Once Keystone is installed, it is configured via a primary configuration file (etc/keystone.conf), a PasteDeploy configuration file (etc/keystone-paste.ini), possibly a separate logging configuration file, and initializing data into Keystone using the command line client.
By default, Keystone starts a service on IANA-assigned port 35357. This may overlap with your system’s ephemeral port range, so another process may already be using this port without being explicitly configured to do so. To prevent this scenario from occurring, it’s recommended that you explicitly exclude port 35357 from the available ephemeral port range. On a Linux system, this would be accomplished by:
$ sysctl -w 'net.ipv4.ip_local_reserved_ports=35357'
To make the above change persistent, net.ipv4.ip_local_reserved_ports = 35357 should be added to /etc/sysctl.conf or to /etc/sysctl.d/keystone.conf.
The Keystone configuration files are an ini file format based on Paste, a common system used to configure Python WSGI based applications. The PasteDeploy configuration entries (WSGI pipeline definitions) can be provided in a separate keystone-paste.ini file, while general and driver-specific configuration parameters are in the primary configuration file keystone.conf.
Note
Since keystone’s PasteDeploy configuration file has been separated from the main keystone configuration file, keystone.conf, all local configuration or driver-specific configuration parameters must go in the main keystone configuration file instead of the PasteDeploy configuration file, i.e. configuration in keystone-paste.ini is not supported.
The primary configuration file is organized into the following sections:
The Keystone primary configuration file is expected to be named keystone.conf. When starting Keystone, you can specify a different configuration file to use with --config-file. If you do not specify a configuration file, Keystone will look in the following directories for a configuration file, in order:
PasteDeploy configuration file is specified by the config_file parameter in [paste_deploy] section of the primary configuration file. If the parameter is not an absolute path, then Keystone looks for it in the same directories as above. If not specified, WSGI pipeline definitions are loaded from the primary configuration file.
One of the most impactful decisions you’ll have to make when configuring keystone is deciding how you want keystone to source your identity data. Keystone supports several different choices that will substantially impact how you’ll configure, deploy, and interact with keystone.
You can also mix-and-match various sources of identity (see Domain-specific Drivers below for an example). For example, you can store OpenStack service users and their passwords in SQL, manage customers in LDAP, and authenticate employees via SAML federation.
| Feature | Status | REMOTE_USER | LDAP | OAuth v1.0a | OpenID Connect | SAML v2 | SQL |
|---|---|---|---|---|---|---|---|
| Local authentication | optional | ✖ | ✔ | ✔ | ✖ | ✖ | ✔ |
| External authentication | optional | ✔ | ✖ | ✖ | ✔ | ✔ | ✖ |
| Identity management | optional | ✖ | ✔ | ✔ | ✖ | ✖ | ✔ |
| PCI-DSS controls | optional | ✔ | ✔ | ✖ | ✖ | ✖ | ✔ |
| Auditing | optional | ✖ | ✔ | ✖ | ✔ | ✔ | ✔ |
Status: optional. Authenticate with keystone by providing credentials directly to keystone.
drivers:
Status: optional. Authenticate with keystone by providing credentials to an external system that keystone trusts (as with federation).
drivers:
Status: optional. Create, update, enable/disable, and delete users via Keystone’s HTTP API.
drivers:
Status: optional. Configure keystone to enforce PCI-DSS compliant security controls.
drivers:
Status: optional. Audit authentication flows using PyCADF.
drivers:
Keystone supports the option (disabled by default) to specify identity driver configurations on a domain by domain basis, allowing, for example, a specific domain to have its own LDAP or SQL server. This is configured by specifying the following options:
[identity]
domain_specific_drivers_enabled = True
domain_config_dir = /etc/keystone/domains
Setting domain_specific_drivers_enabled to True will enable this feature, causing Keystone to look in the domain_config_dir for config files of the form:
keystone.<domain_name>.conf
Options given in the domain specific configuration file will override those in the primary configuration file for the specified domain only. Domains without a specific configuration file will continue to use the options from the primary configuration file.
Keystone also supports the ability to store the domain-specific configuration options in the keystone SQL database, managed via the Identity API, as opposed to using domain-specific configuration files.
Note
Support status for configuring options via the Identity API
Experimental (Kilo, Liberty, Mitaka) Stable (Newton)
This capability (which is disabled by default) is enabled by specifying the following options in the main keystone configuration file:
[identity]
domain_specific_drivers_enabled = true
domain_configurations_from_database = true
Once enabled, any existing domain-specific configuration files in the configuration directory will be ignored and only those domain-specific configuration options specified via the Identity API will be used.
Unlike the file-based method of specifying domain-specific configurations, options specified via the Identity API will become active without needing to restart the keystone server. For performance reasons, the current state of configuration options for a domain are cached in the keystone server, and in multi-process and multi-threaded keystone configurations, the new configuration options may not become active until the cache has timed out. The cache settings for domain config options can be adjusted in the general keystone configuration file (option cache_time in the domain_config group).
Note
It is important to notice that when using either of these methods of specifying domain-specific configuration options, the main keystone configuration file is still maintained. Only those options that relate to the Identity driver for users and groups (i.e. specifying whether the driver for this domain is SQL or LDAP, and, if LDAP, the options that define that connection) are supported in a domain-specific manner. Further, when using the configuration options via the Identity API, the driver option must be set to an LDAP driver (attempting to set it to an SQL driver will generate an error when it is subsequently used).
For existing installations that already use file-based domain-specific configurations who wish to migrate to the SQL-based approach, the keystone-manage command can be used to upload all configuration files to the SQL database:
$ keystone-manage domain_config_upload --all
Once uploaded, these domain-configuration options will be visible via the Identity API as well as applied to the domain-specific drivers. It is also possible to upload individual domain-specific configuration files by specifying the domain name:
$ keystone-manage domain_config_upload --domain-name DOMAINA
Note
It is important to notice that by enabling either of the domain-specific configuration methods, the operations of listing all users and listing all groups are not supported, those calls will need either a domain filter to be specified or usage of a domain scoped token.
Note
Keystone does not support moving the contents of a domain (i.e. “its” users and groups) from one backend to another, nor group membership across backend boundaries.
