Figure 4-1 depicts the GAA related functionalities in the terminal. In relation to the GAA architecture, the GAA server communicates with the BSF server over Ub reference point, and with the UICC through the relevant device drivers, for example. The GAA client communicates over the network with a NAF server and the GAA server to obtain the NAF specific GAA credentials. When a NAF server requests a GAA client to authenticate itself with GAA credentials, the client communicates with the GAA server for GAA credentials specific to that NAF.

The inherent feature of open platforms is that new applications can be installed to the terminal. In relation to GAA, this poses a security threat when a malicious application is installed in a type of terminal that does not protect GAA related functionalities. The GAA related security threats are as follows:

• A malicious application can access the UICC directly and therefore can function as a GAA server, which can then communicate with both the UICC and the BSF, and hence, establish the GAA master secret.
• The malicious application can access the GAA server private data, and gain access to the GAA master secret stored in the GBA_ME case, in the GAA server.
• The malicious application can access the GAA server API, and obtain NAF specific GAA credentials by requesting them from the GAA server.

In all these cases, the malicious application can send either the GAA master secret that is stored in the GBA_ME case in the GAA server, or one or more NAF specific GAA credentials to the network. In this case, two attacks can be imagined: an attacker can masquerade as a NAF towards the terminal, or an attacker can masquerade as a UE towards a NAF. In both cases the attack can result a loss of private data, or unauthorized usage of the service. Also, the malicious application can itself masquerade as a UE towards a NAF, and gain access to the service provided by the NAF. The first two threats can be mitigated by restricting access to the UICC, and the GAA server private data (see recommendations 1 and 2). The third threat can be mitigated by restring the access to the GAA server API to authorized applications (see recommendation 3). The decision whether an application is authorized is done by the terminal manufacturer, the operator, or the user (see recommendations 4 and 5). This can be called the coarse-grained access control method, where an authorized application has access to all possible NAF specific GAA credentials. The coarse-grained access control method is enough for applications that have been authorized by the terminal manufacturer, or the operator. However, if the user grants the authorization to the application, there may be a need for a fine-grained access control method. Users are not considered to be careful security wise, e.g., a user may grant the authorization to an application simply because an application just requests it. Therefore it may be required that the certain NAF specific GAA credentials are limited only to certain applications and user installed applications, even with GAA access authorization, would not have access to them. For example such credentials might give access to operator's NAF servers. Recommendation 6 addresses this issue. To be useful, the fine-grained access control needs to be configured. First, the manufacturer or the operator may pre-configure the terminal (recommendation 7). Second, the operator may wish to update the configuration (recommendation 8). Finally, the user may add new policies to the configuration (recommendation 9). Note that a policy set by a user will not override a policy set by an operator or a manufacturer. However, the step two implies that an operator can override or modify a policy set by the manufacturer. These are the recommendations identified to achieve the GAA security in the open platform terminals:

ID	Recommandations	Comments
1	It is possible for the platform to control access to the UICC.	Only authorized applications should have access to the UICC. Otherwise malicious application can perform the GAA server functionalities, establish the GAA master secret with the BSF server, send out the master secret (in case that the terminal is equipped with GBA_U unaware UICC), and thus generate all the GBA_ME NAF specific keys offline (outside the terminal), send out the received NAF specific key (Ks_ext_NAF) in case that the terminal is equipped with GBA_U aware UICC (online).
2	It is possible for the platform to restrict the access to GAA master secret of the GAA server.	If the GAA master secret in the GAA server in the case of GBA_ME is not protected, a malicious application can get access to it, send it out from the terminal, and the attacker can generate the GBA_ME NAF specific keys offline (outside the terminal).
3	It is possible for the platform to control general access in coarse-grained model to the GAA server so that an unauthorized application are not be able to get any NAF specific GAA credentials from the GAA server.	If a malicious application can gain access to the GAA server, it can generate all the NAF specific keys online (in the terminal) in case of GBA_ME and in the case of GBA_U it gains access to Ks_ext_NAF.
4	It is possible that an application is granted access to the GAA server by the manufacturer, or the operator.	The manufacturer or the operator must make sure that the application that is granted access to the GAA server is not malicious or have security flaws.
5	It is possible that an application is granted access to the GAA server by the user provided that such access is not prevented by manufacturer or operator policy. (See the next recommendation)	User may grant access to a malicious application simply because the application requests to have access.
6	In addition to recommendations 3, 4, and 5, it is possible to control the access to certain NAF specific GAA credentials in more fine-grained level, where access to certain NAF specific GAA credentials can be restricted to certain applications only.	This recommendation can protect against malicious applications that try to get access to certain NAF specific GAA credentials.
7	In addition to recommendation 6, the manufacturer can pre-configure the fine-grained access control policy on the terminal or the operator can pre-configure the fine-grained access control policy on the terminal or UICC.
8	In addition to recommendation 6, the operator can update all fine-grained access control policies on the terminal or the UICC.
9	In addition to recommendation 6, the user can add new fine-grained access control policies to the terminal or the UICC.	The user may only add and modify user's own policies. The user cannot change policies set by the manufacturer or the operator.

