|
|
The following sections describe the various security capabilities available with the Oracle Tuxedo system for ATMI applications:
| Note: | The Oracle Tuxedo product includes environments that allow you to build both Application-to-Transaction Monitor Interfaces (ATMI) and CORBA applications. This topic explains how to implement security in an ATMI application. For information about implementing security in a CORBA application, see Using Security in CORBA Applications. |
Security refers to techniques for ensuring that data stored in a computer or passed between computers is not compromised. Most security measures involve passwords and data encryption, where a password is a secret word or phrase that gives a user access to a particular program or system, and data encryption is the translation of data into a form that is unintelligible without a deciphering mechanism.
Distributed applications such as those used for electronic commerce (e-commerce) offer many access points for malicious people to intercept data, disrupt operations, or generate fraudulent input; the more distributed a business becomes, the more vulnerable it is to attack. Thus, the distributed computing software, or middleware, upon which such applications are built must provide security.
The Oracle Tuxedo product provides several security capabilities for ATMI applications, most of which can be customized for your particular needs.
As shown in the following figure, all but one of the security capabilities available with the ATMI environment of the Oracle Tuxedo product are implemented through a plug-in interface, which allows Oracle Tuxedo customers to independently define and dynamically add their own security plug-ins. A security plug-in is a code module that implements a particular security capability.
The specifications for the security plug-in interface are not generally available, but are available to third-party security vendors. Third-party security vendors can enter into a special agreement with Oracle Systems to develop security plug-ins for Oracle Tuxedo. Oracle Tuxedo customers who want to customize a security capability must contact one of these vendors. For example, an Oracle Tuxedo customer who wants a custom implementation of public key security must contact a third-party security vendor who can provide the appropriate plug-ins. For more information about security plug-ins, including installation and configuration procedures, see your Oracle account executive.
The Oracle Tuxedo system can enforce security in a number of ways, which includes using the security features of the host operating system to control access to files, directories, and system resources. The following table describes the security capabilities available with the ATMI environment of the Oracle Tuxedo product.
The
default authentication plug-in provides security at three levels: no authentication, application password, and user-level authentication. This plug-in works the same way the Oracle Tuxedo implementation of authentication has worked since it was first made available with the Oracle Tuxedo system.
|
|||
The
default authorization plug-in provides security at two levels: optional access control lists and mandatory access control lists. This plug-in works the same way the Oracle Tuxedo implementation of authorization has worked since it was first made available with the Oracle Tuxedo system.
|
|||
Uses
symmetric key encryption to establish data privacy for messages moving over the network links that connect the machines in an ATMI application.
|
|||
Uses public key (or
asymmetric key) encryption to establish end-to-end digital signing and data privacy between ATMI application clients and servers. Complies with the PKCS-7 standard.
|
Implemented as
six interfaces
|
On host operating systems with underlying security features, such as file permissions, the operating-system level of security is the first line of defense. An application administrator can use file permissions to grant or deny access privileges to specific users or groups of users.
Most ATMI applications are managed by an application administrator who configures the application, starts it, and monitors the running application dynamically, making changes as necessary. Because the ATMI application is started and run by the administrator, server programs are run with the administrator’s permissions and are therefore considered secure or “trusted.” This working method is supported by the login mechanism and the read and write permissions on the files, directories, and system resources provided by the underlying operating system.
Client programs are run directly by users with the users’ own permissions. In addition, users running native clients (that is, clients running on the same machine on which the server program is running) have access to the UBBCONFIG configuration file and interprocess communication (IPC) mechanisms such as the bulletin board (a reserved piece of shared memory in which parameters governing the ATMI application and statistics about the application are stored).
For ATMI applications running on platforms that support greater security, a more secure approach is to limit access to the files and IPC mechanisms to the application administrator and to have “trusted” client programs run with the permissions of the administrator (using the setuid command on a UNIX host machine or the equivalent command on another platform). For the most secure operating system security, allow only Workstation clients to access the application; client programs should not be allowed to run on the same machines on which application server and administrative programs run.
Authentication allows communicating processes to mutually prove identification. The authentication plug-in interface in the ATMI environment of the Oracle Tuxedo product can accommodate various security-provider authentication plug-ins using various authentication technologies, including shared-secret password, one-time password, challenge-response, and Kerberos. The interface closely follows the generic security service (GSS) application programming interface (API) where applicable; the GSSAPI is a published standard of the Internet Engineering Task Force. The authentication plug-in interface is designed to make integration of third-party vendor security products with the Oracle Tuxedo system as easy as possible, assuming the security products have been written to the GSSAPI.
The underlying plug-in interface for authentication security is implemented as a single plug-in. The plug-in may be the default authentication plug-in or a custom authentication plug-in.
Direct end-to-end mutual authentication in a distributed enterprise middleware environment such as the Oracle Tuxedo system can be prohibitively expensive, especially when accomplished with security mechanisms optimized for long-duration connections. It is not efficient for clients to establish direct network connections with each server process, nor is it practical to exchange and verify multiple authentication messages as part of processing each service request. Instead, the ATMI applications use a delegated trust authentication model, as shown in the following figure.
A Workstation client authenticates to a trusted system gateway process, the workstation handler (WSH), at initialization time. A native client authenticates within itself, as explained later in this discussion. After a successful authentication, the authentication software assigns a security token to the client. A token is an opaque data structure suitable for transfer between processes. The WSH safely stores the token for the authenticated Workstation client, or the authenticated native client safely stores the token for itself.
As a client request flows through a trusted gateway, the gateway attaches the client’s security token to the request. The security token travels with the client’s request message, and is delivered to the destination server process(es) for authorization checking and auditing purposes.
In this model, the gateway trusts that the authentication software will verify the identity of the client and generate an appropriate token. Servers, in turn, trust that the gateway process will attach the correct security token. Servers also trust that any other servers involved in the processing of a client request will safely deliver the token.
The following figure shows the control flow inside the ATMI environment of the Oracle Tuxedo system while a session is being established between a Workstation client and the WSH. The Workstation client and WSH are attempting to establish a long-term mutually authenticated connection by exchanging messages.
The initiator process (may be thought of as a middleware client process) creates a session context by repeatedly calling the Oracle Tuxedo “initiate security context” function until a return code indicates success or failure. A session context associates identity information with an authenticated user.
When a Workstation client calls tpinit(3c) for C or TPINITIALIZE(3cbl) for COBOL to join an ATMI application, the Oracle Tuxedo system begins its response by first calling the internal “acquire credentials” function to obtain a session credential handle, and then calling the internal “initiate security context” function to obtain a session context. Each invocation of the “initiate security context” function takes an input session token (when one is available) and returns an output session token. A session token carries a protocol for verifying a user’s identity. The initiator process passes the output session token to the session’s target process (WSH), where it is exchanged for another input token. The exchange of tokens continues until both processes have completed mutual authentication.
A security-provider authentication plug-in defines the content of the session context and session token for its security implementation, so ATMI authentication must treat the session context and session token as opaque objects. The number of tokens passed back and forth is not defined, and may vary based on the architecture of the authentication system.
For a native client initiating a session, the initiator process and the target process are the same; the process may be thought of as a middleware client process. The middleware client process calls the security provider’s authentication plug-in to authenticate the native client.
After a successful authentication, the trusted gateway calls two Oracle Tuxedo internal functions that retrieve an authorization token and an auditing token for the client, which the gateway stores for safekeeping. Together, these tokens represent the user identity of a security context. The term security token refers collectively to the authorization and auditing tokens.
When default authentication is used, the authorization token carries two pieces of information:
In addition, when default authentication is used, the auditing token carries the same two pieces of information: principal name and application key.
Like the session token, the authentication and auditing tokens are opaque; their contents are determined by the security provider. The authorization token can be used for performing authorization (permission) checks. The auditing token can be used for recording audit information. In some ATMI applications, it is useful to keep separate user identities for authorization and auditing.
As shown in the following figure, there are situations where a client service request forwarded by a server takes on the identity of the server. The server replaces the client tokens attached to the request with its own tokens and then forwards the service request to the destination service.
| Note: | See Specifying Principal Names for an understanding of how servers acquire their own authorization and auditing tokens and why they need them. |
The feature demonstrated in the preceding figure is known as server permission upgrade, which operates in the following manner: whenever a server calls a dot service (a system-supplied service having a beginning period in its name—such as .TMIB), the service request takes on the identity of the server and thus acquires the access permissions of the server. A server’s access permissions are those of the application (system) administrator. Thus, certain requests that would be denied if the client called the dot service directly would be allowed if the client sent the requests to a server, and the server forwarded the requests to the dot service. For more information about dot services, see the .TMIB service description on the MIB(5) reference page in the Oracle Tuxedo File Formats, Data Descriptions, MIBs, and System Processes Reference.
