- 1 Introduction
- 2 Overview
- 3 Implementation Details
- 4 Additional Research
One of the major requirements for the IPA v2 release is to integrate the Certificate Authority into the IPA server. The main use case for this is provisioning certificates for use by applications, for example web server or IMAP server. There is no requirement to provision user certificates. This is deferred to later version of IPA. Also there is no plan to deploy certificates to all the IPA clients. Instead the clients will be using kerberos keytab to authenticate to IPA server.
Certificate Management Interfaces
Certificate life cycle management includes the following basic operations:
- Requesting certificates. If a new certificate needs to be issued it should be possible to request a new certificate from the system. The assumption is that the keypair is generated on the client system, and the client is submitting a public key to the CA. In IPA v2 we are not planning to support key escrow use case and server side key generation. This will be added later when add support for user certificate authentication.
- Checking request status. In different situations the certificate issuance can be delayed pending administrative approval. There is a need to be able to track and check the status of outstanding requests.
- Getting certificate. When the request is approved there should be a way to retrieve the certificate and store it in the appropriate location.
- Revoking certificates. If the certificate is lost or compromised it should be possible to revoke it so that it can not be used any more. There is also an option to not completely revoke certificate by put it on hold. The example of the situation would be when the user left his smart card at home. His certificates can be put on hold till next day.
- Take certificate off "hold". When the certificate is on hold take it off hold.
All these operations are performed by the Certificate Authority (CA). To prevent direct access to the CA by IPA clients, the CA is front-ended with the Registration Authority (RA). It will be impossible to request anything from the CA bypassing RA. The RA will be embedded into the XML-RPC back end as an XML-RPC back end plugin. This plugin will be responsible for validating the permission of the client and proxying its request to the CA through RA, and then collecting the response and sending it back.
The back end management plugin can be used for management of the certificates in the system through UI and CLI. For v2 we are planning to allow certificates to be associated with a host. If there are multiple services running on the same host there might be a need to issue multiple certificates. There are several approaches how this can be done:
- The simplest approach would be to allow just multiple certificates to be issued for one host. In this case the CA will publish the issued certificates into the multi valued attribute in the host entry. There is a concern about this solution that is currently being investigated. It is related to unpublishing the certificate when it is revoked. I might be that the code responsible for this feature is not capable of removing the certificate from the multi value attribute. If this is the case then we would have to solve the problem by publishing the certificates into separate entries.
- The second approach is to have child entries under host entries where the certificates will be stored. This way there will be no problems with publishing and unpublishing the certificates but we would a bit violate the flat approach to our tree.
- The third option would be to put the certificate into the service entry. Conceptually service entries are now used to store a kerberos principal's keys. Adding auxiliary class to store certificate attribute to the service entry might also be a viable option.
- The last option is not to publish the certificates at all and for any administrative need query the CA using RA. This would be a bit of performance burden and would require RA to expose a more robust querying interface than currently planned.
Currently we are leaning towards the combination of the first and third options. If the first option is not available we might decide to allow one certificate for a host entry and if there are more needed the certificates can be issued to the services running on the host in this case we would have to modify the service to alternatively be a container for the kerberos data and for the certificate.
It should be possible to perform the following tasks via either management interface:
- List the certificates issued to a host. If the certificates are issued to the host itself then the command or UI would be based on the host identity. In CLI the command can be something like:
>ipa-certs list --host=my.host.somewhere.com
In the UI it can be a section in the edit/view host screen. Since the UI will use AJAX we will be able to have this section collapsed by default and retrieve cert information only explicitly on demand if we see performance issues (especially if end up not publishing the certs at all and would have to retrieve information form CA itself). If we put the certificates into the host entry, the command will just retrieve the entry and list the certificates that are stored in the multi valued attribute. If we have to put the certificates in special entries under host, the host management plugin would have to enumerate these entries and build the certificate list this way. If we decide to store the certificates in services, we would have to list the services themselves in the UI. The services will have the fully qualified host name as a part of the principal and this is how they can be matched and associated to the host they are running on. If we decide not to publish certs at all we would have to have the RA request that would ask the CA about all the certificates issued to a given host.
- Get status of a certificate. Display the properties of the certificate including host name, serial number, expiration etc. In UI there should be a button that would allow selecting certificate from the list and clicking “details” button. On command line this should be a command like:
>ipa-certs list --serial=<serial number>
- Revoke a certificate. In the UI there should be a button that would allow selecting certificate from the list and clicking “revoke” button. On command line this should be a command like:
>ipa-certs revoke --serial=<serial number>
- Renew certificate. If the certificate is going to expire there should be a way to issue a replacement certificate. In the UI there should be a button that would allow selecting a certificate from the list and clicking “renew” button. On command line this should be a command like:
>ipa-certs renew --serial=<serial number>
- Get a new certificate. There should be a way to request a certificate for host (or service) from the UI and command line.