Note
When using the file-based domain-specific configuration method, to delete a domain that uses a domain specific backend, it’s necessary to first disable it, remove its specific configuration file (i.e. its corresponding keystone.<domain_name>.conf) and then restart the Identity server. When managing configuration options via the Identity API, the domain can simply be disabled and deleted via the Identity API; since any domain-specific configuration options will automatically be removed.
Note
Although Keystone supports multiple LDAP backends via the above domain-specific configuration methods, it currently only supports one SQL backend. This could be either the default driver or a single domain-specific backend, perhaps for storing service users in a predominantly LDAP installation.
Note
Keystone has deprecated the keystone-manage domain_config_upload option. The keystone team recommends setting domain config options via the API instead.
Due to the need for user and group IDs to be unique across an OpenStack installation and for Keystone to be able to deduce which domain and backend to use from just a user or group ID, it dynamically builds a persistent identity mapping table from a public ID to the actual domain, local ID (within that backend) and entity type. The public ID is automatically generated by Keystone when it first encounters the entity. If the local ID of the entity is from a backend that does not guarantee to generate UUIDs, a hash algorithm will generate a public ID for that entity, which is what will be exposed by Keystone.
The use of a hash will ensure that if the public ID needs to be regenerated then the same public ID will be created. This is useful if you are running multiple keystones and want to ensure the same ID would be generated whichever server you hit.
While Keystone will dynamically maintain the identity mapping, including removing entries when entities are deleted via the Keystone, for those entities in backends that are managed outside of Keystone (e.g. a Read Only LDAP), Keystone will not know if entities have been deleted and hence will continue to carry stale identity mappings in its table. While benign, keystone provides an ability for operators to purge the mapping table of such stale entries using the keystone-manage command, for example:
$ keystone-manage mapping_purge --domain-name DOMAINA --local-id abc@de.com
A typical usage would be for an operator to obtain a list of those entries in an external backend that had been deleted out-of-band to Keystone, and then call keystone-manage to purge those entries by specifying the domain and local-id. The type of the entity (i.e. user or group) may also be specified if this is needed to uniquely identify the mapping.
Since public IDs can be regenerated with the correct generator implementation, if the details of those entries that have been deleted are not available, then it is safe to simply bulk purge identity mappings periodically, for example:
$ keystone-manage mapping_purge --domain-name DOMAINA
will purge all the mappings for DOMAINA. The entire mapping table can be purged with the following command:
$ keystone-manage mapping_purge --all
Generating public IDs in the first run may take a while, and most probably first API requests to fetch user list will fail by timeout. To prevent this, mapping_populate command should be executed. It should be executed right after LDAP has been configured or after mapping_purge.
$ keystone-manage mapping_populate --domain DOMAINA
Keystone supports a customizable public ID generator and it is specified in the [identity_mapping] section of the configuration file. Keystone provides a sha256 generator as default, which produces regeneratable public IDs. The generator algorithm for public IDs is a balance between key size (i.e. the length of the public ID), the probability of collision and, in some circumstances, the security of the public ID. The maximum length of public ID supported by Keystone is 64 characters, and the default generator (sha256) uses this full capability. Since the public ID is what is exposed externally by Keystone and potentially stored in external systems, some installations may wish to make use of other generator algorithms that have a different trade-off of attributes. A different generator can be installed by configuring the following property:
Warning
Changing the generator may cause all existing public IDs to be become invalid, so typically the generator selection should be considered immutable for a given installation.
Note
This feature is only supported by Keystone for the Identity API v3 clients.
Keystone supports authentication plugins and they are specified in the [auth] section of the configuration file. However, an authentication plugin may also have its own section in the configuration file. It is up to the plugin to register its own configuration options.
Keystone provides three authentication methods by default. password handles password authentication and token handles token authentication. external is used in conjunction with authentication performed by a container web server that sets the REMOTE_USER environment variable. For more details, refer to External Authentication.
All authentication plugins must extend the keystone.auth.plugins.base.AuthMethodHandler class and implement the authenticate() method. The authenticate() method expects the following parameters.
If successful, the authenticate() method must provide a valid user_id in auth_context and return None. method_name is used to convey any additional authentication methods in case authentication is for re-scoping. For example, if the authentication is for re-scoping, a plugin must append the previous method names into method_names. Also, a plugin may add any additional information into extras. Anything in extras will be conveyed in the token’s extras field.
If authentication requires multiple steps, the authenticate() method must return the payload in the form of a dictionary for the next authentication step.
If authentication is unsuccessful, the authenticate() method must raise a keystone.exception.Unauthorized exception.
Simply add the new plugin name to the methods list along with your plugin class configuration in the [auth] sections of the configuration file to deploy it.
If the plugin requires additional configurations, it may register its own section in the configuration file.
Plugins are invoked in the order in which they are specified in the methods attribute of the authentication request body. If multiple plugins are invoked, all plugins must succeed in order to for the entire authentication to be successful. Furthermore, all the plugins invoked must agree on the user_id in the auth_context.
The REMOTE_USER environment variable is only set from a containing webserver. However, to ensure that a user must go through other authentication mechanisms, even if this variable is set, remove external from the list of plugins specified in methods. This effectively disables external authentication. For more details, refer to ExternalAuthentication.
Keystone supports customizable token persistence drivers. These can be specified in the [token] section of the configuration file. Keystone provides three non-test persistence backends. These can be set with the [token] driver configuration option.
The drivers Keystone provides are:
Keystone supports customizable token provider and it is specified in the [token] section of the configuration file. Keystone provides both UUID and PKI token providers. However, users may register their own token provider by configuring the following property.
Each token format uses different technologies to achieve various performance, scaling and architectural requirements.
UUID tokens contain randomly generated UUID4 payloads that are issued and validated by the identity service. They are encoded using their hex digest for transport and are thus URL-friendly. They must be persisted by the identity service in order to be later validated. Revoking them is simply a matter of deleting them from the token persistence backend.