You can provide authentication for your ATMI application by using the default plug-in or a custom plug-in. You choose a plug-in by configuring the Oracle Tuxedo registry, a tool that controls all security plug-ins.
If you want to use the default authentication plug-in, you do not need to configure the registry. If you want to use a custom authentication plug-in, however, you must configure the registry for your plug-in before you can install it. For more detail about the registry, see Setting the Oracle Tuxedo Registry.
Authorization allows administrators to control access to ATMI applications. Specifically, an administrator can use authorization to allow or disallow principals (authenticated users) to use resources or facilities in an ATMI application.
A fanout is an umbrella plug-in to which individual plug-in implementations are connected. As shown in the following figure, the authorization plug-in interface is implemented as a fanout.
The default authorization implementation consists of a fanout plug-in and a default authorization plug-in. A custom implementation consists of the fanout plug-in, the default authorization plug-in, and one or more custom authorization plug-ins.
In a fanout plug-in model, a caller sends a request to the fanout plug-in. The fanout plug-in passes the request to each of the subordinate plug-ins, and receives a response from each. Finally, the fanout plug-in forms a composite response from the individual responses, and sends the composite response to the caller.
The purpose of an authorization request is to determine whether a client operation should be allowed or whether the results of an operation should be kept unchanged. Each authorization plug-in returns one of three responses: permit, deny, or abstain. The abstain response gives writers of authorization plug-ins a graceful way to handle situations that are not accommodated by the original plug-in, such as names of operations that are added to the system after the plug-in is installed.
The authorization fanout plug-in forms a composite response as described in the following table. For default authorization, the composite response is determined solely by the default authorization plug-in.
As an example of custom authorization, consider a banking application in which a user is identified as a member of the Customer group, and the following conditions are in effect:
Customer group to withdraw money from a particular account.Customer group to withdraw money from a particular account but only on Monday through Friday between 9:00 A.M. and 5:00 P.M.Customer group to withdraw money from a particular account but only if the amount being withdrawn is less than $10,000.
So, if a user in the Customer group attempts to withdraw $500.00 on Monday at 10 A.M., the operation is allowed. If the same user attempts the same withdrawal on Saturday morning, the operation is not allowed.
Many other custom authorization scenarios are possible. Feel free to improvise; define the conditions that best serve the needs of your business.
Authorization decisions are based partly on user identity, which is stored in an authorization token. Because authorization tokens are generated by the authentication security plug-in, providers of authentication and authorization plug-ins need to ensure that these plug-ins work together.
An Oracle Tuxedo system process or server (such as /Q server TMQUEUE(5) or EventBroker server TMUSREVT(5)) calls the authorization plug-in when it receives a client request. In response, the authorization plug-in performs a pre-operation check and returns whether the operation should be allowed.
If the client operation is allowed, the Oracle Tuxedo system process or server may call the authorization plug-in after the client operation completes. In response, the authorization plug-in performs a post-operation check and returns whether the results of the operation are acceptable.
These calls are system-level calls, not application-level calls. An ATMI application cannot call the authorization plug-in.
The authorization process is somewhat different for (1) users of the default authorization plug-in provided by the Oracle Tuxedo system and (2) users of one or more custom authorization plug-ins. The default plug-in does not support post-operation checks. If the default authorization plug-in receives a post-operation check request, it returns immediately and does nothing.
The custom plug-ins support both pre-operation and post-operation checks.
When default authorization is called by an ATMI process to perform a pre-operation check in response to a client request, the authorization plug-in performs the following tasks.
Because the authorization token is created by the authentication plug-in, the authorization plug-in has no record of the token’s content. This information is necessary for the authorization process.
The authorization plug-in determines whether that operation should be allowed by examining the client’s authorization token, the access control list (ACL), and the configured security level (optional or mandatory ACL) of the ATMI application.
The authorization fanout plug-in receives a decision (permit or deny) from the default authorization plug-in and operates on its behalf.
Users of one or more custom authorization plug-ins may take advantage of additional functionality offered by the ATMI environment of the Oracle Tuxedo product. Specifically, the custom plug-ins may perform an additional check after an operation occurs.
When custom authorization is called by an ATMI process to perform a pre-operation check in response to a client request, the authorization plug-in performs the following tasks.
The authorization plug-in determines whether the operation should be allowed by examining the operation, the client’s authorization token, and associated data. “Associated data” may include user data and the security level of the ATMI application.
If necessary, in order to satisfy authorization requirements, the authorization plug-in may modify the user data before the operation is performed.
The authorization fanout plug-in makes the ultimate decision by checking the individual responses (permit, deny, abstain) of its subordinate plug-ins.
If the client operation is allowed, custom authorization may be called by the ATMI process to perform a post-operation check after the client operation completes. If so, the authorization plug-in performs the following tasks.
The authorization plug-in determines whether the operation results are acceptable by examining the operation, the client’s authorization token, and associated data. “Associated data” may include user data and the security level of the ATMI application.
The authorization fanout plug-in makes the ultimate decision by checking the individual responses (permit, deny, abstain) of its subordinate plug-ins.
A post-operation check is useful for label-based security models. For example, suppose that a user is authorized to access CONFIDENTIAL documents but performs an operation that retrieves a TOP SECRET document. (Often, a document’s classification label is not easily determined until after the document has been retrieved.) In this case, the post-operation check is an efficient means to either deny the operation or modify the output data by expunging any restricted information.
You can provide authorization for your ATMI application by using the default plug-in or adding one or more custom plug-ins. You choose a plug-in by configuring the Oracle Tuxedo registry, a tool that controls all security plug-ins.
If you want to use the default authorization plug-in, you do not need to configure the registry. If you want to add one or more custom authorization plug-ins, however, you must configure the registry for your additional plug-ins before you can install them. For more detail about the registry, see Setting the Oracle Tuxedo Registry.
Auditing provides a means to collect, store, and distribute information about operating requests and their outcomes. Audit-trail records may be used to determine which principals performed, or attempted to perform, actions that violated the security levels of an ATMI application. They may also be used to determine which operations were attempted, which ones failed, and which ones successfully completed.
How auditing is done (that is, how information is collected, processed, protected, and distributed) depends on the auditing plug-in.
A fanout is an umbrella plug-in to which individual plug-in implementations are connected. As shown in the following figure, the auditing plug-in interface is implemented as a fanout.
The default auditing implementation consists of a fanout plug-in and a default auditing plug-in. A custom implementation consists of the fanout plug-in, the default auditing plug-in, and one or more custom auditing plug-ins.
In a fanout plug-in model, a caller sends a request to the fanout plug-in. The fanout plug-in passes the request to each of the subordinate plug-ins, and receives a response from each. Finally, the fanout plug-in forms a composite response from the individual responses, and sends the composite response to the caller.
The purpose of an auditing request is to record an event. Each auditing plug-in returns one of two responses: success (the audit succeeded—logged the event) or failure (the audit failed—did not log the event). The auditing fanout plug-in forms a composite response in the following manner: if all responses are success, the composite response is success; otherwise, the composite response is failure.
For default auditing, the composite response is determined solely by the default auditing plug-in. For custom auditing, the composite response is determined by the fanout plug-in after collecting the responses of the subordinate plug-ins. For more insight into how fanouts work, see Authorization Plug-in Architecture.
Auditing decisions are based partly on user identity, which is stored in an auditing token. Because auditing tokens are generated by the authentication security plug-in, providers of authentication and auditing plug-ins need to ensure that these plug-ins work together.
An ATMI system process or server (such as /Q server TMQUEUE(5) or EventBroker server TMUSREVT(5)) calls the auditing plug-in when it receives a client request. Because it is called before an operation begins, the auditing plug-in can audit operation attempts and store data if that data will be needed later for a post-operation audit. In response, the auditing plug-in performs a pre-operation audit and returns whether the audit succeeded.
The ATMI system process or server may call the auditing plug-in after the client operation is performed. In response, the auditing plug-in performs a post-operation audit and returns whether the audit succeeded.
