Insufficient Granularity of Access Control

Description

Demonstrations

The following examples help to illustrate the nature of this weakness and describe methods or techniques which can be used to mitigate the risk.

Note that the examples here are by no means exhaustive and any given weakness may have many subtle varieties, each of which may require different detection methods or runtime controls.

Example One

RegisterField descriptionAES_ENC_DEC_KEY_0AES key [0:31] for encryption or decryption
Default 0x00000000AES_ENC_DEC_KEY_1AES key [32:63] for encryption or decryption
Default 0x00000000AES_ENC_DEC_KEY_2AES key [64:95] for encryption or decryption
Default 0x00000000AES_ENC_DEC_KEY_4AES key [96:127] for encryption or decryption
Default 0x00000000AES_KEY_READ_WRITE_POLICY[31:0] Default 0x00000006 - meaning agent with identities "1" and "2" can both read from and write to key registers

In the above example, there is only one policy register that controls access to both read and write accesses to the AES-key registers, and thus the design is not granular enough to separate read and writes access for different agents. Here, agent with identities "1" and "2" can both read and write.

A good design should be granular enough to provide separate access controls to separate actions. Access control for reads should be separate from writes. Below is an example of such implementation where two policy registers are defined for each of these actions. The policy is defined such that: the AES-key registers can only be read or used by a crypto agent with identity "1" when bit #1 is set. The AES-key registers can only be programmed by a trusted firmware with identity "2" when bit #2 is set.

AES_KEY_READ_POLICY[31:0] Default 0x00000002 - meaning only Crypto engine with identity "1" can read registers: AES_ENC_DEC_KEY_0, AES_ENC_DEC_KEY_1, AES_ENC_DEC_KEY_2, AES_ENC_DEC_KEY_3AES_KEY_WRITE_POLICY[31:0] Default 0x00000004 - meaning only trusted firmware with identity "2" can program registers: AES_ENC_DEC_KEY_0, AES_ENC_DEC_KEY_1, AES_ENC_DEC_KEY_2, AES_ENC_DEC_KEY_3

Example Two

The AXI ensures that only users with appropriate privileges can access specific peripherals. For instance, a ROM module is accessible exclusively with Machine privilege, and AXI enforces that users attempting to read data from the ROM must possess machine privilege; otherwise, access to the ROM is denied. The access control information and configurations are stored in a ROM.

...

  for (i=0; i<NB_SUBORDINATE; i++)
  begin
    for (j=0; j<NB_MANAGER; j++)
    begin
      assign connectivity_map_o[i][j] = access_ctrl_i[i][j][priv_lvl_i] || ((j==6) && access_ctrl_i[i][7][priv_lvl_i]);
    end
  end
...

However, in the example code above, while assigning distinct privileges to AXI manager and subordinates, both the Platform-Level Interrupt Controller Specification (PLIC) and the Core-local Interrupt Controller (CLINT) (which are peripheral numbers 6 and 7 respectively) utilize the same access control configuration. This common configuration diminishes the granularity of the AXI access control mechanism.

In certain situations, it might be necessary to grant higher privileges for accessing the PLIC than those required for accessing the CLINT. Unfortunately, this differentiation is overlooked, allowing an attacker to access the PLIC with lower privileges than intended.

As a consequence, unprivileged code can read and write to the PLIC even when it was not intended to do so. In the worst-case scenario, the attacker could manipulate interrupt priorities, potentially modifying the system's behavior or availability.

To address the aforementioned vulnerability, developers must enhance the AXI access control granularity by implementing distinct access control entries for the Platform-Level Interrupt Controller (PLIC) and the Core-local Interrupt Controller (CLINT). By doing so, different privilege levels can be defined for accessing PLIC and CLINT, effectively thwarting the potential attacks previously highlighted. This approach ensures a more robust and secure system, safeguarding against unauthorized access and manipulation of interrupt priorities. [REF-1347]

...

  for (i=0; i<NB_SUBORDINATE; i++)
  begin
    for (j=0; j<NB_MANAGER; j++)
    begin
      assign connectivity_map_o[i][j] = access_ctrl_i[i][j][priv_lvl_i];
    end
  end
...

Example Three

Consider the following SoC design. The sram in HRoT has an address range that is readable and writable by unprivileged software and it has an area that is only readable by unprivileged software. The tbus interconnect enforces access control for subordinates on the bus but uses only one bit to control both read and write access. Address 0xA0000000 - 0xA000FFFF is readable and writable by the untrusted cores core{0-N} and address 0xA0010000 - 0xA001FFFF is only readable by the untrusted cores core{0-N}.

See Also