The certificate operations can be performed only via RA. The management plugin that interacts with RA would have to check if the user is allowed to perform certificate operations. In IPA v2 we decided that we will not have a granular access control over individual operations. Instead we will check that a user who is trying to execute a certificate command is a member of the specific group. If he is a member of the group then he would be able to perform any of the certificate related operations. The name of the group will default to "administrators" but would be changeable in a specific configuration entry (see below).
Since in v2 we are not planning to support certificate operations with user certificates not all operations that will be implemented would actually be used and exposed via UI and CLI. For example pulling on hold and talking off hold will not possible in v2 despite the fact that RA and management plugin will be aware of this functionality. Also there will be no approval process. This means that any request for a certificate will be automatically approved. In this situation a request for new or renewed certificate should be satisfied by CA within seconds. It would be a responsibility of the management plugin to issue request and then immediately retrieve the certificate. If might require some polling logic to make sure that we do not time out a bit prematurely if it took a bit longer for CA to issue a certificate.
Automatic Cert Provisioning
So far we have been talking about the administrative tasks performed by the administrator via UI and CLI. In those scenarios the administrator is the actor and his access control properties are checked before a certificate operation is granted. Potentially an administrator (depending upon his privileges of cause) can perform all the certificate operations including issuance, renewal and revocation. Any of these operations is manual and requires user authentication before it can be performed.
There is one more use case when the certificate shall be issued or renewed but the request is not authorized by user (administrator) but rather by a host. The use case would be as following:
- IPA enabled host runs a web application that needs certificate for its functioning. The certificate provisioning can be scripted as a part of the installation.
- IPA enabled host (if told to) would renew certificate on behalf of the application if it detects that the certificate is about to expire.
We will talk about how it would be actually accomplished below.
After some evaluation we agreed that there will be limitations of what a host can request. We decided that it can request only certificate for itself (or service running on the same host) and not for any other host. It can also can't automatically revoke certificates since in case of a bug or some misconfiguration it can have unrepairable impact on the customer environment.
IPA Client Design
To be able to request a certificate from the IPA client automatically there should be a utility tightly integrated with the IPA client that would aid requesting the certificate for a service running on the host. The following diagram shows all the IPA client components involved in the operation.
- The cert utility, named ipa-get-cert on the diagram, will accept command line parameters and store them in the LDB. This is the only thing it will do.
- The XML-RPC Client (formerly known as policy downloader) will be a daemon. It will be event based as any other service on the client designed so far. It will wake up periodically and see if there is anything it needs to do. Originally we planned that the XML-RPC Client will be responsible for only dealing with policies, but now its responsibilities will broaden (thus the name change). The XML-RPC Client being a component capable of establishing XML-RPC connection will now be in charge of requesting certificates too, since that can be done only through XML-RPC interface. So when XML-RPC Client wakes up it will check if it is time to get policies or do something with certificates. Here we really have several options on how we will implement the flow of the operations. One scenario is shown on the diagram but there are others that we will discuss below.
- So as shown on the diagram the XML-RPC Client will wake up and if it is time for check for cert tasks ask the Data Provider is there is anything to do with the certs. The data provider will get data from LDB.
- Before sending it back it might check if the cert is already provisioned into the LDAP entry and pull it from there or do some other checks if needed. This is really an optional step and implementation would decide if it is needed.
- The IPA Data Provider will respond to the XML-RPC Client. It can respond that there is nothing to do (which might be caused by the fact that there is actually nothing to do or by the fact that the client is offline), it can hand the XML-RPC Client back the parameters of the request it should submit over XML-RPC or it can return the certificate if we decide that step 4 makes sense.
- If there is work to do the XML-RPC Client will formulate a request to the IPA server over the XML-RPC connection.
- The results of the request need to be recorded in the LDB so the XML-RPC Client would call the Data Provider to request update to the LDB.
Alternatively if we think that the XML-RPC Client would not need the Data Provider to do any lookups the XML-RPC Client will directly interact with LDB but it still needs to ask data provider about the offline status. The Data Provider is the only service that has authority to determine the offline status. So the XML-RPC Client logic might be:
- Wake up and determine what operation it is supposed to do: certs or policies (let us say certs)
- Ask data provider is it is offline. If yes – sleep. If no continue.
- Check if there is the cert related data to process. If no – sleep. If yes continue.
- If the data shows that XML-RPC request needs to be issued, then issue request and process the results potentially updating LDB. If the data shows that there needs to be an LDAP lookup, then request the LDAP lookup from IPA Data Provider over DBUS interface. Process results. Update LDB.
The exact logic will be figured out during the implementation but the following main points should be assumed and not changed:
- The XML-RPC Client will be a daemon (fork-exec by the service controller).