In addition, an ATMI system process or server may call the auditing plug-in when a potential security violation occurs. (Suspicion of a security violation arises when a pre-operation or post-operation authorization check fails, or when an attack on security is detected.) In response, the auditing performs a post-operation audit and returns whether the audit succeeded.
These calls are system-level calls, not application-level calls. An ATMI application cannot call the auditing plug-in.
The auditing process is somewhat different for (1) users of the default auditing plug-in provided by the Oracle Tuxedo system and (2) users of one or more custom auditing plug-ins. The default plug-in does not support pre-operation audits. If the default auditing plug-in receives a pre-operation audit request, it returns immediately and does nothing.
The custom plug-ins support both pre-operation and post-operation audits.
The default auditing implementation consists of the Oracle Tuxedo EventBroker component and userlog (ULOG). These utilities report only security violations; they do not report which operations were attempted, which ones failed, and which ones successfully completed.
When default auditing is called by an ATMI process to perform a post-operation audit when a security violation is suspected, the auditing plug-in performs the following tasks.
Because the auditing token is created by the authentication plug-in, the auditing plug-in has no record of the token’s content. This information is necessary for the auditing process.
The auditing plug-in examines the client’s auditing token and the security violation delivered in the post-operation audit request.
The auditing fanout plug-in receives a decision (success or failure) from the default auditing plug-in and operates on its behalf.
Users of one or more custom auditing plug-ins may take advantage of additional functionality offered by the ATMI environment of the Oracle Tuxedo product. Specifically, the custom plug-ins may perform an additional audit before an operation occurs.
When custom auditing is called by an ATMI process to perform a pre-operation audit in response to a client request, the auditing plug-in performs the following tasks.
The auditing plug-in examines the client’s auditing token and may store user data if that data will be needed later for a post-operation audit.
The auditing fanout plug-in makes the ultimate decision by checking the individual responses (success or failure) from its subordinate plug-ins.
Custom auditing may be called by the ATMI process to perform a post-operation audit after the client operation is performed. If so, the auditing plug-in performs the following tasks.
The auditing plug-in examines the client’s auditing token, the completion status delivered in the post-operation audit request, and any data stored during the pre-operation audit.
The auditing fanout plug-in decides if the post-operation audit succeeded or failed by checking the individual responses (success or failure) from its subordinate plug-ins.
An operation is considered successful if it passes both pre- and post-operation audits, and the operation itself is successful. Some companies collect and store both pre- and post-operation auditing data, even though such data can occupy a lot of disk space.
You can provide auditing for your ATMI application by using the default plug-in or adding one or more custom plug-ins. You choose a plug-in by configuring the Oracle Tuxedo registry, a tool that controls all security plug-ins.
If you want to use the default auditing plug-in, you do not need to configure the registry. If you want to add one or more custom auditing plug-ins, however, you must configure the registry for your additional plug-ins before you can install them. For more detail about the registry, see Setting the Oracle Tuxedo Registry.
Link-level encryption (LLE) establishes data privacy for messages moving over the network links that connect the machines in an ATMI application. It employs the symmetric key encryption technique (specifically, RC4), which uses the same key for encryption and decryption.
When LLE is being used, the Oracle Tuxedo system encrypts data before sending it over a network link and decrypts it as it comes off the link. The system repeats this encryption/decryption process at every link through which the data passes. For this reason, LLE is referred to as a point-to-point facility.
LLE can be used on the following types of ATMI application links:
There are three levels of LLE security: 0-bit (no encryption), 56-bit (International), and 128-bit (United States and Canada). The International LLE version allows 0-bit and 56-bit encryption. The United States and Canada LLE version allows 0, 56, and 128-bit encryption.
LLE control parameters and underlying communication protocols are different for various link types, but the setup is basically the same in all cases:
The first configuration parameter is the minimum encryption level that a process will accept. It is expressed as a key length: 0, 56, or 128 bits.
The second configuration parameter is the maximum encryption level a process can support. It also is expressed as a key length: 0, 56, or 128 bits.
For convenience, the two parameters are denoted as (min, max) in the discussion that follows. For example, the values “(56, 128)” for a process mean that the process accepts at least 56-bit encryption but can support up to 128-bit encryption.
When two processes at the opposite ends of a network link need to communicate, they must first agree on the size of the key to be used for encryption. This agreement is resolved through a two-step process of negotiation.
A Tuxedo process will process the MINENCRYTPBITS and MAXENCRYPTBITS using the following steps.
min-max values accommodate the default min-max values, then the local software assigns those values as the min-max values for the process.min-max values is not configured, then the default value will be used for the missing value. For instance (0, max-value-configured) or (min-value-configured, 128) will be used.min-max values specified in the configurations for a particular link type, then the local software assigns 0 as the minimum value and assigns the highest bit-encryption rate possible for the default min-max values as the maximum value, that is, (0, 128) for the LLE.
After the min-max values are determined for the two processes, the negotiation of key size begins. The negotiation process need not be encrypted or hidden. Once a key size is agreed upon, it remains in effect for the lifetime of the network connection.
The following table shows which key size, if any, is agreed upon by two processes when all possible combinations of min-max values are negotiated. The header row holds the min-max values for one process; the far left column holds the min-max values for the other.
The ATMI environment of the Oracle Tuxedo product offers some backward compatibility for LLE.
The following table shows which key size, if any, is agreed upon by two ATMI applications when one of them is running under release 6.5 and the other under release 7.1 or later. The header row holds the min-max values for the process running under release 7.1 or later; the far left column holds the min-max values for the process running under release 6.5.
If your current Oracle Tuxedo installation is configured for (0, 56), (0, 128), (56, 56), or (56, 128), and you want to interoperate with a release 6.5 ATMI application that is configured for a maximum LLE level of 40 bits, then any negotiation results in an automatic upgrade to 56.
The negotiation result in this case is the same as the negotiation result for two sites running release 6.5 and configured for a maximum LLE level of 40 bits. In both scenarios, the negotiation results in an automatic upgrade to 56.
The following table shows which key size, if any, is agreed upon by two ATMI applications when one of them is running under pre-release 6.5 and the other under release 7.1 or later. The header row holds the min-max values for the process running under release 7.1 or later; the far left column holds the min-max values for the process running under pre-release 6.5.
If your current Oracle Tuxedo installation is configured for (0, 56) or (0, 128), and you want to interoperate with a pre-release 6.5 ATMI applications that is configured for a maximum LLE level of 40 bits, then the result of any negotiation is 40.
If your current Oracle Tuxedo installation is configured for (56, 56), (56, 128), or (128, 128), then your system cannot interoperate with a pre-release 6.5 ATMI application that is configured for a maximum LLE level of 40 bits. Attempts to negotiate a common key size fail.
The length of time a Workstation client can take for initialization is limited. By default, this interval is 30 seconds in an ATMI application not using LLE, and 60 seconds in an ATMI application using LLE. The 60-second interval includes the time needed to negotiate an encrypted link. This time limit can be changed when LLE is configured by changing the value of the MAXINITTIME parameter for the workstation listener (WSL) server in the UBBCONFIG file, or the value of the TA_MAXINITTIME attribute in the T_WSL class of the WS_MIB(5).
The Oracle Tuxedo product provides the industry-standard SSL protocol to establish secure communications between client and server applications. When using the SSL protocol, principals use digital certificates to prove their identity to a peer.
| Note: | The actual network protocol used is TLS 1.0, which is the successor to the SSL protocol, but this document will follow common usage and refer to this protocol as SSL Encryption. |
Like LLE, the SSL protocol can be used with password authentication to provide confidentiality and integrity to communication between the client application and the Oracle Tuxedo domain. When using the SSL protocol with password authentication, you are prompted for the password of the Listener/Handler (IIOP, Workstation, or JOLT) defined by the SEC_PRINCIPAL_NAME parameter when you enter the tmloadcf command.
SSL can be used on the following types of ATMI application links:
There are four levels of SSL security: 0-bit (no encryption), 56-bit, 128-bit, and 256-bit. Choose the level that you need during Oracle Tuxedo installation.
The SSL protocol works in the following manner:
The initiating application can then use either password or certificate authentication to authenticate itself to the Oracle Tuxedo domain.
Figure 1-7 illustrates how the SSL protocol works.
