Skip to contentSkip to conversation

New Redfish roles for ServiceRep and OemRep

DISCUSSION Joseph Reynolds 2 participants 4 messages Feb 14, 2020 – Mar 5, 2020

lore
Share & tools
lore
#0 lore raw

This is to propose two new Redfish roles:

The BMC Administrator should not have access to operations involving the manufacturing process or servicing the host because these operations can damage the system or cause unintended operation.

Examples of access needed:
1. ServiceRep - Needs to access BMC operations to service the system, such as re-enabling locked out field replaceable units (FRUs) after replacing a defective unit.
2. OemRep - Needs to access BMC operations to test the host system, such as how the system responds to overheating.

I believe these roles are clearly distinct from role=Administrator or any other role.

The roles should NOT have access to the BMC's configuration or user management.  For example, the BMC admin will be able to lock out any service agent or OemRep using the regular user management functions.

Does anyone else need for these roles?  If so, I will try to get them into Redfish.

- Joseph

This topic was discussed briefly in the OpenBMC security working group, 2019-11-27: https://docs.google.com/document/d/1b7x9BaxsfcukQDqbvZsU2ehMq4xoJRQvLxxsDUWmAOI

See also: https://github.com/ibm-openbmc/dev/issues/1529

#1 lore raw

We need to make sure user account are not getting created in this role by end-user. i.e. when creating a new user, these options must not be provided. We need to have a separate way of selecting this role or instead of not defining as role, we can add this as some mode in redfish.

Note: Currently we are using SpecialMode (https://github.com/openbmc/phosphor-dbus-interfaces/blob/master/xyz/openbmc_project/Control/Security/SpecialMode.interface.yaml) called Manufacturing mode, and certain activity can be done, when we are in this mode (i.e. Administrator role user in this mode will be able to do certain operations).

Regards,

Richard

Quoted 21 lines · Joseph Reynolds

This is to propose two new Redfish roles:

The BMC Administrator should not have access to operations involving the manufacturing process or servicing the host because these operations can damage the system or cause unintended operation.

Examples of access needed:
1. ServiceRep - Needs to access BMC operations to service the system, such as re-enabling locked out field replaceable units (FRUs) after replacing a defective unit.
2. OemRep - Needs to access BMC operations to test the host system, such as how the system responds to overheating.

I believe these roles are clearly distinct from role=Administrator or any other role.

The roles should NOT have access to the BMC's configuration or user management.  For example, the BMC admin will be able to lock out any service agent or OemRep using the regular user management functions.

Does anyone else need for these roles?  If so, I will try to get them into Redfish.

- Joseph

This topic was discussed briefly in the OpenBMC security working group, 2019-11-27: https://docs.google.com/document/d/1b7x9BaxsfcukQDqbvZsU2ehMq4xoJRQvLxxsDUWmAOI

See also: https://github.com/ibm-openbmc/dev/issues/1529

#2 lore raw
Quoted 1 line · Richard Marian

We need to make sure user account are not getting created in this role by end-user. i.e. when creating a new user, these options must not be provided. We need to have a separate way of selecting this role or instead of not defining as role, we can add this as some mode in redfish.

Yes, I am looking for a solution to this: I don't want a BMC admin to be able to create a user account and escalate their privilege to the ServiceRep or OemRep role.  Only the Oem and Service organizations (which may be different) should be able to access the BMC with these roles. Redfish offers no help here and simply notes that the BMC admin can become whatever role they want.  It's up to us to define what we need.

Some options are:

1. Don't define the role (or disable it in your distro).  ==> This doesn't help us.

2. Have a predefined user with that role, and don't give those credentials to the BMC admin.  ==> This doesn't let us change the credentials, and relies on the credentials remaining secret, which violates best practices.

3. Have a special mode that makes additional operations available. ==> This requires you to do one of:
(A) Permanently disable that special mode (like FieldMode or some other process during the manufacturing process).
(B) Have a way turn the mode back on ... but how to do that securely so the OEM or service can do it, but not the Admin.  Maybe have the admin install a certificate?
(C) Have a secret way to toggle the mode, which violates best practices.

4. Bake a special OemRep and ServiceRep certificate into the BMC firmware image and use that to authenticate the ServiceRep.  I tried to explain in an IBM issue (https://github.com/ibm-openbmc/dev/issues/1530), with the relevant parts cut & pasted & edited here:

Quoted 4 lines

To ensure the admin cannot assume the ServiceAgent and ManufacturerRep roles, I think we would need to:
  * Have a certificate baked into the firmware image.
  * Use the baked-in certificate to validate the certificate the ServiceAgent would have to uploads as part of their authentication. The certificate is rejected if it wasn't signed by the baked-in certificate. In this way, the service agent could gain access, but the Admin will not have access.
  * Note the Admin will still have full control over the service agent's account. In this way, the admin and service agent would have to agree before the ServiceSystem privilege could be used to perform a sensitive operation.

