
and access list identities. Periodic tidying is activated by default and will
have a safety buffer of 72 hours, meaning only those entries are deleted which
were expired before 72 hours from when the tidy operation is being performed.
This can be configured via config/tidy/roletag-denylist and config/tidy/identity-accesslist
endpoints.
»Varying Public Certificates
Note: this only applies to the ec2 auth method.
The AWS public certificate, which contains the public key used to verify the
PKCS#7 signature, varies for different AWS regions. The primary AWS public
certificate, which covers most AWS regions, is already included in Vault and
does not need to be added. Instances whose PKCS#7 signatures cannot be
verified by the default public certificate included in Vault can register a
different public certificate which can be found here,
via the auth/aws/config/certificate/<cert_name> endpoint.
»Dangling Tokens
An EC2 instance, after authenticating itself with the method, gets a Vault token. After that, if the instance terminates or goes down for any reason, the method will not be aware of such events. The token issued will still be valid, until it expires. The token will likely be expired sooner than its lifetime when the instance fails to renew the token on time.
»Cross Account Access
To allow Vault to authenticate IAM principals and EC2 instances in other
accounts, Vault supports using AWS STS (Security Token Service) to assume AWS
IAM Roles in other accounts. For each target AWS account ID, you configure the
IAM Role for Vault to assume using the auth/aws/config/sts/<account_id> and
Vault will use credentials from assuming that role to validate IAM principals
and EC2 instances in the target account.
The account in which Vault is running (i.e. the master account) must be listed as
a trusted entity in the IAM Role being assumed on the remote account. The Role itself
should allow the permissions specified in the Recommended Vault IAM
Policy except it doesn't need any further
sts:AssumeRole permissions.
Furthermore, in the master account, Vault must be granted the action sts:AssumeRole
for the IAM Role to be assumed.
»AWS Instance Metadata Timeout
Affects Vault 1.4 and later
Anytime Vault uses the instance metadata service on an EC2 instance, such as for getting credentials from the instance profile, there may be a delay with the introduction of v2 of the instance metadata service (IMDSv2). The AWS SDK used by Vault first attempts to connect to IMDSv2, and if that times out, it falls back to v1. In Vault 1.4, this timeout can take up to 2 minutes. In Vault 1.5.5 and later, it can take up to 2 seconds with this fix: #10133.
The timeout occurs in situations where there is a proxy between Vault and IMDSv2, and the instance hop limit is set to less than the number of "hops" between Vault and IMDSv2. For example, if Vault is running in docker on an EC2 instance with the instance hop limit set to 1, the AWS SDK client will attempt to connect to IMDSv2, timeout, and fall back to IMDSv1 because of the extra network hop between docker and IMDS.
To avoid the timeout behavior, the hop limit may be adjusted on the underlying EC2 instances. With the docker example, setting the hop limit to 2 will allow the AWS SDK in Vault to connect to IMDSv2 without delay.
»Authentication
»Via the CLI
»Enable AWS EC2 authentication in Vault.
»Configure the credentials required to make AWS API calls
If not specified, Vault will attempt to use standard environment variables
(AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY) or IAM EC2 instance role
credentials if available.
The IAM account or role to which the credentials map must allow the
ec2:DescribeInstances action. In addition, if IAM Role binding is used (see
bound_iam_role_arn below), iam:GetInstanceProfile must also be allowed.
»Configure the policies on the role.
»Configure a required X-Vault-AWS-IAM-Server-ID Header (recommended)
»Perform the login operation
For the iam auth method, generating the signed request is a non-standard operation. The Vault cli supports generating this for you:
This assumes you have AWS credentials configured in the standard locations AWS
SDKs search for credentials (environment variables, ~/.aws/credentials, IAM
instance profile, or ECS task role, in that order). If you do not have IAM
credentials available at any of these locations, you can explicitly pass them
in on the command line (though this is not recommended), omitting
aws_security_token if not applicable.
The region used defaults to us-east-1, but you can specify a custom region like so:
When using a custom region, be sure the designated region corresponds to that of the STS endpoint you're using.
An example of how to generate the required request values for the login method
can be found found in the vault cli
source code.
Using an approach such as this, the request parameters can be generated and
passed to the login method:
»Via the API
»Enable AWS authentication in Vault.
»Configure the credentials required to make AWS API calls.
»Configure the policies on the role.
»Perform the login operation
The response will be in JSON. For example:
»API
The AWS auth method has a full HTTP API. Please see the AWS Auth API for more details.