The implementation of the SSL protocol is flexible enough to fit into most public key infrastructures. The default implementation of the plug-in frame work in the Oracle Tuxedo product requires that digital certificates are stored in an LDAP-enabled directory. You can choose any LDAP-enabled directory service. You also need to choose the certificate authority from which to obtain digital certificates and private keys used in a Tuxedo application. You must have an LDAP-enabled directory service and a certificate authority in place before using the SSL protocol in a Tuxedo application.
When two processes at the opposite ends of a network link need to communicate, they must first agree on the size of the key to be used for encryption. This agreement is resolved through a two-step process of negotiation.
A Tuxedo process will process the MINENCRYTPBITS and MAXENCRYPTBITS using the following steps.
min-max values accommodate the default min-max values, then the local software assigns those values as the min-max values for the process.min-max values is not configured, then the default value will be used for the missing value. For instance (0, max-value-configured) or (min-value-configured, 128) will be used.min-max values specified in the configurations for a particular link type, then the local software assigns 0 as the minimum value and assigns 128 as the maximum value.| Notes: |
MAXENCRYPTBITS is 256; it will be adjusted according to the actual link level security configured.
After the min-max values are determined for the two processes, the negotiation of key size begins. The negotiation process need not be encrypted or hidden. Once a key size is agreed upon, it remains in effect for the lifetime of the network connection.
The following table shows which key size, if any, is agreed upon by two processes when all possible combinations of min-max values are negotiated. The header row holds the min-max values for one process; the far left column holds the min-max values for the other.
In order to use SSL between two Tuxedo processes, both processes must be running Tuxedo 10.0 or later (except when using the CORBA SSL capabilities described in "Using Security in CORBA Applications." It is possible to specify both non-SSL and SSL ports for WSL and JSL processes and to specify SSL or LLE connectivity for individual entries in the *DM_TDOMAIN section of a DMCONFIG file. In this way, it is possible to gradually migrate a workstation or domain application to use SSL as individual workstation clients and Tuxedo domains are upgraded to Tuxedo 10.
| Note: | It is not possible to use SSL between BRIDGE and tlisten processes in an MP mode application until all machines in the Tuxedo domain are upgraded to Tuxedo 10.0 or later. |
The length of time a Workstation client can take for initialization is limited. By default, this interval is 60. The 60-second interval includes the time needed to negotiate an encrypted link. This time limit can be changed when WSL is configured by changing the value of the MAXINITTIME parameter for the workstation listener (WSL) server in the UBBCONFIG file, or the value of the TA_MAXINITTIME attribute in the T_WSL class of the WS_MIB(5).
A cipher suite is a SSL encryption method that includes the key exchange algorithm, the symmetric encryption algorithm, and the secure hash algorithm used to protect the integrity of the communication. For example, the cipher suite RSA_WITH_RC4_128_MD5 uses RSA for key exchange, RC4 with a 128-bit key for bulk encryption, and MD5 for message digest.
The ATMI security environment supports the cipher suites described in Table 1-8.
The appropriate encryption software for SSL is automatically included in Oracle Tuxedo installer. You can choose whether or not to use SSL encryption in your application during Oracle Tuxedo installation.
For more information, see Installing the Oracle Tuxedo System.
Public key security provides two capabilities that make end-to-end digital signing and data encryption possible:
Message-based digital signature allows the recipient (or recipients) of a message to identify and authenticate both the sender and the sent message. Digital signature provides solid proof of the originator and content of a message; a sender cannot falsely repudiate responsibility for a message to which that sender’s digital signature is attached. Thus, for example, Bob cannot issue a request for a withdrawal from his bank account and later claim that someone else issued that request.
In addition, message-based encryption protects the confidentiality of messages by ensuring that only designated recipients can decrypt and read them.
Informal but recognized industry standards for public key software have been issued by a group of leading communications companies, led by RSA Laboratories. These standards are called Public-Key Cryptography Standards, or PKCS. The public key software in the ATMI environment of the Oracle Tuxedo software complies with the PKCS-7 standard.
PKCS-7 is a hybrid cryptosystem architecture. A symmetric key algorithm with a random session key is used to encrypt a message, and a public key algorithm is used to encrypt the random session key. A random number generator creates a new session key for each communication, which makes it difficult for a would-be attacker to reuse previous communications.
All the algorithms on which public key security is based are well known and commercially available. To select the algorithms that will best serve your ATMI application, consider speed and degree of security factor.
The public key security in the ATMI environment of the Oracle Tuxedo product supports any public key algorithms supported by the underlying plug-ins, including RSA, ElGamal, and Rabin. (RSA stands for Rivest, Shamir, and Adelman, the inventors of the RSA algorithm.) All these algorithms can be used for digital signatures and encryption.
Public key (or asymmetric key) algorithms such as RSA are implemented through a pair of different but mathematically related keys:
The public key security in the ATMI environment of the Oracle Tuxedo product supports any digital signature algorithms supported by the underlying plug-ins, including RSA, ElGamal, Rabin, and Digital Signature Algorithm (DSA). With the exception of DSA, all these algorithms can be used for digital signatures and encryption. DSA can be used for digital signatures but not for encryption.
Digital signature algorithms are simply public key algorithms used to provide digital signatures. DSA is also a public key algorithm (implemented through public-private key pairs), but it can only be used to provide digital signatures, not encryption.
Public key security supports the following three symmetric key algorithms:
DES-CBC is a 64-bit block cipher run in Cipher Block Chaining (CBC) mode. It provides 56-bit keys (8 parity bits are stripped from the full 64-bit key) and is exportable outside the United States.
Two-key triple-DES is a 128-bit block cipher run in Encrypt-Decrypt-Encrypt (EDE) mode. Two-key triple-DES provides two 56-bit keys (in effect, a 112-bit key) and is not exportable outside the United States.
For some time it has been common practice to protect and transport a key for DES encryption with triple-DES, which means that the input data (in this case the single-DES key) is encrypted, decrypted, and then encrypted again (an encrypt-decrypt-encrypt process). The same key is used for the two encryption operations.
RC2 is a variable key-size block cipher with a key size range of 40 to 128 bits. It is faster than DES and is exportable with a key size of 40 bits. A 56-bit key size is allowed for foreign subsidiaries and overseas offices of United States companies. In the United States, RC2 can be used with keys of virtually unlimited length, although the ATMI public key security restricts the key length to 128 bits.
Oracle Tuxedo customers cannot expand or modify this list of algorithms.
In symmetric key algorithms, the same key is used to encrypt and decrypt a message. The public key encryption system uses symmetric key encryption to encrypt a message sent between two communicating entities. Symmetric key encryption operates at least 1000 times faster than public key cryptography.
A block cipher is a type of symmetric key algorithm that transforms a fixed-length block of plaintext (unencrypted text) data into a block of ciphertext (encrypted text) data of the same length. This transformation takes place in accordance with the value of a randomly generated session key. The fixed length is called the block size.
Public key security supports any message digest algorithms supported by the underlying plug-ins, including MD5, SHA-1 (Secure Hash Algorithm 1), and many others. Both MD5 and SHA-1 are well known, one-way hash algorithms. A one-way hash algorithm takes a message and converts it into a fixed string of digits, which is referred to as a message digest or hash value.
MD5 is a high-speed, 128-bit hash; it is intended for use with 32-bit machines. SHA-1 offers more security by using a 160-bit hash, but is slower than MD5.
Message-based digital signatures enhance ATMI security by allowing a message originator to prove its identity, and by binding that proof to a specific message buffer. Mutually authenticated and tamper-proof communication is considered essential for ATMI applications that transport data over the Internet, either between companies or between a company and the general public. It also is critical for ATMI applications deployed over insecure internal networks.
The scope of protection for a message-based digital signature is end-to-end: a message buffer is protected from the time it leaves the originating process until the time it is received at the destination process. It is protected at all intermediate transit points, including temporary message queues, disk-based queues, and system processes, and during transmission over inter-server network links.
The following figure shows how end-to-end message-based digital signature works.
Message-based digital signature involves generating a digital signature by computing a message digest on the message, and then encrypting the message digest with the sender’s private key. The recipient verifies the signature by decrypting the encrypted message digest with the signer’s public key, and then comparing the recovered message digest to an independently computed message digest. The signer’s public key either is contained in a digital certificate included in the signer information, or is referenced by an issuer-distinguished name and issuer-specific serial number that uniquely identify the certificate for the public key.
Digital certificates are electronic files used to uniquely identify individuals and resources over networks such as the Internet. A digital certificate securely binds the identity of an individual or resource, as verified by a trusted third party known as a Certification Authority, to a particular public key. Because no two public keys are ever identical, a public key can be used to identify its owner.