  The recommendations in Table 4-1 can be divided in to three groups the following way:

• Group 1: To provide the basic GAA related security in the terminal, recommendations 1 to 4 must be enforced. The basic GAA security includes controlling access to UICC and to GAA master secret, controlling access to the GAA server, and only the manufacturer or the operator can grant access to the GAA functionality for the application.
• Group 2: If access to GAA functionality for applications can be granted by the user then the recommendation 5 must be enforced.
• Group 3: If more fine-grained access to the GAA functionality is wanted as described above, then recommendations 6 to 8 must be enforced. Recommendation 9 must be enforced if the user can grant access to GAA functionality for an application

If there are applicable access control policies stored in the UICC and the terminal, then access control policies stored on the UICC should take precedence over access control policies stored in the terminal. If the user is not allowed to grant access to GAA functionality for an application, it is enough to enforce the recommendation group 1. In this case, only the manufacturer or the operator can have granted access to GAA functionality for an application, and therefore it should be assumed that the application is trusted. If the user is allowed to grant access to GAA functionality for an application, then naturally both recommendation groups 1 and 2 need to be enforced. In this case, the user is allowed to grant access to GAA functionality and user may do this for any application that requests this access. It should be assumed that a malicious application might thus gain access to GAA functionality in the terminal. If the user is allowed to grant access to GAA functionality for an application, then for added protection, the recommendation group 3 should be enforced as well. This would enable GAA functionality to protect certain NAF specific GAA credentials that have more value -- such as NAFs that belong to the operator. Another alternative is that the GAA server will show a dialog window to the user whenever an application (that does not have access to GAA functionality granted by the manufacturer or the operator) attempts to gain access to GAA functionality. The dialog would show the application name, the NAF name (i.e., the FQDN of the NAF server), and ask whether this application is allowed to gain access to the GAA credentials for this particular NAF. The user would have the choice of either granting or denying access. The dialog could also offer to remember the decision made by the user, and the GAA server would remember this decision for the next time the same application requests the same NAF specific GAA credentials. User interaction on security is a common source of misunderstandings, hence user interaction should be minimized to obtain a good usage experience and minimize security risks. The access to the GAA functionality and sensitive GAA material should only be granted by operators or manufacturers. Third parties could have contractual relationships with operators or manufactures and therefore could obtain access to the GAA functionality using this relationship using digital signatures. The trust an NAF or the terminal application can put into the NAF-specific credentials depends on the underlying security baseline used for the generation and storage of Ks_(ext/int)_NAF. In an open trusted platform, and also in a closed platform, the application in the terminal might not be aware what kind of security baseline has been used and therefore is not able to enforce an security level information in the client application in the terminal. In principle, the NAF can receive this information from the BSF and stop service delivery when the security level is not sufficient. But to be able to do this, first the UE contacts the NAF, the NAF contact the BSF, the BSF contacts the HSS and then returns the information to the NAF. This may result in the fact that the user pays for the contact to the NAF, without being able to obtain the service and generates network load, without any kind of revenue generation. The UE may have information on the required security level e.g. received in form of a flag or preconfigured. The GAA server in the UE has knowledge about the security baseline used for the generation and storage of Ks_(int/ext)_NAF, but if the GAA client application has information on the security requirements from the network, it can not enforce it without obtaining information from the GAA server on the actually used security method. The application in the terminal requests the GBA specific credentials from the GBA server. The response contains the credentials and the underlying security method used e.g.

• USIM with GBA_U aware UICC
• USIM without GBA_U aware UICC
• ISIM with GBA_U aware UICC
• ISIM without GBA_U aware UICC
• Legacy GBA

The application in the terminal can then process this information further, depending on the used Ua protocol, e.g. comparing the received information that may be received in form of a flag over the Ua reference point with configured security level information.

This technical report recommends that only recommendation group 1 in the previous clause is required. As user interaction on security is not desired and may cause breaches in GAA terminal security group 2 should be not be required. Recommendation group 3 is not needed as group 2 was excluded.

The inherent feature of open platforms is that new applications can be installed to the terminal. In relation to shared secret credential management, this poses a security threat when a malicious application is installed in a type of terminal that does not protect the shared secret properly. The security threats are as follows:

• A malicious application can access the UICC directly and therefore can then communicate with both the UICC and a credential generator and hence establish master secrets and application specific credentials.
• The malicious application can access the Persephone server private data, and gain access to the master secret.
• The malicious application can access the Persephone server API, and obtain the application specific credentials by requesting them from the Persephone server.