Both PKI and PKIZ tokens contain JSON payloads that represent the entire token validation response that would normally be retrieved from keystone. The payload is then signed using Cryptographic Message Syntax (CMS). The combination of CMS and the exhaustive payload allows PKI and PKIZ tokens to be verified offline using keystone’s public signing key. The only reason for them to be persisted by the identity service is to later build token revocation lists (explicit lists of tokens that have been revoked), otherwise they are theoretically ephemeral when supported by token revocation events (which describe invalidated tokens rather than enumerate them). PKIZ tokens add zlib compression after signing to achieve a smaller overall token size. To make them URL-friendly, PKI tokens are base64 encoded and then arbitrarily manipulated to replace unsafe characters with safe ones whereas PKIZ tokens use conventional base64url encoding. Due to the size of the payload and the overhead incurred by the CMS format, both PKI and PKIZ tokens may be too long to fit in either headers or URLs if they contain extensive service catalogs or other additional attributes. Some third-party applications such as web servers and clients may need to be recompiled from source to customize the limitations that PKI and PKIZ tokens would otherwise exceed). Both PKI and PKIZ tokens require signing certificates which may be created using keystone-manage pki_setup for demonstration purposes (this is not recommended for production deployments: use certificates issued by an trusted CA instead).
Fernet tokens contain a limited amount of identity and authorization data in a MessagePacked payload. The payload is then wrapped as a Fernet message for transport, where Fernet provides the required web safe characteristics for use in URLs and headers. Fernet tokens require symmetric encryption keys which can be established using keystone-manage fernet_setup and periodically rotated using keystone-manage fernet_rotate.
Warning
UUID, PKI, PKIZ, and Fernet tokens are all bearer tokens, meaning that they must be protected from unnecessary disclosure to prevent unauthorized access.
| Feature | Status | Fernet tokens | PKI tokens | PKIZ tokens | UUID tokens |
|---|---|---|---|---|---|
| Create unscoped token | mandatory | ✔ | ✔ | ✔ | ✔ |
| Create project-scoped token | mandatory | ✔ | ✔ | ✔ | ✔ |
| Create domain-scoped token | optional | ✔ | ✔ | ✔ | ✔ |
| Create trust-scoped token | optional | ✔ | ✔ | ✔ | ✔ |
| Create a token given an OAuth access token | optional | ✔ | ✔ | ✔ | ✔ |
| Create a token with a bind attribute | optional | ✖ | ✔ | ✔ | ✔ |
| Revoke a token | optional | ✔ | ✔ | ✔ | ✔ |
| Online validation | mandatory | ✔ | ✔ | ✔ | ✔ |
| Offline validation | optional | ✖ | ✔ | ✔ | ✖ |
| Non-persistent | optional | ✔ | ✔ | ✔ | ✖ |
Status: mandatory. All token providers must be capable of issuing tokens without an explicit scope of authorization.
CLI commands:
drivers:
Status: mandatory. All token providers must be capable of issuing project-scoped tokens.
CLI commands:
drivers:
Status: optional. Domain-scoped tokens are not required for all use cases, and for some use cases, projects can be used instead.
CLI commands:
drivers:
Status: optional. Tokens scoped to a trust convey only the user impersonation and project-based authorization attributes included in the delegation.
CLI commands:
drivers:
Status: optional. OAuth access tokens can be exchanged for keystone tokens.
drivers:
Status: optional. Tokens can express a binding to an additional authentication method, such as kerberos or x509.
drivers:
Status: optional. Tokens may be individually revoked, such as when a user logs out of Horizon. Under certain circumstances, it’s acceptable for more than just a single token may be revoked as a result of this operation (such as when the revoked token was previously used to create additional tokens).
CLI commands:
drivers:
Status: mandatory. Keystone must be able to validate the tokens that it issues when presented with a token that it previously issued.
drivers:
Status: optional. Services using Keystone for authentication may want to validate tokens themselves, rather than calling back to keystone, in order to improve performance and scalability.
drivers:
Status: optional. If a token format does not require persistence (such as to a SQL backend), then there is no scalability limit to the number of tokens that keystone can issue at once, and there is no need to perform clean up operations such as `keystone-manage token_flush`.
drivers:
Keystone supports a caching layer that is above the configurable subsystems (e.g. token, identity, etc). Keystone uses the dogpile.cache library which allows for flexible cache backends. The majority of the caching configuration options are set in the [cache] section. However, each section that has the capability to be cached usually has a caching boolean value that will toggle caching for that specific section. The current default behavior is that subsystem caching is enabled, but the global toggle is set to disabled.
enabled - enables/disables caching across all of keystone
debug_cache_backend - enables more in-depth logging from the cache backend (get, set, delete, etc)
backend - the caching backend module to use e.g. dogpile.cache.memcached
Note
A given backend must be registered with dogpile.cache before it can be used. The default backend is the Keystone no-op backend (keystone.common.cache.noop). If caching is desired a different backend will need to be specified. Current functional backends are:
dogpile.cache.memcached - Memcached backend using the standard python-memcached library (recommended for use with Apache httpd with mod_wsgi)
dogpile.cache.pylibmc - Memcached backend using the pylibmc library
dogpile.cache.bmemcached - Memcached using python-binary-memcached library.
dogpile.cache.redis - Redis backend
dogpile.cache.dbm - local DBM file backend
dogpile.cache.memory - in-memory cache
oslo_cache.mongo - MongoDB as caching backend
oslo_cache.memcache_pool - Memcache with pooling. This implementation also provides client connection re-use.
Warning
dogpile.cache.memory is not suitable for use outside of unit testing as it does not cleanup its internal cache on cache expiration, does not provide isolation to the cached data (values in the store can be inadvertently changed without extra layers of data protection added), and does not share cache between processes. This means that caching and cache invalidation will not be consistent or reliable when using Keystone and the dogpile.cache.memory backend under any real workload.
expiration_time - int, the default length of time to cache a specific value. A value of 0 indicates to not cache anything. It is recommended that the enabled option be used to disable cache instead of setting this to 0.
backend_argument - an argument passed to the backend when instantiated backend_argument should be specified once per argument to be passed to the backend and in the format of <argument name>:<argument value>. e.g.: backend_argument = host:localhost
proxies - comma delimited list of ProxyBackends e.g. my.example.Proxy, my.example.Proxy2
The token system has a separate cache_time configuration option, that can be set to a value above or below the global expiration_time default, allowing for different caching behavior from the other systems in Keystone. This option is set in the [token] section of the configuration file.
The Token Revocation List cache time is handled by the configuration option revocation_cache_time in the [token] section. The revocation list is refreshed whenever a token is revoked. It typically sees significantly more requests than specific token retrievals or token validation calls.