Digital certificates allow verification of the claim that a specific public key does in fact belong to a specific subscriber. A recipient of a certificate can use the public key listed in the certificate to verify that the digital signature was created with the corresponding private key. If such verification is successful, this chain of reasoning provides assurance that the corresponding private key is held by the subscriber named in the certificate, and that the digital signature was created by that particular subscriber.
A certificate typically includes a variety of information, such as:
The most widely accepted format for certificates is defined by the ITU-T X.509 international standard. Thus, certificates can be read or written by any ATMI application complying with X.509. The public key security in the ATMI environment of the Oracle Tuxedo product recognizes certificates that comply with X.509 version 3, or X.509v3.
Certificates are issued by a Certification Authority, or CA. Any trusted third-party organization or company that is willing to vouch for the identities of those to whom it issues certificates and public keys can be a CA. When it creates a certificate, the CA signs the certificate with its private key, to obtain a digital signature. The CA then returns the certificate with the signature to the subscriber; these two parts—the certificate and the CA’s signature—together form a valid certificate.
The subscriber and others can verify the issuing CA’s digital signature by using the CA’s public key. The CA makes its public key readily available by publicizing that key or by providing a certificate from a higher-level CA attesting to the validity of the lower-level CA’s public key. The second solution gives rise to hierarchies of CAs.
The recipient of an encrypted message can develop trust in the CA’s private key recursively, if the recipient has a certificate containing the CA’s public key signed by a superior CA whom the recipient already trusts. In this sense, a certificate is a stepping stone in digital trust. Ultimately, it is necessary to trust only the public keys of a small number of top-level CAs. Through a chain of certificates, trust in a large number of users’ signatures can be established.
Thus, digital signatures establish the identities of communicating entities, but a signature can be trusted only to the extent that the public key for verifying the signature can be trusted.
Note that Oracle Systems has no plans to become a CA. By offering a public key plug-in interface, Oracle Systems extends the opportunity to all Oracle Tuxedo customers to select a CA of their choice.
To make a public key and its identification with a specific subscriber readily available for use in verification, the digital certificate may be published in a repository or made available by other means. Repositories are databases of certificates and other information available for retrieval and use in verifying digital signatures. Retrieval can be accomplished automatically by having the verification program directly request certificates from the repository as needed.
The Public-Key Infrastructure (PKI) consists of protocols, services, and standards supporting applications of public key cryptography. Because the technology is still relatively new, the term PKI is somewhat loosely defined: sometimes “PKI” simply refers to a trust hierarchy based on public key certificates; in other contexts, it embraces digital signature and encryption services provided to end-user applications as well.
There is no single standard public key infrastructure today, though efforts are underway to define one. It is not yet clear whether a standard will be established or multiple independent PKIs will evolve with varying degrees of interoperability. In this sense, the state of PKI technology today can be viewed as similar to local and wide-area network technology in the 1980s, before there was widespread connectivity via the Internet.
The following services are likely to be found in a PKI:
The following figure shows the PKI process flow.
Note that Oracle Systems has no plans to become a PKI vendor. By offering a public key plug-in interface, Oracle Systems extends the opportunity to all Oracle Tuxedo customers to use a PKI security solution with the PKI software from their vendor of choice.
Message-based encryption keeps data private, which is essential for ATMI applications that transport data over the Internet, whether between companies or between a company and its customers. Data privacy is also critical for ATMI applications deployed over insecure internal networks.
Message-based encryption also helps ensure message integrity, because it is more difficult for an attacker to modify a message when the content is obscured.
The scope of protection provided by message-based encryption is end-to-end; a message buffer is protected from the time it leaves the originating process until the time it is received at the destination process. It is protected at all intermediate transit points, including temporary message queues, disk-based queues, and system processes, and during transmission over interserver network links.
The following figure shows how end-to-end message-based encryption works.
The message is encrypted by a symmetric key algorithm and a session key. Then, the session key is encrypted by the recipient’s public key. Next, the recipient decrypts the encrypted session key with the recipient’s private key. Finally, the recipient decrypts the encrypted message with the session key to obtain the message content.
| Note: | The figure does not show two other steps in this process: (1) the data is compressed immediately before the message is encrypted; and (2) the data is uncompressed immediately after the message is decrypted. |
Because the unit of encryption is an ATMI message buffer, message-based encryption is compatible with all existing ATMI programming interfaces and communication paradigms. The encryption process is always the same, whether it is being performed on messages shipped between two processes in a single machine, or on messages sent between two machines through a network.
The underlying plug-in interface for public key security consists of six component interfaces, each of which requires one or more plug-ins. By instantiating these interfaces with your preferred plug-ins, you can bring custom message-based digital signature and message-based encryption to your ATMI application.
The six component interfaces are:
The public key initialization interface allows public key software to open public and private keys. For example, gateway processes may need to have access to a specific private key in order to decrypt messages before routing them. This interface is implemented as a fanout.
The key management interface allows public key software to manage and use public and private keys. Note that message digests and session keys are encrypted and decrypted using this interface, but no bulk data encryption is performed using public key cryptography. Bulk data encryption is performed using symmetric key cryptography.
The certificate lookup interface allows public key software to retrieve X.509v3 certificates for a given principal. Principals are authenticated users. The certificate database may be stored using any appropriate tool, such as Lightweight Directory Access Protocol (LDAP), Microsoft Active Directory, Netware Directory Service (NDS), or local files.
The certificate parsing interface allows public key software to associate a simple principal name with an X.509v3 certificate. The parser analyzes a certificate to generate a principal name to be associated with the certificate.
The certificate validation interface allows public key software to validate an X.509v3 certificate in accordance with specific business logic. This interface is implemented as a fanout, which allows Oracle Tuxedo customers to use their own business rules to determine the validity of a certificate.
The proof material mapping interface allows public key software to access the proof materials needed to open keys, provide authorization tokens, and provide auditing tokens.
You can provide public key security for your ATMI application by using custom plug-ins. You choose a plug-in by configuring the Oracle Tuxedo registry, a tool that controls all security plug-ins.
If you want to use custom public key plug-ins, you must configure the registry for your public key plug-ins before you can install them. For more detail about the registry, see Setting the Oracle Tuxedo Registry.
The default public key implementation supports the following algorithms:
The default authentication and authorization plug-ins provided by the ATMI environment of the Oracle Tuxedo product work in the same manner that implementations of authentication and authorization have worked since they were first made available with the Oracle Tuxedo system.
An application administrator can use the default authentication and authorization plug-ins to configure an ATMI application with one of five levels of security. The five levels are:
At the lowest level, no authentication is provided. At the highest level, an access control checking feature determines which users can execute a service, post an event, or enqueue (or dequeue) a message on an application queue. The security levels are briefly described in the following table.
| Note: | The term client is synonymous with client process, meaning a specific instance of a client program in execution. An ATMI client program can exist in active memory in any number of individual instances. |
An application administrator can designate a security level by setting the SECURITY parameter in the UBBCONFIG configuration file to the appropriate value.
The default is NONE. If SECURITY is set to USER_AUTH, ACL, or MANDATORY_ACL, then the application administrator must configure a system-supplied authentication server named AUTHSVR. AUTHSVR performs per-user authentication.
An application developer can replace AUTHSVR with an authentication server that has logic specific to the ATMI application. For example, a company may want to develop a custom authentication server so that it can use the popular Kerberos mechanism for authentication.
Upon joining an ATMI application, a client process has two names: a combined user-client name and a unique client identifier known as an application key.
Two client names are reserved for special semantics: tpsysadm and tpsysop. tpsysadm is treated as the application administrator, and tpsysop is treated as the application operator.
When an authenticated client joins an ATMI application, it passes a username and client name to tpinit(3c) in a TPINIT buffer if the application is written in C, or to TPINITIALIZE(3cbl) in a TPINFDEF-REC record if the application is written in COBOL. The username and client name, as well as other security-related fields in the TPINIT buffer/ TPINFDEF-REC record, are described in the following table.
For an authenticated security level (USER_AUTH, ACL, or MANDATORY_ACL), the username, client name, and user-specific data are transferred to AUTHSVR without interpretation by the Oracle Tuxedo system. The only manipulation of this information is its encryption when transmitted over the network from a Workstation client.
