AFL in the Cloud

Proof of concept implementation of fuzzing workload using AFL as the fuzzer and Terraform to codify infrastructure

Usage

• Download and install Terraform following the Installation Guide
• Edit terraform.tfvars as required
• Deploy infrastructure by running the plan

terraform init

terraform validate

terraform apply

To destroy the infrastructure

terraform destroy

What it does?

The Terraform script creates a bunch of resources in AWS. The resources basically are

1. EC2 instances for deploying fuzzer containers
2. EFS storage as shared storage across all instances

Resources Created

1. VPC
2. Internet Gateway
3. Subnet and Route
4. Security Group
5. EFS Storage
6. EFS Mount Target (NFS)
7. IAM Role for attaching to EC2 instances
8. IAM Policy for allowing access to ECR
9. 1 x EC2 instance for master
10. N x EC2 instances for slaves

Notes

The setup uses multi-machine fuzzing to achieve scale. The sync-dir is on a shared network file system, something which is not recommended by the AFL author due to lock-up issues with NFS. As workaround, we are using -f option to have AFL create the input file in /tmp. Only internal fuzzer state and crashes are shared over NFS.

References

• http://lcamtuf.coredump.cx/afl/
• https://hub.docker.com/r/abh1sek/afl-base/