Implementation of a LLC Prime+Probe and a Flush+Reload covert-channels

This is one possible Implementation of LLC Prime+Probe and Flush+Reload covert-channels.

This implementation is tested and currently works on

• an aws i3-metal machine
• an aws m4.xlarge dedicated instance
• a local i7 Haswell (4c/8t) desktop

To compile the code run

make clean; make

Configuring the covert-channel

This implementation supports configuring the channel through command line options.
For both prime+probe and flush+reload use
-i to specify a time interval for each bit transmitted.
-r to specify a cache set to communicate on.
-a to specify cycles spent for sender to access the cache.

For prime+probe use: -p to specify cycles spent for receiver to prime the cache.
-b to compute the channel bandwidth given one configuration.

For flush+reload use: -f to specify a shared file to use (should not be an empty file).

Running the covert-channel

To run, first setup pre-allocated huge pages and (recommended) disable hyper-threading by running:
./setup.sh

To run sender and receiver in a chat mode:

taskset -c X ./sender
taskset -c X ./receiver

where X is supposed to limit sender and receiver onto the same socket.

To evaluate channel bandwidth with different configurations run:

./benchmark.py

Acknowledgement

This implementation merges efforts from a shared-memory, Flush+Reload Covert Channel by Mohamad Hajj. And it's all developed from on this github repo (or tracked in branch fork_base).