The resource system has a separate cache_time configuration option, that can be set to a value above or below the global expiration_time default, allowing for different caching behavior from the other systems in Keystone. This option is set in the [resource] section of the configuration file.
Currently resource has caching for project and domain specific requests (primarily around the CRUD actions). The list_projects and list_domains methods are not subject to caching.
Warning
Be aware that if a read-only resource backend is in use, the cache will not immediately reflect changes on the back end. Any given change may take up to the cache_time (if set in the [resource] section of the configuration) or the global expiration_time (set in the [cache] section of the configuration) before it is reflected. If this type of delay (when using a read-only resource backend) is an issue, it is recommended that caching be disabled on resource. To disable caching specifically on resource, in the [resource] section of the configuration set caching to False.
Currently role has caching for get_role, but not for list_roles. The role system has a separate cache_time configuration option, that can be set to a value above or below the global expiration_time default, allowing for different caching behavior from the other systems in Keystone. This option is set in the [role] section of the configuration file.
Warning
Be aware that if a read-only role backend is in use, the cache will not immediately reflect changes on the back end. Any given change may take up to the cache_time (if set in the [role] section of the configuration) or the global expiration_time (set in the [cache] section of the configuration) before it is reflected. If this type of delay (when using a read-only role backend) is an issue, it is recommended that caching be disabled on role. To disable caching specifically on role, in the [role] section of the configuration set caching to False.
PKI stands for Public Key Infrastructure. Tokens are documents, cryptographically signed using the X509 standard. In order to work correctly token generation requires a public/private key pair. The public key must be signed in an X509 certificate, and the certificate used to sign it must be available as Certificate Authority (CA) certificate. These files can be either externally generated or generated using the keystone-manage utility.
The files used for signing and verifying certificates are set in the Keystone configuration file. The private key should only be readable by the system user that will run Keystone. The values that specify the certificates are under the [signing] section of the configuration file. The configuration values are:
You may use a signing certificate issued by an external CA instead of generated by keystone-manage. However, certificate issued by external CA must satisfy the following conditions:
The basic workflow for using a signing certificate issued by an external CA involves:
One way to request a signing certificate from an external CA is to first generate a PKCS #10 Certificate Request Syntax (CRS) using OpenSSL CLI.
First create a certificate request configuration file (e.g. cert_req.conf):
[ req ]
default_bits = 2048
default_keyfile = keystonekey.pem
default_md = default
prompt = no
distinguished_name = distinguished_name
[ distinguished_name ]
countryName = US
stateOrProvinceName = CA
localityName = Sunnyvale
organizationName = OpenStack
organizationalUnitName = Keystone
commonName = Keystone Signing
emailAddress = keystone@openstack.org
Then generate a CRS with OpenSSL CLI. Do not encrypt the generated private key. The -nodes option must be used.
For example:
$ openssl req -newkey rsa:2048 -keyout signing_key.pem -keyform PEM -out signing_cert_req.pem -outform PEM -config cert_req.conf -nodes
If everything is successfully, you should end up with signing_cert_req.pem and signing_key.pem. Send signing_cert_req.pem to your CA to request a token signing certificate and make sure to ask the certificate to be in PEM format. Also, make sure your trusted CA certificate chain is also in PEM format.
Assuming you have the following already:
Copy the above to your certificate directory. For example:
$ mkdir -p /etc/keystone/ssl/certs
$ cp signing_cert.pem /etc/keystone/ssl/certs/
$ cp signing_key.pem /etc/keystone/ssl/certs/
$ cp cacert.pem /etc/keystone/ssl/certs/
$ chmod -R 700 /etc/keystone/ssl/certs
Make sure the certificate directory is root-protected.
If your certificate directory path is different from the default /etc/keystone/ssl/certs, make sure it is reflected in the [signing] section of the configuration file.
keystone-manage pki_setup is a development tool. We recommend that you do not use keystone-manage pki_setup in a production environment. In production, an external CA should be used instead. This is because the CA secret key should generally be kept apart from the token signing secret keys so that a compromise of a node does not lead to an attacker being able to generate valid signed Keystone tokens. This is a low probability attack vector, as compromise of a Keystone service machine’s filesystem security almost certainly means the attacker will be able to gain direct access to the token backend.
When using the keystone-manage pki_setup to generate the certificates, the following configuration options in the [signing] section are used:
If keystone-manage pki_setup is not used then these options don’t need to be set.
keystone-manage fernet_setup will attempt to create a key repository as configured in the [fernet_tokens] section of keystone.conf and bootstrap it with encryption keys.
A single 256-bit key is actually composed of two smaller keys: a 128-bit key used for SHA256 HMAC signing and a 128-bit key used for AES encryption. See the Fernet token specification for more detail.
keystone-manage fernet_rotate will rotate encryption keys through the following states:
When a new primary key is created, all new tokens will be encrypted using the new primary key. The old primary key is demoted to a secondary key, which can still be used for validating tokens. Excess secondary keys (beyond [fernet_tokens] max_active_keys) are revoked. Revoked keys are permanently deleted.
Rotating keys too frequently, or with [fernet_tokens] max_active_keys set too low, will cause tokens to become invalid prior to their expiration.
Keystone provides two configuration options for your service catalog.
A dynamic database-backed driver fully supporting persistent configuration.
keystone.conf example:
[catalog]
driver = sql
Note
A template_file does not need to be defined for the sql.Catalog driver.
To build your service catalog using this driver, see the built-in help:
$ openstack --help
$ openstack help service create
$ openstack help endpoint create
You can also refer to an example in Keystone (tools/sample_data.sh).
The templated catalog is an in-memory backend initialized from a read-only template_file. Choose this option only if you know that your service catalog will not change very much over time.
Note
Attempting to change your service catalog against this driver will result in HTTP 501 Not Implemented errors. This is the expected behavior. If you want to use these commands, you must instead use the SQL-based Service Catalog driver.
keystone.conf example:
[catalog]
driver = templated
template_file = /opt/stack/keystone/etc/default_catalog.templates
The value of template_file is expected to be an absolute path to your service catalog configuration. An example template_file is included in Keystone, however you should create your own to reflect your deployment.
Another such example is available in devstack (files/default_catalog.templates).
Endpoint Filtering enables creation of ad-hoc catalogs for each project-scoped token request.