Every time a client joins an ATMI application, it is assigned a 32-bit application key by the Oracle Tuxedo system. The client cannot reset the key other than by terminating its association and joining the ATMI application as a different user.
The assigned application key is the client’s security credential. The client provides its application key with every service invocation as part of the TPSVCINFO structure in the appkey field. (See tpservice(3c) in the Oracle Tuxedo ATMI C Function Reference for more information about TPSVCINFO.)
The following table shows how the application key is set for various security levels and clients. All application key assignments are hardcoded except the last item in the table.
Messages from native ATMI clients that must be run by the administrator (like tmadmin(1))
|
||
In addition, any message that originates from tpsvrinit(3c) or tpsvrdone(3c) in a C program (TPSVRINIT(3cbl) or TPSVRDONE(3cbl) in COBOL) is assigned the application key of the administrator: 0x80000000. The application key of the client is assigned to messages that pass through a server but originate at a client; an exception to this rule is described in Replacing Client Tokens with Server Tokens.
A user identifier (UID) is an integer, between 0 and 128K, that is used by the application to refer to a particular user. A group identifier (GID) is an integer, between 0 and 16K, that is used by the application to refer to an application group.
To use access control, an application administrator must maintain lists of (1) users, (2) groups, and (3) and mappings of groups to application entities (such as services, events, and application queues). The third type of list, a mapping of groups to application entities, is known as the access control list (ACL).
When a client tries to access an application resource, such as a service, the system checks the client’s application key and thus identifies the group to which the user belongs. Next, the system checks the ACL for the target resource and determines whether the client’s group has access permission. The application administrator, application operator, and processes or service requests running with the privileges of the application administrator or operator are not subject to ACL permission checking.
The user, group, and ACL files are located in the application_root directory, where application _root is the first pathname defined for the APPDIR variable. The following figure identifies these files and specifies the administrative commands available for controlling each list.
| Note: | For an ATMI application running on the Compaq VMS operating system, the names of the user, group, and ACL files have .dat extensions: tpusr.dat, tpgrp.dat, and tpacl.dat. |
The files are colon-delimited, flat text files that can be read and written only by the application administrator—the owner of the TUXCONFIG file referenced by the TUXCONFIG variable. The format of the files is irrelevant, since the files are fully administered with a set of dedicated commands. Only the application administer is allowed to use these commands.
An application administer can use the tpaclcvt(1) command to convert security data files to the format needed by the ACL checking feature. For example, on a UNIX host machine, an administrator can use tpaclcvt to convert the /etc/password file and store the converted version in the tpusr file. The same administrator can use tpaclcvt to convert the /etc/group file and store the converted version in the tpgrp file.
The AUTHSVR server uses the user information stored in the tpusr file to authenticate users who want to join the ATMI application.
The ACL and MANDATORY_ACL security levels constitute the default authorization implementation for the ATMI environment in the Oracle Tuxedo product.
When the security level is ACL, if there is no entry in the tpacl file associated with the target application entity, the client is permitted to access the entity. This security level enables an administrator to configure access for only those resources that need more security. That is, there is no need to add entries to the tpacl file for services, events, or application queues that are open to everyone.
When the security level is MANDATORY_ACL, if there is no entry in the tpacl file associated with the target application entity, the client is not permitted to access the entity. For this reason, this level is called mandatory. There must be an entry in the tpacl file for each and every application entity that the client needs to access.
For both the ACL and MANDATORY_ACL security levels, if an entry for an application entity exists in the tpacl file and the client attempts to access that entity, the user associated with that client must be a member of a group that is allowed to access that entity; otherwise, permission is denied.
For some ATMI applications, it may be necessary to use both system-level and application-level authorization. An entry in the tpacl file can be used to control which users can access a service, and application logic can control data-dependent access, for example, which users can handle transactions for more than a million dollars.
Note that there is no ACL permission checking for administrative services, events, and application queues with names that begin with a dot (.). For example, any client can subscribe to administrative events such as .SysMachineBroadcast, .SysNetworkConfig, and .SysServerCleaning. In addition, there is no ACL permission checking for the application administrator, application operator, or processes or service requests running with the privileges of the application administrator or operator.
Application developers and administrators must be aware of certain security issues when configuring ATMI applications to interoperate with Oracle Tuxedo pre-release 7.1 (6.5 or earlier) software.
Interoperability, as defined in this discussion, is the ability of the current release of Oracle Tuxedo software to communicate over a network with a previous release of Oracle Tuxedo software. Specifically, inter-domain interoperability and intra-domain interoperability have the following meanings:
Involves one ATMI application running Oracle Tuxedo release 7.1 or later software, and another ATMI application running Oracle Tuxedo pre-release 7.1 software. See the diagram Inter-Domain Interoperability for clarification.
Involves one machine in a multiple-machine ATMI application running Oracle Tuxedo release 7.1 or later software, and another machine in the same application running Oracle Tuxedo pre-release 7.1 software. See the figure Intra-Domain Interoperability for clarification.
Interoperating with Oracle Tuxedo pre-release 7.1 software is allowed or disallowed at the authentication security level. Authentication, as implemented by Oracle Tuxedo release 7.1 or later software, allows communicating processes to mutually prove their identities.
By default, interoperability with a machine running Oracle Tuxedo pre-release 7.1 software is not allowed. To change the default, an application administrator can use the CLOPT -t option to allow workstation handlers (WSHs), domain gateways (GWTDOMAINs), and servers in the release 7.1 or later ATMI application to interoperate with Oracle Tuxedo pre-release 7.1 software. Mandating Interoperability Policy provides instructions for using the CLOPT -t option as well as the security ramifications for authentication and authorization when using CLOPT -t.
Whenever a network link is established between machines running Oracle Tuxedo software, link-level encryption may be used to encrypt data before sending it over the network link, and decrypt it as it comes off the link. Of course, link-level encryption is possible only if LLE is installed on both the sending and receiving machines.
LLE interoperability with Oracle Tuxedo pre-release 7.1 software is described in Backward Compatibility of LLE.
SSL encryption can be used over network links between machines running Oracle Tuxedo software only if both machines are running Tuxedo 10.0 or later. LLE encryption can be used over network links to machines running earlier releases of Tuxedo.
The following interoperability rules for public key security apply to a machine running release 7.1 or later Oracle Tuxedo software that is configured to interoperate with a machine running Oracle Tuxedo pre-release 7.1 software. To clarify the rules, each rule has an accompanying example scenario involving a Workstation client running Oracle Tuxedo pre-release 7.1 software.
See Setting Digital Signature Policy for a description of the
SIGNATURE_REQUIRED configuration parameter.
|
For inter-domain interoperability, release 7.1 or later domain gateway (GWTDOMAIN) processes enforce the interoperability rules for public key security.
For intra-domain interoperability, release 7.1 or later native clients, workstation handlers (WSHs), or server processes communicating with the local bridge process enforce the interoperability rules for public key security, as shown in the following figure. A bridge process operates only as a conduit; it does not decrypt message buffer content or verify digital signatures.
| Note: | Typically, a release 7.1 or later WSH does not verify digital signatures. But when routing a digitally signed message buffer to a process running Oracle Tuxedo pre-release 7.1 software, the WSH verifies any digital signatures before removing them. |
For an ATMI application running Oracle Tuxedo release 7.1 or later software, it is possible to have any combination of default or custom authentication, authorization, auditing, and public key security. In addition, any combination of these four security capabilities is compatible with link-level encryption.
It is possible to have default authentication and custom authorization, or custom authentication and default authorization, as long as the application developer is aware of the following restriction: the authorization security token must carry at a minimum (1) an authenticated username, or principal name, and (2) an application key value as defined in Application Key.
Authorization decisions are based partly on user identity, which is stored in an authorization token. Because authorization tokens are generated by the authentication security plug-in, providers of authentication and authorization plug-ins need to ensure that these plug-ins work together. (See Authentication and Authorization for more detail.)
It is possible to have default authentication and custom auditing, or custom authentication and default auditing, as long as the application developer is aware of the following restriction: the auditing security token must carry at a minimum (1) an authenticated username, or principal name, and (2) an application key value as defined in Application Key.
Auditing decisions are based partly on user identity, which is stored in an auditing token. Because auditing tokens are generated by the authentication security plug-in, providers of authentication and auditing plug-ins need to ensure that these plug-ins work together. (See Authentication and Auditing for more detail.)
