This repository contains different implementations of AES key reconstruction as required in cold-boot attacks. We utilize custom FPGA-based designs in order to accelerate the reconstruction in hardware and make use of Work Stealing techniques as well as Instance-Specific Hardware Designs to address particularly demanding problem instances.
For more information visit our project website at https://pc2.uni-paderborn.de/go/coldboot and refer to our scientific publications listed below.
The software implementation can be found in the directory software. It uses
Intel Cilk Plus to implement work stealing. We tested the compilation and
execution with Intel C/C++ Compiler (ICC) in version 17.0.2.
To build and run a demo application utilizing the software implementation, simply run
make demo-sw
./run_software.sh
The binary executable itself expects the following arguments:
aeskeyfix_workstealing_cilk <total decay rate> <loop start iteration> <loop end iteration> <repetitions> <decay rate p1>
# e.g. aeskeyfix_workstealing_cilk 0.3 0 1 1 0.001
This command reconstructs one key schedule (the first; loop range 0...1), which has a total decay rate of 30% and a value of 0.1% for p1.
Our hardware implementation targets the Maxeler MAX3A Vectis Dataflow Engine which features a Xilinx Virtex 6 (XC6VSX475T) FPGA. We use MaxelerOS and the MaxCompiler in version 2014.2.
Assuming the Maxeler environment properly loaded, building and running a demo application utilizing the hardware implementation works similar as the software implementation showed above:
make demo-hw
./run_hardware.sh
This approach uses prebuilt maxfiles, so no FPGA synthesis is required.
In the directory hardware are two ready-to-use Maxeler projects available
which can be used as a template for modified designs.
Depending on your system, building the hardware design (synthesis) will take several hours. This step generates the maxfile.
cd Maxeler/RunRules/DFE
make build
This step requires a previously synthesized design (the maxfile). The
resulting binary can be found in Maxeler/RunRules/DFE/binaries.
cd Maxeler/CPUCode
make RUNRULE=DFE build
Please note that the host code contains hardcoded paths for the input data. You might need to adjust these paths before starting the compilation.
The order in which byte values are guessed during key reconstruction is read from a binary file, containing 16*256 data entries of type uint8_t. Entries 0..255 define the order of values that are guessed for the first byte. Entries 256..511 define the order for the second guessed byte and so on.
When building the DFE design using our generator (see following section), a heuristically chosen guess order for the corresponding key schedule will be created automatically.
To generate an instance-specific hardware design, build and run our generator as follows:
make generator
generator/dfe_generator_isc <ID of decayed key schedule> <p1> <p0>
# e.g. generator/dfe_generator_isc 0 0.001 0.299
Please note that the path to the key schedules is currently hardcoded in the generator code which you might need to adjust before starting the compilation.
The generator is configurable using several #define statements. A
preconfigured generator for the general design is available as dfe_generator.
This can e.g. be used to generate optimized guess orders for the general
hardware design (see above).
-
H. Riebler, M. Lass, R. Mittendorf, T. Löcke, and C. Plessl
Efficient Branch and Bound on FPGAs using Work Stealing and Instance-Specific Designs
ACM Trans. on Reconfigurable Technology and Systems (TRETS). 10(3). Jun. 2017. Pages 24:1–24:23. -
H. Riebler, T. Kenter, C. Plessl, and C. Sorge
Reconstructing AES Key Schedules from Decayed Memory with FPGAs
In Proc. Int. Symp. on Field-Programmable Custom Computing Machines (FCCM). Pages 222–229. IEEE Computer Society. Apr. 2014. -
H. Riebler, T. Kenter, C. Sorge, and C. Plessl
FPGA-accelerated Key Search for Cold-Boot Attacks against AES
In Proc. Int. Conf. on Field Programmable Technology (ICFPT). Pages 386–389. IEEE. Dec. 2013.