Configure the endpoint filter catalog driver in the [catalog] section. For example:
[catalog]
driver = catalog_sql
In the [endpoint_filter] section, set return_all_endpoints_if_no_filter to False to return an empty catalog if no associations are made. For example:
[endpoint_filter]
return_all_endpoints_if_no_filter = False
See API Specification for Endpoint Filtering for the details of API definition.
Note
Support status for Endpoint Filtering
Experimental (Icehouse, Juno) Stable (Kilo)
Logging is configured externally to the rest of Keystone. Configure the path to your logging configuration file using the [DEFAULT] log_config_append option of keystone.conf. If you wish to route all your logging through syslog, set the [DEFAULT] use_syslog option.
A sample log_config_append file is included with the project at etc/logging.conf.sample. Like other OpenStack projects, Keystone uses the Python logging module, which includes extensive configuration options for choosing the output levels and formats.
A secure deployment should have Keystone running in a web server (such as Apache httpd), or behind an SSL terminator.
For the V2.0 API, Keystone provides an additional capability that allows users to use a HTTP PATCH to change their own password.
Each user can then change their own password with a HTTP PATCH :
$ curl -X PATCH http://localhost:5000/v2.0/OS-KSCRUD/users/<userid> -H "Content-type: application/json" \
-H "X_Auth_Token: <authtokenid>" -d '{"user": {"password": "ABCD", "original_password": "DCBA"}}'
In addition to changing their password all of the user’s current tokens will be revoked.
Keystone provides an optional capability to assign roles on a project or domain that, rather than affect the project or domain itself, are instead inherited to the project subtree or to all projects owned by that domain. This capability is enabled by default, but can be disabled by including the following in keystone.conf:
[os_inherit]
enabled = False
The Endpoint Policy feature provides associations between service endpoints and policies that are already stored in the Identity server and referenced by a policy ID.
Configure the endpoint policy backend driver in the [endpoint_policy] section. For example:
[endpoint_policy]
driver = sql
See API Specification for Endpoint Policy for the details of API definition.
Note
Support status for Endpoint Policy
Experimental (Juno) Stable (Kilo)
The OAuth 1.0a feature provides the ability for Identity users to delegate roles to third party consumers via the OAuth 1.0a specification.
To enable OAuth1:
[oauth1]
driver = sql
[auth]
methods = external,password,token,oauth1
WSGIPassAuthorization On
See API Specification for OAuth 1.0a for the details of API definition.
Note
Support status for OAuth 1.0a
Experimental (Havana, Icehouse) Stable (Juno)
The Revocation Events feature provides a list of token revocations. Each event expresses a set of criteria which describes a set of tokens that are no longer valid.
Add the revoke backend driver to the [revoke] section in keystone.conf. For example:
[revoke]
driver = sql
See API Specification for Revocation Events for the details of API definition.
Note
Support status for Revocation Events
Experimental (Juno) Stable (Kilo)
Token binding refers to the practice of embedding information from external authentication providers (like a company’s Kerberos server) inside the token such that a client may enforce that the token only be used in conjunction with that specified authentication. This is an additional security mechanism as it means that if a token is stolen it will not be usable without also providing the external authentication.
To activate token binding you must specify the types of authentication that token binding should be used for in keystone.conf e.g.:
[token]
bind = kerberos
Currently only kerberos is supported.
To enforce checking of token binding the enforce_token_bind parameter should be set to one of the following modes:
disabled disable token bind checking
permissive enable bind checking, if a token is bound to a mechanism that is unknown to the server then ignore it. This is the default.
strict enable bind checking, if a token is bound to a mechanism that is unknown to the server then this token should be rejected.
required enable bind checking and require that at least 1 bind mechanism is used for tokens.
named enable bind checking and require that the specified authentication mechanism is used. e.g.:
[token]
enforce_token_bind = kerberos
Do not set enforce_token_bind = named as there is not an authentication mechanism called named.
Keystone provides a method of setting a limit to the number of entities returned in a collection, which is useful to prevent overly long response times for list queries that have not specified a sufficiently narrow filter. This limit can be set globally by setting list_limit in the default section of keystone.conf, with no limit set by default. Individual driver sections may override this global value with a specific limit, for example:
[resource]
list_limit = 100
If a response to list_{entity} call has been truncated, then the response status code will still be 200 (OK), but the truncated attribute in the collection will be set to true.
In the future, keystone may offer the ability to identify a project in a hierarchy via a URL style of naming from the root of the hierarchy (for example specifying ‘projectA/projectB/projectC’ as the project name in an authentication request). In order to prepare for this, keystone supports the optional ability to ensure both projects and domains are named without including any of the reserverd characters specified in section 2.2 of rfc3986.
The safety of the names of projects and domains can be controlled via two configuration options:
[resource]
project_name_url_safe = off
domain_name_url_safe = off
When set to off (which is the default), no checking is done on the URL safeness of names. When set to new, an attempt to create a new project or domain with an unsafe name (or update the name of a project or domain to be unsafe) will cause a status code of 400 (Bad Request) to be returned. Setting the configuration option to strict will, in addition to preventing the creation and updating of entities with unsafe names, cause an authentication attempt which specifies a project or domain name that is unsafe to return a status code of 401 (Unauthorized).
It is recommended that installations take the steps necessary to where they can run with both options set to strict as soon as is practical.
The etc/ folder distributed with Keystone contains example configuration files for each Server application.
Like most OpenStack projects, Keystone supports the protection of its APIs by defining policy rules based on an RBAC approach. These are stored in a JSON policy file, the name and location of which is set in the main Keystone configuration file.
Each Keystone v3 API has a line in the policy file which dictates what level of protection is applied to it, where each line is of the form:
<api name>: <rule statement> or <match statement>
where:
<rule statement> can contain <rule statement> or <match statement>
<match statement> is a set of identifiers that must match between the token provided by the caller of the API and the parameters or target entities of the API call in question. For example:
"identity:create_user": "role:admin and domain_id:%(user.domain_id)s"
Indicates that to create a user you must have the admin role in your token and in addition the domain_id in your token (which implies this must be a domain scoped token) must match the domain_id in the user object you are trying to create. In other words, you must have the admin role on the domain in which you are creating the user, and the token you are using must be scoped to that domain.