Public key security is compatible with all features and processes supported by Oracle Tuxedo release 7.1 or later software except the compression feature. Encrypted message buffers cannot be compressed using the compression feature. But, because the public key software compresses the message content just before it encrypts the message buffer, any size savings are still achieved.
This topic describes the compatibility/interaction of public key security with the following ATMI features and processes:
Central to the data-dependent routing feature is the ability of a process to examine the content of incoming message buffers. If an incoming message buffer is encrypted, a process configured for data-dependent routing must have opened a recipient’s private key so that the public key software can use that key to decrypt the message buffer. For data-dependent routing, the public key software does not verify digital signatures.
If a decryption key is not available, the routing operation fails. The system generates an ERROR userlog(3c) message to report the failure.
If a decryption key is available, the process makes a routing decision based on a decrypted copy of the encrypted message buffer. The chain of events is as follows:
The system then transmits the original encrypted message buffer in accordance with the routing decision.
Public-private keys are represented and manipulated via handles. A handle has data associated with it that is used by the public key application programming interface (API) to locate or access the item named by the handle. A process opens a key handle for digital signature generation, message encryption, or message decryption.
A key handle is a process resource; it is not bound to any specific thread or context. Any communication necessary to open a key is performed within the thread’s currently active context. Thereafter, the key is available to any context in the process, whether or not the context is associated with the same ATMI application.
A key’s internal data structures are thread safe. As such, a key may be accessed concurrently by multiple threads.
In general, a TMUSREVT(5) system server handles encrypted message buffers without decrypting them, that is, both digital signatures and encryption envelopes remain intact as messages flow through the Oracle Tuxedo EventBroker component. However, the following cases require that the EventBroker component decrypt posted message buffers:
If the EventBroker does not have access to a suitable decryption key, the subscription’s filter expression is assumed to be false, and the subscription is not considered a match.
If the EventBroker does not have access to a suitable decryption key, the subscription’s notification action fails, and the system generates an ERROR userlog(3c) message to report the failure.
If the EventBroker does not have access to a suitable decryption key, the subscription’s notification action fails, and the system generates an ERROR userlog() message to report the failure.
For a transactional subscription, the system also marks the transaction as rollback-only.
If the EventBroker does not have access to a suitable decryption key, the tppost(3c) operation fails, and the system generates an ERROR userlog() message to report the failure.
If the EventBroker does not have access to a suitable decryption key, the tppost(3c) operation fails, and the system generates an ERROR userlog() message to report the failure.
In general, a TMQUEUE(5) or TMQFORWARD(5) system server handles encrypted message buffers without decrypting them, that is, both signatures and encryption envelopes remain intact as messages flow through the Oracle Tuxedo /Q component. However, the following cases require that the /Q component decrypt enqueued message buffers:
TMQFORWARD operations that, based on system configurations, need to access message content for data-dependent routing.
If TMQFORWARD does not have access to a suitable decryption key, the forward operation fails. The system returns the message to the queue and generates an ERROR userlog(3c) message to report the failure.
After a number of periodic retry attempts, TMQFORWARD might place the unreadable message on an error queue.
If the /Q component does not have access to a suitable decryption key, the tpenqueue(3c) operation fails, and the system generates an ERROR userlog() message to report the failure.
If the /Q component does not have access to a suitable decryption key, the tpenqueue(3c) operation fails, and the system generates an ERROR userlog() message to report the failure.
A non-transactional tpdequeue(3c) operation has the side effect of destroying an encrypted queued message if the invoking process does not hold a valid decryption key.
If a message with an invalid signature is placed in a queue (or if the message is corrupted or tampered with while on the queue), any attempt to dequeue it fails. A non-transactional tpdequeue() operation has the side effect of destroying such a message. A transactional tpdequeue() operation causes transaction rollback, and all future transactional attempts to dequeue the message will continue to fail.
Public key security operations—opening and closing keys, requesting a digital signature, or requesting encryption—are not transactional, and are not undone by transaction rollback. However, transactions might rollback due to failure conditions associated with the following public key operations:
Domain gateway (GWTDOMAIN) processes connecting two ATMI applications running Oracle Tuxedo release 7.1 or later software preserve digital signatures and encryption envelopes. In addition, the domain gateway processes verify digital signatures and enforce administrative system policies regarding digital signatures and encryption.
The following figure is an aid to understanding how domain gateway processes interact with local and remote ATMI applications. The table following the figure describes how release 7.1 or later domain gateway processes handle digitally signed and encrypted message buffers.
|
If the data-dependent routing feature applies and the domain gateway process does not have a suitable decryption key, the gateway process rejects the message. (See Compatibility/Interaction with Data-dependent Routing for clarification.)
|
||
If the domain gateway process is running within a domain, machine, or group requiring encryption, the gateway process rejects the message. If a service advertised by the domain gateway requires encryption, the gateway process rejects the message. (See Setting Encryption Policy for clarification.)
|
||
If the domain gateway process is running within a domain, machine, or group requiring digital signatures, the gateway process rejects the message. If a service advertised by the domain gateway requires digital signatures, the gateway process rejects the message. (See Setting Digital Signature Policy for clarification.)
|
||
|
If the data-dependent routing feature applies and the domain gateway process does not have a suitable decryption key, the gateway process rejects the message. (See Compatibility/Interaction with Data-dependent Routing for clarification.)
If the encrypted message is destined for a process running Oracle Tuxedo pre-release 7.1 (6.5 or earlier) software, the domain gateway process rejects the message. (See Interoperating with Pre-Release 7.1 Software and Interoperability for Public Key Security for clarification.)
|
||
If the domain gateway process is running within a domain, machine, or group requiring encryption, the gateway process rejects the message. If a service advertised by the domain gateway requires encryption, the gateway process rejects the message. (See Setting Encryption Policy for clarification.)
|
||
The domain gateway process verifies the digital signature and forwards the message with digital signature attached over the network.
If the message is destined for a process running Oracle Tuxedo pre-release 7.1 software and assuming interoperability with Oracle Tuxedo pre-release 7.1 software is allowed, the domain gateway process verifies and then removes the digital signature before forwarding the message over the network. (See Interoperating with Pre-Release 7.1 Software and Interoperability for Public Key Security for clarification.)
|
||
If the domain gateway process is running within a domain, machine, or group requiring digital signatures, the gateway process rejects the message. If a service advertised by the domain gateway requires digital signatures, the gateway process rejects the message. (See Setting Digital Signature Policy for clarification.)
|
A domain gateway (GWTDOMAIN) process connecting a release 7.1 or later ATMI application to another vendor’s gateway process operates on outbound message buffers as follows:
In addition, the domain gateway process enforces the administrative system policies regarding encryption and digital signatures for the ATMI application. As an example, if encryption and/or digital signatures are required at the domain level for the ATMI application, the local domain gateway process rejects any message coming from the other vendor’s gateway process.
With more distributed multi-domain Tuxedo applications extending their reach to public networks and less secure environments, the Tuxedo domain gateway is required to better defend against potential threats. These environments may contain insecure networks and untrusted participants, who can initiate or propagate malicious attacks such as Denial-of-Service (DoS) attacks as shown in Figure 1-16.
The Tuxedo TDomain gateway (GWTDOMAIN) uses the following features to defend against DoS attacks.
Limited/Restricted Connection Numbers
Message Authentication Code (MAC) Usage
GWTDOMAIN is a daemon server that waits on a well-known TCP port to accept incoming connection requests. This opens the vulnerability to connection flood attack, a type of DoS attack where the attacker continuously tries to establish many connections with GWTDOMAIN at the same timeusing particular tools (for example, a port scanning program). This causes the domain gateway to waste computing power (time, memory, and so on) to accept the connection requests and allocate resources for each connection.
By limiting the number of connections, GWTDOMAIN can avoid this problem. For more GWTDOMAIN information, see
GWTDOMAIN(5).
The Limited/Restricted Connection Numbers feature requires modification of the *SERVERS section in the UBBCONFIG file.
The CLOPT used to specify the parameter for GWTDOMAIN is "-x" using the following syntax: -x limit[:{[duration][:period]}]. A colon ( : ) is used to separate each option.
| Notes: | The colon ( : ) can only be used between two options. For example, configurations like ":duration" or "limit::" are invalid. |
| Note: | The default value(s) for the duration and period options are used if they are not specified. |
| Note: | Please be aware that the timing is not exact for performance reason. There may be a one-second difference. |
If the number of current active connections plus the number of closed connections in a specified previous period is greater than the limit, GWTDOMAIN is suspended for a duration specified in seconds.
| Note: | The number of current active connections includes both active incoming connections and active outgoing connections. The number of closed connections in a previous period includes both closed incoming connections and closed outgoing connections. However, when GWTDOMAIN is suspended, none of the closed connections are counted. |
limit, duration, and period are defined as follows:
The maximum number of connections. The minimum limit value is 0, and the maximum value is 2,147,483,647.