- The XML-RPC Client will have a main loop that will be event based and events will be triggered by time elapsed or emitted at the start of the daemon.
- The XML-RPC Client will be a client of the IPA data provider as all other services (PAM responder, NSS responder etc.) are.
- The XML-RPC Client is in charge of the XML-RPC connection no other process should be aware how to do XML-RPC.
- The XML-RPC Client is in charge of all operations that can be done only via XML-RPC. This now includes not only getting policies but also requesting operations from CA via RA.
- The XML-RPC Client may or may not directly connect to LDB. It definitely would directly connect to the configuration instance of LDB. If we would save the certificate related data there we would not need to use data provider to connect to it. If we decide to store this data in the LDB instance that stores other information we might decide that it would be beneficial to access this data through data provider.
Server Policy about Clients
Though the clients would have the capability to request new certificates or track and renew old ones it does not mean that the server would blindly respect and execute these requests. The management plugin that interfaces with RA would have to determine that the current XML-RPC request is executed under host principal and not by an administrative user (see above). In the same configuration entry that will hold the name of the group of user that would have access to certificate operations, there will be an attribute that would hold the policy that will control how server would react to the client requests to issue or renew a certificate. In IPA v2 there will be 3 supported values.
- Always - always respect host's requests
- Never (default) - automatically ignore all requests coming from hosts
- Renew - respect only renew requests. The RA uses one and the same call for issuance and renewal so it is hard to differentiate the two scenarios. To react differently in case of certificate renewal the plugin will check the host (or service entry) for a certificate. If the certificate already exists in the entry, then the request will be treated as a renewal. If there is no certificate in the entry then it would be treated as a new request.
The list of the values can be later extended if needed.
Command Line Utility
Now it is time to talk about the data that utility would collect. At its core, the tool's function is very similar to that of ipa-getkeytab, so we're aiming for a command-line interface which feels similar.
There will be only four commands that the utility would be able to issue:
- Request certificate and track (by default or not track if told not to) its expiration.
- Start tracking expiration of an already provisioned certificate
- Stop tracking expiration of an already provisioned certificate
- Status – list current pending requests and/or currently tracked certs
In the case of requesting a certificate the command might look like this (the specific details will be determined at the implementation phase):
ipa-getcert request [options] * If the client can conceive of more than one CA: -c location of CA (format TBD) * If we're generating a key: -g generate a new key (default: use an already-generated key) -G size size of new key (default TBD) * Whether we're generating a key or not: -d DIR database containing / for storing private key and cert (NSS) -n NAME nickname to give issued certificate (only valid with -d) -k PATH file containing / for storing PEM private key -f PATH file for storing issued PEM certificate (only valid with -k) * Whether or not to track expiration: --no-track-expiration * Optional bits: -S NAME requested subject name (default: CN=<fqdn>) -u USAGE requested usage/eku (default: tls-server) -s NAME requested service name part (default: host)
In case of starting to track a certificate's expiration, one shall provide the following command line:
ipa-getcert start-tracking [options] * General options: -d DIR database containing private key and cert (NSS) -n NAME nickname of issued certificate (only valid with -d) -k PATH file containing PEM private key -f PATH file containing / for storing issued PEM certificate (only valid with -k) * If the client can conceive of more than one CA: -c ARG location of CA (format TBD)
The utility will make sure that a certificate in the given format and place already exists and then save the information about it in the LDB.
In case of stopping the tracking of a certificate's expiration, it would be:
ipa-getcert stop-tracking [options] * General options: -s NUM serial number of certificate -S KEYID subject key identifier for certificate * In case the serial number corresponds to more than one certificate, and the key identifier is not known: -d DIR database containing private key and cert (NSS) -n NAME nickname of issued certificate (only valid with -d) -k PATH file containing PEM private key -f PATH file containing / for storing issued PEM certificate (only valid with -k)
To stop tracking of the cert expiration the utility will just remove an entry that corresponds to the given cert from the LDB.
The status commend will list the contents of the cert data in the LDB.
ipa-getcert list [options] * General options: --requests List only information about outstanding requests --tracking List only information about tracked certificates
Data Stored in LDB
The data stored in the LDB will look like this:
- Entry that will contain request for a new certificate
- Location of the CA (if more than one can be known to the client)
- Server-supplied identifier for tracking the state of the request
- Date when the request was submitted
- Format that the certificate shall be saved in
- Path where the certificate shall be saved
- Should its expiration be tracked or not once it's issued
- Format that the private key is stored in (for later)
- Path where the private key can be found (for later)
- Entry that will contain expiration tracking information
- Serial number – retrieved from the issued certificate
- Subject name – retrieved from the issued certificate
- Subject key identifier – retrieved / calculated from the issued certificate
- Certificate expiration date – retrieved from the issued certificate
- Format that the private key is stored in (for generating a new request)
- Path where the private key can be found (for generating a new request)
- Format that the certificate is saved in
- Path where the certificate was saved
The policy downloader will look at the data taken from these LDB entries and take appropriate action.