Each component of a match statement is of the form:
<attribute from token>:<constant> or <attribute related to API call>
The following attributes are available
Attributes from token: user_id, the domain_id or project_id depending on the scope, and the list of roles you have within that scope
Attributes related to API call: Any parameters that are passed into the API call are available, along with any filters specified in the query string. Attributes of objects passed can be referenced using an object.attribute syntax (e.g. user.domain_id). The target objects of an API are also available using a target.object.attribute syntax. For instance:
"identity:delete_user": "role:admin and domain_id:%(target.user.domain_id)s"
would ensure that the user object that is being deleted is in the same domain as the token provided.
Every target object (except token) has an id and a name available as target.<object>.id and target.<object>.name. Other attributes are retrieved from the database and vary between object types. Moreover, some database fields are filtered out (e.g. user passwords).
List of object attributes:
The default policy.json file supplied provides a somewhat basic example of API protection, and does not assume any particular use of domains. For multi-domain configuration installations where, for example, a cloud provider wishes to allow administration of the contents of a domain to be delegated, it is recommended that the supplied policy.v3cloudsample.json is used as a basis for creating a suitable production policy file. This example policy file also shows the use of an admin_domain to allow a cloud provider to enable cloud administrators to have wider access across the APIs.
A clean installation would need to perhaps start with the standard policy file, to allow creation of the admin_domain with the first users within it. The domain_id of the admin domain would then be obtained and could be pasted into a modified version of policy.v3cloudsample.json which could then be enabled as the main policy file.
Ensure that your keystone.conf is configured to use a SQL driver:
[identity]
driver = sql
You may also want to configure your [database] settings to better reflect your environment:
[database]
connection = sqlite:///keystone.db
idle_timeout = 200
Note
It is important that the database that you specify be different from the one containing your existing install.
You should now be ready to initialize your new database without error, using:
$ keystone-manage db_sync
To test this, you should now be able to start keystone:
$ uwsgi --http 127.0.0.1:35357 --wsgi-file $(which keystone-wsgi-admin)
And use the OpenStack Client to list your projects (which should successfully return an empty list from your new database):
$ openstack --os-token ADMIN --os-url http://127.0.0.1:35357/v2.0/ project list
Note
We’re providing the default OS_TOKEN and OS_URL values from keystone.conf to connect to the Keystone service. If you changed those values, or deployed Keystone to a different endpoint, you will need to change the provided command accordingly.
keystone-manage is designed to execute commands that cannot be administered through the normal REST API. At the moment, the following calls are supported:
Invoking keystone-manage by itself will give you additional usage information.
The private key used for token signing can only be read by its owner. This prevents unauthorized users from spuriously signing tokens. keystone-manage pki_setup Should be run as the same system user that will be running the Keystone service to ensure proper ownership for the private key file and the associated certificates.
Keystone APIs are protected by the rules in the policy file. The default policy rules require admin credentials to administer users, projects, and roles. See section Keystone API protection with Role Based Access Control (RBAC) for more details on policy files.
The Keystone command line interface packaged in python-keystoneclient only supports the Identity v2.0 API. The OpenStack common command line interface packaged in python-openstackclient supports both v2.0 and v3 APIs.
With both command line interfaces there are two ways to configure the client to use admin credentials, using either an existing token or password credentials.
Note
As of the Juno release, it is recommended to use python-openstackclient, as it supports both v2.0 and v3 APIs. For the purpose of backwards compatibility, the CLI packaged in python-keystoneclient is not being removed.
Note
If your Keystone deployment is brand new, you will need to use this authentication method, along with your [DEFAULT] admin_token.
To authenticate with Keystone using a token and python-openstackclient, set the following flags.
To administer a Keystone endpoint, your token should be either belong to a user with the admin role, or, if you haven’t created one yet, should be equal to the value defined by [DEFAULT] admin_token in your keystone.conf.
You can also set these variables in your environment so that they do not need to be passed as arguments each time:
$ export OS_URL=http://localhost:35357/v2.0
$ export OS_TOKEN=ADMIN
Instead of python-openstackclient, if using python-keystoneclient, set the following:
To authenticate with Keystone using a password and python-openstackclient, set the following flags, note that the following user referenced below should be granted the admin role.
You can also set these variables in your environment so that they do not need to be passed as arguments each time:
$ export OS_USERNAME=my_username
$ export OS_PASSWORD=my_password
$ export OS_PROJECT_NAME=my_project
$ export OS_AUTH_URL=http://localhost:35357/v2.0
If using python-keystoneclient, set the following instead:
python-openstackclient is set up to expect commands in the general form of:
$ openstack [<global-options>] <object-1> <action> [<object-2>] [<command-arguments>]
For example, the commands user list and project create can be invoked as follows:
# Using token authentication, with environment variables
$ export OS_URL=http://127.0.0.1:35357/v2.0/
$ export OS_TOKEN=secrete_token
$ openstack user list
$ openstack project create demo
# Using token authentication, with flags
$ openstack --os-token=secrete --os-url=http://127.0.0.1:35357/v2.0/ user list
$ openstack --os-token=secrete --os-url=http://127.0.0.1:35357/v2.0/ project create demo
# Using password authentication, with environment variables
$ export OS_USERNAME=admin
$ export OS_PASSWORD=secrete
$ export OS_PROJECT_NAME=admin
$ export OS_AUTH_URL=http://localhost:35357/v2.0
$ openstack user list
$ openstack project create demo
# Using password authentication, with flags
$ openstack --os-username=admin --os-password=secrete --os-project-name=admin --os-auth-url=http://localhost:35357/v2.0 user list
$ openstack --os-username=admin --os-password=secrete --os-project-name=admin --os-auth-url=http://localhost:35357/v2.0 project create demo
In the SQL backend expired tokens are not automatically removed. These tokens can be removed with:
$ keystone-manage token_flush
The memcache backend automatically discards expired tokens and so flushing is unnecessary and if attempted will fail with a NotImplemented error.