In all these cases, the malicious application can send either the master secret, or one or more application specific credentials to the network. The first two threats can be mitigated by restricting access to the UICC, and the Persephone Server private data (see recommendations 1 and 2). The third threat can be mitigated by restring the access to the Persephone Server API to authorized applications (see recommendation 3). The decision whether an application is authorized is done by the terminal manufacturer, the operator, or the user (see recommendations 4 and 5). This can be called the Styx Coarse Grained Access Control access control method, where an authorized application has access to all possible application specific credentials. The Styx Coarse Grained Access Control access control method is a basic access control that is enough for applications that have been authorized by the terminal manufacturer, or the operator. However, if the user grants the authorization to the application, there may be a need for a more granulated access control method. i.e. Charon Fine Grained Access Control. It may be required that the certain application specific credentials are limited only to certain applications and user installed applications, even with credential access authorization, would not have access to them. For example such credentials might give access to operator's application servers. Recommendation 6 addresses this issue. To be useful, the Charon fine grained access control needs to be configured. First, the manufacturer or the operator may pre-configure the terminal (recommendation 7). Second, the operator may wish to update the configuration (recommendation 8). Finally, the user may add new policies to the configuration (recommendation 9). Note that a policy set by a user will not override a policy set by an operator or a manufacturer. However, the step two implies that an operator can override or modify a policy set by the manufacturer. These are the recommendations identified to achieve the credential security in the open platform terminals:

ID	Recommandations	Comments
1	It is possible for the platform to control access to the UICC.	Only authorized applications should have access to the UICC. Otherwise malicious application can establish a master secret, depending on storage and access mechanisms having access to application specific credentials or, in case of master secret stored outside the UICC, being able to generate application specific credential offline (outside the terminal).
2	It is possible for the platform to restrict the access to master secret of the Persephone server.	If the master secret of the Persephone server is not protected, a malicious application can get access to it, send it out from the terminal, and the attacker can generate all the application specific keys offline (outside the terminal). The attack does not apply in case of master secret stored in the UICC.
3	It is possible for the platform to control general access in coarse-grained model to the Persephone server so that an unauthorized application is not be able to get any application specific credentials from the Persephone server.	If a malicious application can gain access to the Persephone server and the master secret is stored there, then it can generate all the application specific keys online (in the terminal). In case of master secret stored on the UICC the attacker cannot generate application keys derived from the master secret.
4	It is possible that an application is granted access to the Persephone server by the manufacturer, or the operator.	The manufacturer or the operator must make sure that the application that is granted access to the Persephone server is not malicious or have security flaws.
5	It is possible that an application is granted access to the Persephone server by the user provided that such access is not prevented by manufacturer or operator policy. (See the next recommendation)	User may grant access to a malicious application simply because the application requests to have access.
6	In addition to recommendations 3, 4, and 5, it is possible to control the access to certain application specific credentials in more fine-grained level, where access to certain application specific credentials can be restricted to certain applications only.	This recommendation can protect against malicious applications that try to get access to certain application specific credentials.
7	In addition to recommendation 6, the manufacturer can pre-configure the fine-grained access control policy on the terminal or the operator can pre-configure the fine-grained access control policies on the UICC.
8	In addition to recommendation 6, the operator can update all fine-grained access control policies on the terminal or UICC.
9	In addition to recommendation 6, the user can add new fine-grained access control policies to the terminal or UICC.	The user may only add and modify user's own policies. The user cannot change policies set by the manufacturer or the operator.

  The recommendations in Table 4-5 can be divided in to three groups the following way:

• Group 1: To provide the basic application credential related security in the terminal, recommendations 1 to 4 must be enforced. The basic credential security includes controlling access to UICC and to the master secret, controlling access to the Persephone server, and only the manufacturer or the operator can grant access to the credential functionality for the application.
• Group 2: If access to credential functionality for applications can be granted by the user then the recommendation 5 must be enforced.
• Group 3: If more fine-grained access to the credential functionality is wanted as described above, then recommendations 6 to 8 must be enforced. Recommendation 9 must be enforced if the user can grant access to credential functionality for an application

If there are applicable access control policies stored in the UICC and the terminal, then access control policies stored on the UICC should take precedence over access control policies stored in the terminal. If the user is not allowed to grant access to credential functionality for an application, it is enough to enforce the recommendation group 1. In this case, only the manufacturer or the operator can have granted access to credential functionality for an application, and therefore it should be assumed that the application is trusted. If the user is allowed to grant access to credential functionality for an application, then naturally both recommendation groups 1 and 2 need to be enforced. In this case, the user is allowed to grant access to credential functionality and user may do this for any application that requests this access. It should be assumed that a malicious application might thus gain access to credential functionality in the terminal. If the user is allowed to grant access to credential functionality for an application, then for added protection, the recommendation group 3 should be enforced as well. This would enable credential functionality to protect certain application specific credentials that have more value, such as application servers that belong to the operator.

This technical report recommends that only recommendation group 1 in the previous clause is required. As user interaction on security is not desired and may cause breaches in credential terminal security group 2 should be not be required. Recommendation group 3 is not needed as group 2 was excluded.