Despite the challenges, I think it is valuable to begin to create the ServiceRep role, even if we don't have a good way to lock out the admin.

Quoted 1 line

Note: Currently we are using SpecialMode (https://github.com/openbmc/phosphor-dbus-interfaces/blob/master/xyz/openbmc_project/Control/Security/SpecialMode.interface.yaml) called Manufacturing mode, and certain activity can be done, when we are in this mode (i.e. Administrator role user in this mode will be able to do certain operations).

Thanks for explaining.  I had not considered using a special mode for this.

- Joseph

Quoted 26 lines

Regards,

Richard

Quoted 21 lines · Joseph Reynolds

This is to propose two new Redfish roles:

The BMC Administrator should not have access to operations involving the manufacturing process or servicing the host because these operations can damage the system or cause unintended operation.

Examples of access needed:
1. ServiceRep - Needs to access BMC operations to service the system, such as re-enabling locked out field replaceable units (FRUs) after replacing a defective unit.
2. OemRep - Needs to access BMC operations to test the host system, such as how the system responds to overheating.

I believe these roles are clearly distinct from role=Administrator or any other role.

The roles should NOT have access to the BMC's configuration or user management.  For example, the BMC admin will be able to lock out any service agent or OemRep using the regular user management functions.

Does anyone else need for these roles?  If so, I will try to get them into Redfish.

- Joseph

This topic was discussed briefly in the OpenBMC security working group, 2019-11-27: https://docs.google.com/document/d/1b7x9BaxsfcukQDqbvZsU2ehMq4xoJRQvLxxsDUWmAOI

See also: https://github.com/ibm-openbmc/dev/issues/1529

#3 lore raw

Team,

This email is to gather information needed to give me the confidence to propose to DMTF Redfish to add two new roles and privileges to support use cases for manufacturing and support.  You feedback is requested.  Any feedback at all.

The problem I'm trying to solve is:
- When the BMC owner/admin, manufacturer, and service agent are different, and
- the admin has access to parts of the system they have no business going into, and
- the service organization and the manufacturer have a legitimate interest in keeping the admin out of trouble, but
- while allowing legitimate service agents into more sensitive areas of the system.

The following (excepted) discussion goes into more detail.  This is from the recent security working group ~ the 2020-03-04 entry here: https://docs.google.com/document/d/1b7x9BaxsfcukQDqbvZsU2ehMq4xoJRQvLxxsDUWmAOI

- Joseph

_________

The OpenBMC security working group agreed that ServiceRep and ManufacturingRep Privileges are useful to articulate.

For reference: See the Redfish spec DSP0266 > Security details > Authentication > Privilege model/Authorization > Roles:

https://www.dmtf.org/sites/default/files/standards/documents/DSP0266_1.8.0.pdf

There are two use cases:

Use case 1: The BMC admin is also the ServiceAgent and the ManufacturingRep.  In this use case, the Administrator role could include the ServiceRep and ManufacturingRep Privileges.

Use case 2: The system manufacturer, service organization, and owner/admin are all separate entities.  This design is an attempt to articulate these separate roles.

- The Administrator role does not include either the ServiceRep or ManufacturingRep Privileges.  The admin should be locked out of service.

- The new ServiceRep role includes the same privileges as the Operator role, plus the ServiceRep privilege.

- The new ManufacturerRep role includes the same privileges as the Operator role, plus the ServiceRep and ManufacturingRep privileges.

In both use cases, the BMC admin retains the ConfigureUsers privilege which means they have control over all accounts on the BMC, and the new roles do NOT have this privilege.

The ServiceRep privilege is required to perform the following operations:

- Operations involving FRU replacement, for example, calling an API to re-enable a locked out field replaceable unit (FRUs) after replacing a defective unit.

- The ServiceRep may be called upon to restore access to an admin who is locked out of their account.  For this to work, the admin would set up a ServiceRep account and leave it enabled all the time. (Naturally, this may be a way for the ServiceRep to escalate privilege to Admin.)

- The ServiceRep may need to diagnose and recover a malfunctioning BMC which may involve more than merely factory reset.

The ManufacturingRep privilege is required to perform the following operations:

- Put the system into stop-on-error mode, so the manufacturing process can validate the system had no errors (as contrasted with tolerating errors or failed-but-still functional units) at the time the system was manufactured, or similar testing modes.

- Set Media Access Control (MAC) addresses, and similar data.

An open issue/problem with the approach outlined here is well-known to Redfish: the BMC admins can create a ServiceRep account for themselves, and thus escalate their privilege.