Proposed Administrative Interfaces
The following administrative utilities are proposed. These commands are lacking details. For example the certificate request should contain information whether we are requesting the certificate for the service or for the host itself.
./ipa request-certificate [--ca=<ca_plugin_name>] [--request_type=<request_type>] <request> where <ca_plugin_name> 'ipa-ca' is default backend plugin accessing IPA's internal CA <request_type> 'pkcs10' is a default request type supported by default CA plugin <request> certificate request returning error_code, error_message, issued_certificate ./ipa revoke-certificate [--ca=<ca_plugin_name>] [--reason=<revocation_reason>] <serial_number> where <ca_plugin_name> 'ipa-ca' is default backend plugin accessing IPA's internal CA <serial_number> certificate serial number of the certificate to be revoked <revocation_reason> certificate revocation reason returning error_code, error_message ./ipa take-certificate-off-hold [--ca=<ca_plugin_name>] <serial_number> where <ca_plugin_name> 'ipa-ca' is default backend plugin accessing IPA's internal CA <serial_number> certificate serial number of the certificate to be taken off hold returning error_code, error_message ./ipa check_request_status [--ca=<ca_plugin_name>] <request_id> where <ca_plugin_name> 'ipa-ca' is default backend plugin accessing IPA's internal CA <request_id> is request id of the request to be verified returning error_code, error_message, certificate_serial_number ./ipa get-certificate [--ca=<ca_plugin_name>] <serial_number> where <ca_plugin_name> 'ipa-ca' is default backend plugin accessing IPA's internal CA <serial_number> certificate serial number (of previously generated certificate) to be retrieved returning error_code, error_message, issued_certificate
Multiple CA Support
You can customize this list to provide required certificate management capabilities to IPA.
The IPA server is very extensible and pluggable. In IPA v2 we plan to embed the CA but that does not mean that other CA's can't be used. To support other CAs the only change required is a different management plugin that will implement interfaces described below. The --ca option allows selecting the plugin that will serve the request. This means that one would be able to add support to any number of CAs without changes to UI or CLI.
These are preliminary interfaces. They might need to be extended to reflect other parameters. For example we would need to add the name of the service a certificate is requested for.
def request_certificate(self, certificate_request=None, request_type="pkcs10"): # . . . return (error_code, error_message, issued_certificate) def revoke_certificate(self, serial_number=None, revocation_reason=0): # . . . return (error_code, error_message) def take_certificate_off_hold(self, serial_number=None): # . . . return (error_code, error_message) def check_request_status(self, request_id=None): # . . . return (error_code, error_message, certificate_serial_number) def get_certificate(self, serial_number=None): # . . . return (error_code, error_message, issued_certificate)
The configuration entry most likely will be created under cn=etc. The configuration entry will contain two attributes:
Standard "manager", "member" or "owner" attribute. This attribute has a DN syntax. It will point to DN of the group of users that are allowed to issue certificate commands. By default it will point to DN of the "admins" group that is created by default at the installation. If not present reference to "admins" group should be assumed. If we choose "manager" and want the referential integrity plugin to be able to track changes we would have to add "manager" attribute to the list of the attributes tracked by the referential integrity plugin. The "member" and "owner" attributes are already listed in the referential integrity attribute.
attribute ( 2.16.840.1.1137126.96.36.199.TBD NAME 'hostCApolicy' DESC 'Policy on how to treat host requests for cert operations.' EQUALITY caseIgnoreMatch ORDERING caseIgnoreMatch SUBSTR caseIgnoreSubstringsMatch SYNTAX 188.8.131.52.4.1.14184.108.40.206.15 SINGLE-VALUE X-ORIGIN 'IPA v2')
If not present "Never" should be assumed.
Object class will look like this:
objectclass ( 2.16.840.1.1137220.127.116.11.TBD NAME 'ipaCAaccess' STRUCTURAL MAY (member $ hostCApolicy) X-ORIGIN 'IPA v2' )
CA provides three authentication methods to access publishing directory:
1. basic authentication - using specified DN with password
2. basic authentication over SSL - using specified DN with password over SSL
3. client authentication - using client certificate over SSL
We have decided to automatically create a special CA administrative user in pretty much the same way we create the "kdc" account during installation.
This account will look like this:
# CAadmin, sysaccounts, etc, example.com dn: uid=CAadmin,cn=sysaccounts,cn=etc,dc=example,dc=com objectClass: account objectClass: simplesecurityobject objectClass: top uid: CAadmin userPassword: ...
The account like this is currently used by KDC to connect to the DS. Similar approach should be taken by the CA. CA will use this account to bind to DS.