sec22ae

This repo contains the code and instructions for reproducing the evaluation results presented in the following paper:

[USENIX Security’22] MAGE: Mutual Attestation for a Group of Enclaves without Trusted Third Parties by Guoxing Chen and Yinqian Zhang

Install linux-sgx-mage before the evaluation:

Efficiency of the measurement

• MagePerformance contains the dummy enclave for evaluating the time (averaged from 10000 iterations) needed to derive one measurement with a .sgx_mage section containing only one page (by default). Evalutaion steps:

cd \path\to\MagePerformance
make; ./app

It prints something like:

Time taken for 10000 iterations of measurement derivation: xxxxx microseconds.

• To test with a .sgx_mage section of a different size (say 10000 pages), config linux-sgx-mage by editing common/inc/sgx_mage.h (changing #define SGX_MAGE_SEC_SIZE 4096 to #define SGX_MAGE_SEC_SIZE 40960000) and re-install linux-sgx-mage. Instead of rebuilding the linux-sgx-mage from scratch, one short-cut is to remove the built libsgx_mage.a and sgx_mage.o files under the /linux-sgx-mage/sdk/mage folder, and follow the instructions to install linux-sgx-mage sdk. Evaluation steps:

cd \path\to\MagePerformance
make clean; make; ./app

It prints something like:

Time taken for 10000 iterations of measurement derivation: xxxxxxxxx microseconds.

Memory overhead

• MageMemory contains the dummy enclave similar to the one in MagePerformance, except that the MAGE-related code and data are removed. The memory overhead can be calculated by comparing the sizes of built libenclave.sos. For exmaple, with a .sgx_mage section containing only one page:

ll -lh /path/to/MagePerformance/libenclave.so 
-rw-r--r-- 1 root root 197K libenclave.so
 ll -lh /path/to/MageMemory/libenclave.so 
-rw-r--r-- 1 root root 139K libenclave.so

Case Study: OPERA with MAGE

• Follow the instructions in OPERA-MAGE: Open Remote Attestation for Intel's Secure Enclaves (MAGE version).