When the limit is reached (or exceeded) and there is an incoming request, GWTDOMAIN is suspended for the given duration. At the same time, the current incoming request which triggers the suspending is not accepted. Polling is resumed after duration has elapsed.
Setting the limit to 0 prohibits the domain gateway from accepting any incoming connection requests. In other words, this is an "OUTGOING_ONLY" connection policy.
The duration in seconds to suspend polling for incoming connection when limit is reached. The default value is (SCANUNIT * SANITYSCAN) seconds. The minimum duration value is 5, and the maximum value is 65,535.
The time interval (in seconds) proceeding GWTDOMAIN check point to count the closed connections in the past. When not specified, the default value is the same as duration. The minimum period value is 0, and the maximum value is 65,535.
If period is specified as 0, the number of closed connections in a prior period will always be 0, limit only counts active connections.
Listing 1-1 shows an example where the GWTDOMAIN limit is set to 512 concurrent socket connections. When the 512 limit is reached and there is an incoming request, GWTDOMAIN will stop polling and accepting new incoming connection requests for a duration of 300 seconds (or, 5 minutes). Since period is specified as 0, only the active connections are counted.
# UBBCONFIG
...
*SERVERS
GWTDOMAIN SRVGRP=GWGRP1 SRVID=2 CLOPT= “-A -- -x 512:300:0”
Listing 1-2 shows an example where the GWTDOMAIN limit is set to 200 concurrent socket connections. When the 200 limit is reached, (for example:
and since the duration value is not specified, GWTDOMAIN will stop polling and accepting new incoming connection requests for the duration default value SCANUNIT * SANITYSCAN seconds.
| Note: | The current incoming connection that triggered the suspension is also not accepted, and is closed at the end of the suspended duration. |
# UBBCONFIG
...
*SERVERS
GWTDOMAIN SRVGRP=GWGRP1 SRVID=2 CLOPT= “-A -- -x 200::60”
The following conditions will post messages to USERLOG:
<LIBGW_CAT 5359> "WARN: The number of connections for <ldom-name> exceeds limit <%d>, start to suspend for <%d> seconds"
<LIBGW_CAT 5360> "INFO: Resume accepting connection request"
| Note: | These two messages can be controlled using the "throttle message" mechanism to avoid the potential of flooding the USERLOG. |
<LIBGW_CAT 5361> "INFO: The connection limit for <ldom-name> is set to 0. No incoming connection request will be accepted."
The sanity check of message is strengthened with this feature, to protect GWTDOMAIN from crash when under attack. This feature is deployed automatically after installed, no configuration work needed.
By associating the message authentication code (MAC) with messages, a Tuxedo domain gateway can validate and authenticate them. With MAC, the domain gateway can defend against various types of DoS attacks (for example, message tampering, message forging, and message replay attack).
This feature can only take effect when LLE and/or domain SECURITY is configured. MAC works after connection is established. When a MAC message from a remote domain gateway fails validation and authentication, the corresponding connection is dropped. All pending messages are also dropped, and all on-going service requests fail.
GWTDOMAIN determines whether MAC is turned on for the session during the session negotiation phase. MAC can only be enabled when either LLE and/or SECURITY is supported and activated for the session.
| Note: | SSL does not support MAC usage. |
It is not necessary to turn on the SECURITY feature to enable MAC; however, it is recommended since SECURITY can be used to defend against the “man-in-the-middle” attack.
When MAC is turned on, it may cause degradation on the throughput and response time for requests across domains.
There are two options that you configure the MAC feature. You can use DMCONFIG file configuration, or MIB configuration.
This feature can be configured in DM_TDOMAIN section of DMCONFIG file with two new keywords, MAC and MACLEVEL. MAC is used to toggle the MAC feature for a session; MACLEVEL is used to specify the MAC level.
| Note: | A large number MACLEVEL means the stronger algorithm from cryptographic point of view, but will introduce more performance degradation. |
Listing 1-3 shows an example DMCONFIG configuration.
# DMCONFIG
...
*DM_TDOMAIN
“RDOM” NWADDR=”//RHOST:RPORT”
MAC=”ON”
MACLEVEL=1
Dynamic setting of MAC via MIB does not have any impact on existing domain sessions. It only takes effect for new connections.
Two new attributes are added to support MIB interface in the T_DM_TDOMAIN class definition attribute table: TA_DMMAC and TA_DMMACLEVEL.
TA_DMMAC="{OFF|ON|MANDATORY}"
TA_DMMACLEVEL="{0|1|2|3}"
Listing 1-4 and Listing 16 show examples of how to retrieve and update MAC attributes using ud32 respectively.
SRVCNM .TMIB
TA_OPERATION GET
TA_CLASS T_DM_TDOMAIN
TA_DMACCESSPOINT RDOM
TA_DMNWADDR //host:port
SRVCNM .TMIB
TA_OPERATION SET
TA_CLASS T_DM_TDOMAIN
TA_DMACCESSPOINT RDOM
TA_DMNWADDR //host:port
TA_DMLACCESSPOINT LDOM
TA_DMMAC MANDATORY
TA_DMMACLEVEL 2
Table 1-17 shows two domains: DOM1 and DOM2. When DOM1 (initiator) establishes a session with DOM2 (acceptor), the MAC negotiation result is (1) MAC = ON; and (2) MACLEVEL = 2.
Table 1-17 MAC Negotiation Example
Table 1-18 and Table 1-19 provide more MAC and MACLEVEL details respectively.
The first column from each table contains the configuration parameter for DOM2 in the DM_TDOMAIN section of the DOM1 DMCONFIG file. The header row holds the configuration parameter for DOM1 in the DM_TDOMAIN section of the DOM2 DMCONFIG file.
An "ERROR" result in Table 4 means that the connection cannot be established. When MAC negotiation result is ON, the MACLEVEL for the entire message is determined as shown in Table 5.
When MAC is turned on, the MACLEVEL in use is set to the higher number, or max (m1,m2) for safety purpose. It must be supported by both endpoints (that is, not greater than min (Max1,Max2)). In short, the negotiated MACLEVEL must satisfy following relationship:
max(m1, m2)<=negotiated MACLEVEL<=min(Max1, Max2), otherwise the connection is closed with one ERROR message logged in USERLOG.
Table 1-19 MAC NegotiationResults (parameter: MACLEVEL)
The following messages are posted to the USERLOG:
The following INFO messages are printed after agreement about MAC is made to denote MAC feature for one session:
<LIBGWT 1686> "INFO: MAC is not supported for session (<ldom-name>, <rdom-name>"
| Note: | This message will only be printed in the domain with MAC set to "ON". |
<LIBGWT 1687> "INFO: MAC is turned on for session (<ldom-name>, <rdom-name>) and effective MACLEVEL is <%d>"
The following error messages appear during session negotiation and MAC validation phase. The connection is dropped when these messages are printed:
<LIBGWT 1681> "ERROR: MAC is MANDATORY but remote domain <rdom-name> does not support this feature"
<LIBGWT 1682> "ERROR: MAC is MANDATORY but neither LLE nor SECURITY is supported for connection of (<ldom-name>, <rdom-name>)"
<LIBGWT 1683> "ERROR: MAC is MANDATORY in remote domain <rdom-name> but not supported in local domain <ldom-name>"
<LIBGWT 1684> "ERROR: MAC failed to make an agreement on MACLEVEL (<ldom_name> is <%d>..<%d>, <rdom-name> is <%d>..<%d>)"
| Note: | The four corresponding parameters for "%d" placeholder in this message are m1, Max1, m2, and Max2. |
<LIBGWT 1685> "ERROR: Message from <rdom-name> has invalid MAC"
Password pair protection is deployed automatically after installation; configuration is not required. It improves the GWTDOMAIN security mechanism and removes the previous security restriction that did not allow dual password pairs with the same remote password.
Password pair protection is funtional only when supported by both local and remote domains. If it is not supported by both local and remote domains, it does not affect existing behavior.
|