As an alternative to the SQL Database backing store, Keystone can use a directory server to provide the Identity service. An example Schema for OpenStack would look like this:
dn: dc=openstack,dc=org
dc: openstack
objectClass: dcObject
objectClass: organizationalUnit
ou: openstack
dn: ou=Projects,dc=openstack,dc=org
objectClass: top
objectClass: organizationalUnit
ou: groups
dn: ou=Users,dc=openstack,dc=org
objectClass: top
objectClass: organizationalUnit
ou: users
dn: ou=Roles,dc=openstack,dc=org
objectClass: top
objectClass: organizationalUnit
ou: roles
The corresponding entries in the Keystone configuration file are:
[ldap]
url = ldap://localhost
user = dc=Manager,dc=openstack,dc=org
password = badpassword
suffix = dc=openstack,dc=org
use_dumb_member = False
allow_subtree_delete = False
user_tree_dn = ou=Users,dc=openstack,dc=org
user_objectclass = inetOrgPerson
The default object classes and attributes are intentionally simplistic. They reflect the common standard objects according to the LDAP RFCs. However, in a live deployment, the correct attributes can be overridden to support a preexisting, more complex schema. For example, in the user object, the objectClass posixAccount from RFC2307 is very common. If this is the underlying objectclass, then the uid field should probably be uidNumber and username field either uid or cn. To change these two fields, the corresponding entries in the Keystone configuration file are:
[ldap]
user_id_attribute = uidNumber
user_name_attribute = cn
There is a set of allowed actions per object type that you can modify depending on your specific deployment. For example, the users are managed by another tool and you have only read access, in such case the configuration is:
[ldap]
user_allow_create = False
user_allow_update = False
user_allow_delete = False
There are some configuration options for filtering users, tenants and roles, if the backend is providing too much output, in such case the configuration will look like:
[ldap]
user_filter = (memberof=CN=openstack-users,OU=workgroups,DC=openstack,DC=org)
In case that the directory server does not have an attribute enabled of type boolean for the user, there is several configuration parameters that can be used to extract the value from an integer attribute like in Active Directory:
[ldap]
user_enabled_attribute = userAccountControl
user_enabled_mask = 2
user_enabled_default = 512
In this case the attribute is an integer and the enabled attribute is listed in bit 1, so the if the mask configured user_enabled_mask is different from 0, it gets the value from the field user_enabled_attribute and it makes an ADD operation with the value indicated on user_enabled_mask and if the value matches the mask then the account is disabled.
It also saves the value without mask to the user identity in the attribute enabled_nomask. This is needed in order to set it back in case that we need to change it to enable/disable a user because it contains more information than the status like password expiration. Last setting user_enabled_mask is needed in order to create a default value on the integer attribute (512 = NORMAL ACCOUNT on AD)
In case of Active Directory the classes and attributes could not match the specified classes in the LDAP module so you can configure them like:
[ldap]
user_objectclass = person
user_id_attribute = cn
user_name_attribute = cn
user_description_attribute = displayName
user_mail_attribute = mail
user_enabled_attribute = userAccountControl
user_enabled_mask = 2
user_enabled_default = 512
user_attribute_ignore = tenant_id,tenants
For additional information on LDAP connections, performance (such as slow response time), or field mappings, setting debug_level in the [ldap] section is used to enable debugging:
debug_level = 4095
This setting in turn sets OPT_DEBUG_LEVEL in the underlying python library. This field is a bit mask (integer), and the possible flags are documented in the OpenLDAP manpages. Commonly used values include 255 and 4095, with 4095 being more verbose.
Warning
Enabling debug_level will negatively impact performance.
Some directory servers do not provide any enabled attribute. For these servers, the user_enabled_emulation attribute has been created. It is enabled by setting the respective flags to True. Then the attribute user_enabled_emulation_dn may be set to specify how the enabled users are selected. This attribute works by using a groupOfNames entry and adding whichever users or that you want enabled to the respective group with the member attribute. For example, this will mark any user who is a member of enabled_users as enabled:
[ldap]
user_enabled_emulation = True
user_enabled_emulation_dn = cn=enabled_users,cn=groups,dc=openstack,dc=org
The default values for user enabled emulation DN is cn=enabled_users,$user_tree_dn.
If a different LDAP schema is used for group membership, it is possible to use the group_objectclass and group_member_attribute attributes to determine membership in the enabled emulation group by setting the user_enabled_emulation_use_group_config attribute to True.
If you are using a directory server to provide the Identity service, it is strongly recommended that you utilize a secure connection from Keystone to the directory server. In addition to supporting LDAP, Keystone also provides Transport Layer Security (TLS) support. There are some basic configuration options for enabling TLS, identifying a single file or directory that contains certificates for all the Certificate Authorities that the Keystone LDAP client will recognize, and declaring what checks the client should perform on server certificates. This functionality can easily be configured as follows:
[ldap]
use_tls = True
tls_cacertfile = /etc/keystone/ssl/certs/cacert.pem
tls_cacertdir = /etc/keystone/ssl/certs/
tls_req_cert = demand
A few points worth mentioning regarding the above options. If both tls_cacertfile and tls_cacertdir are set then tls_cacertfile will be used and tls_cacertdir is ignored. Furthermore, valid options for tls_req_cert are demand, never, and allow. These correspond to the standard options permitted by the TLS_REQCERT TLS option.
Many environments typically have user and group information in directories that are accessible by LDAP. This information is for read-only use in a wide array of applications. Prior to the Havana release, we could not deploy Keystone with read-only directories as backends because Keystone also needed to store information such as projects, roles, domains and role assignments into the directories in conjunction with reading user and group information.
Keystone now provides an option whereby these read-only directories can be easily integrated as it now enables its identity entities (which comprises users, groups, and group memberships) to be served out of directories while resource (which comprises projects and domains), assignment and role entities are to be served from different Keystone backends (i.e. SQL). To enable this option, you must have the following keystone.conf options set:
[identity]
driver = ldap
[resource]
driver = sql
[assignment]
driver = sql
[role]
driver = sql
With the above configuration, Keystone will only lookup identity related information such users, groups, and group membership from the directory, while resources, roles and assignment related information will be provided by the SQL backend. Also note that if there is an LDAP Identity, and no resource, assignment or role backend is specified, they will default to LDAP. Although this may seem counter intuitive, it is provided for backwards compatibility. Nonetheless, the explicit option will always override the implicit option, so specifying the options as shown above will always be correct. Finally, it is also worth noting that whether or not the LDAP accessible directory is to be considered read only is still configured as described in a previous section above by setting values such as the following in the [ldap] configuration section:
[ldap]
user_allow_create = False
user_allow_update = False
user_allow_delete = False
Note
While having identity related information backed by LDAP while other information is backed by SQL is a supported configuration, as shown above; the opposite is not true. If either resource or assignment drivers are configured for LDAP, then Identity must also be configured for LDAP.