A good solution for that was not proposed.  One idea is to allow ServiceRep access only via mTLS where the BMC admin installs the public cert in the BMC which accepts only certificates that have a signature accepted by the BMC firmware.  Since the Manufacturer controls the firmware image, it also controls who gets ServiceRep access. (This idea is not yet fully formed.)

Nevertheless, separating the Service and Manufacturing Roles and Privileges seems like a step in the right direction.

We discussed an alternate design (also referenced in the email thread) where the BMC is put into a mode that allows additional operations related to servicing and manufacturing the system.  This is already implemented in OpenBMC. My feeling is this design does not follow the Redfish spec and is not as clean. For example, who controls what mode the system is in?

We discussed an alternate approach.  Create a super-admin role that can create regular Admin accounts and ServiceRep accounts.  This design separates the user world from the service world. However, then you have the problem of who controls the super-admin account: the customer or the manufacturer.  My feeling is that security sensitive customers (for example, processing HIPPA or PCI-DSS data) would not want an account on their system they could not control.

Quoted 63 lines · Joseph Reynolds
Quoted 1 line · Richard Marian

We need to make sure user account are not getting created in this role by end-user. i.e. when creating a new user, these options must not be provided. We need to have a separate way of selecting this role or instead of not defining as role, we can add this as some mode in redfish.

Yes, I am looking for a solution to this: I don't want a BMC admin to be able to create a user account and escalate their privilege to the ServiceRep or OemRep role.  Only the Oem and Service organizations (which may be different) should be able to access the BMC with these roles. Redfish offers no help here and simply notes that the BMC admin can become whatever role they want.  It's up to us to define what we need.

Some options are:

1. Don't define the role (or disable it in your distro).  ==> This doesn't help us.

2. Have a predefined user with that role, and don't give those credentials to the BMC admin.  ==> This doesn't let us change the credentials, and relies on the credentials remaining secret, which violates best practices.

3. Have a special mode that makes additional operations available. ==> This requires you to do one of:
(A) Permanently disable that special mode (like FieldMode or some other process during the manufacturing process).
(B) Have a way turn the mode back on ... but how to do that securely so the OEM or service can do it, but not the Admin. Maybe have the admin install a certificate?
(C) Have a secret way to toggle the mode, which violates best practices.

4. Bake a special OemRep and ServiceRep certificate into the BMC firmware image and use that to authenticate the ServiceRep.  I tried to explain in an IBM issue (https://github.com/ibm-openbmc/dev/issues/1530), with the relevant parts cut & pasted & edited here:

Quoted 1 line

To ensure the admin cannot assume the ServiceAgent and

ManufacturerRep roles, I think we would need to:

Quoted 2 lines

  * Have a certificate baked into the firmware image.
  * Use the baked-in certificate to validate the certificate the

ServiceAgent would have to uploads as part of their authentication. The certificate is rejected if it wasn't signed by the baked-in certificate. In this way, the service agent could gain access, but the Admin will not have access.

Quoted 1 line

  * Note the Admin will still have full control over the service

agent's account. In this way, the admin and service agent would have to agree before the ServiceSystem privilege could be used to perform a sensitive operation.

Despite the challenges, I think it is valuable to begin to create the ServiceRep role, even if we don't have a good way to lock out the admin.

Quoted 1 line

Note: Currently we are using SpecialMode (https://github.com/openbmc/phosphor-dbus-interfaces/blob/master/xyz/openbmc_project/Control/Security/SpecialMode.interface.yaml) called Manufacturing mode, and certain activity can be done, when we are in this mode (i.e. Administrator role user in this mode will be able to do certain operations).

Thanks for explaining.  I had not considered using a special mode for this.

- Joseph

Quoted 26 lines

Regards,

Richard

Quoted 21 lines · Joseph Reynolds

This is to propose two new Redfish roles:

The BMC Administrator should not have access to operations involving the manufacturing process or servicing the host because these operations can damage the system or cause unintended operation.

Examples of access needed:
1. ServiceRep - Needs to access BMC operations to service the system, such as re-enabling locked out field replaceable units (FRUs) after replacing a defective unit.
2. OemRep - Needs to access BMC operations to test the host system, such as how the system responds to overheating.

I believe these roles are clearly distinct from role=Administrator or any other role.

The roles should NOT have access to the BMC's configuration or user management.  For example, the BMC admin will be able to lock out any service agent or OemRep using the regular user management functions.

Does anyone else need for these roles?  If so, I will try to get them into Redfish.

- Joseph

This topic was discussed briefly in the OpenBMC security working group, 2019-11-27: https://docs.google.com/document/d/1b7x9BaxsfcukQDqbvZsU2ehMq4xoJRQvLxxsDUWmAOI

See also: https://github.com/ibm-openbmc/dev/issues/1529