Various LDAP backends in Keystone use a common LDAP module to interact with LDAP data. By default, a new connection is established for each LDAP operation. This can become highly expensive when TLS support is enabled, which is a likely configuration in an enterprise setup. Reuse of connectors from a connection pool drastically reduces overhead of initiating a new connection for every LDAP operation.
Keystone provides connection pool support via configuration. This will keep LDAP connectors alive and reused for subsequent LDAP operations. The connection lifespan is configurable as other pooling specific attributes.
In the LDAP identity driver, Keystone authenticates end users via an LDAP bind with the user’s DN and provided password. This kind of authentication bind can fill up the pool pretty quickly, so a separate pool is provided for end user authentication bind calls. If a deployment does not want to use a pool for those binds, then it can disable pooling selectively by setting use_auth_pool to false. If a deployment wants to use a pool for those authentication binds, then use_auth_pool needs to be set to true. For the authentication pool, a different pool size (auth_pool_size) and connection lifetime (auth_pool_connection_lifetime) can be specified. With an enabled authentication pool, its connection lifetime should be kept short so that the pool frequently re-binds the connection with the provided credentials and works reliably in the end user password change case. When use_pool is false (disabled), then the authentication pool configuration is also not used.
Connection pool configuration is part of the [ldap] configuration section:
[ldap]
# Enable LDAP connection pooling. (boolean value)
use_pool=false
# Connection pool size. (integer value)
pool_size=10
# Maximum count of reconnect trials. (integer value)
pool_retry_max=3
# Time span in seconds to wait between two reconnect trials.
# (floating point value)
pool_retry_delay=0.1
# Connector timeout in seconds. Value -1 indicates indefinite wait for
# response. (integer value)
pool_connection_timeout=-1
# Connection lifetime in seconds. (integer value)
pool_connection_lifetime=600
# Enable LDAP connection pooling for end user authentication. If use_pool
# is disabled, then this setting is meaningless and is not used at all.
# (boolean value)
use_auth_pool=false
# End user auth connection pool size. (integer value)
auth_pool_size=100
# End user auth connection lifetime in seconds. (integer value)
auth_pool_connection_lifetime=60
Multiple LDAP server URLs can be provided to keystone to provide high-availability support for a single LDAP backend. To specify multiple LDAP servers, simply change the url option in the [ldap] section. The new option should list the different servers, each separated by a comma. For example:
[ldap]
url = "ldap://localhost,ldap://backup.localhost"
As of the Newton release, keystone encrypts all credentials stored in the default sql backend. Credentials are encrypted with the same mechanism used to encrypt Fernet tokens, fernet. Keystone provides only one type of credential encryption but the encryption provider is pluggable in the event you wish to supply a custom implementation.
This document details how credential encryption works, how to migrate existing credentials in a deployment, and how to manage encryption keys for credentials.
The configuration for credential encryption is straightforward. There are only two configuration options needed:
[credential]
provider = fernet
key_repository = /etc/keystone/credential-keys/
[credential] provider defaults to the only option supplied by keystone, fernet. There is no reason to change this option unless you wish to provide a custom credential encryption implementation. The [credential] key_repository location is a requirement of using fernet but will default to the /etc/keystone/credential-keys/ directory. Both [credential] key_repository and [fernet_tokens] key_repository define locations for keys used to encrypt things. One holds the keys to encrypt and decrypt credentials and the other holds keys to encrypt and decrypt tokens. It is imperative that these repositories are managed separately and they must not share keys. Meaning they cannot share the same directory path. The [credential] key_repository is only allowed to have three keys. This is not configurable and allows for credentials to be re-encrypted periodically with a new encryption key for the sake of security.
The implementation of this feature did not change any existing credential API contracts. All changes are transparent to the user unless you’re inspecting the credential backend directly.
When creating a credential, keystone will encrypt the blob attribute before persisting it to the backend. Keystone will also store a hash of the key that was used to encrypt the information in that credential. Since Fernet is used to encrypt credentials, a key repository consists of multiple keys. Keeping track of which key was used to encrypt each credential is an important part of encryption key management. Why this is important is detailed later in the Encryption key management section.
When updating an existing credential’s blob attribute, keystone will encrypt the new blob and update the key hash.
When listing or showing credentials, all blob attributes are decrypted in the response. Neither the cipher text, nor the hash of the key used to encrypt the blob are exposed through the API. Furthermore, the key is only used internally to keystone.
When upgrading a Mitaka deployment to Newton, three database migrations will ensure all credentials are encrypted. The process is as follows:
If performing a rolling upgrade, please note that a limited service outage will take affect during this migration. When the migration is in place, credentials will become read-only until the database is contracted. After the contract phase is complete, credentials will be writeable to the backend. A [credential] key_repository location must be specified through configuration and bootstrapped with keys using keystone-manage credential_setup prior to migrating any existing credentials. If a new key repository isn’t setup using keystone-manage credential_setup keystone will assume a null key to encrypt and decrypt credentials until a proper key repository is present. The null key is a key consisting of all null bytes and its only purpose is to ease the upgrade process from Mitaka to Newton. It is highly recommended that the null key isn’t used. It is no more secure than storing credentials in plain text. If the null key is used, you should migrate to a proper key repository using keystone-manage credential_setup and keystone-manage credential_migrate.
Key management of [credential] key_repository is handled with three keystone-manage commands:
keystone-manage credential_setup will populate [credential] key_repository with new encryption keys. This must be done in order for proper credential encryption to work, with the exception of the null key. This step should only be done once.
keystone-manage credential_rotate will create and rotate a new encryption key in the [credential] key_repository. This will only be done if all credential key hashes match the hash of the current primary key. If any credential has been encrypted with an older key, or secondary key, the rotation will fail. Failing the rotation is necessary to prevent overrotation, which would leave some credentials indecipherable since the key used to encrypt it no longer exists. If this step fails, it is possible to forcibly re-key all credentials using the same primary key with keystone-manage credential_migrate.
keystone-manage credential_migrate will check the backend for credentials whose key hash doesn’t match the hash of the current primary key. Any credentials with a key hash mismatching the current primary key will be re-encrypted with the current primary key. The new cipher text and key hash will be updated in the backend.