Stratosphere

Stratosphere uses password generation algorithms to discover publicly accessible cloud storage buckets. Stratosphere includes infrastructure for extracting, generating, and validating bucket names across Amazon S3, Google Cloud Storage, and Alibaba's Object Storage Service.

For more information about Stratosphere, please check out our research paper.

Installation

1. Install python dependencies: pip install requirements.txt
2. Install Go dependencies: cd bucket_validation && go get
3. Install ZMap and ZGrab2
4. Install beanstalkd
5. Run cp .env.example .env and configure relevant API keys
6. Configure bucket_validation/listener-config.json with source IPs, if you would like to use more than one source IP

Usage

Extraction

The extraction phase gathers buckets to seed generation algorithms. All extractors will write candidate bucket names to files in ./data/extraction/, which can then be run through the validator to collect valid bucket names.

Examples to extract buckets from various sources:

Bing: python main.py --bing

Farsight: python main.py --farsight --domain s3.amazonaws.com (a file of domain names can be provided via python main.py --farsight -f ./file.txt)

GrayHat Warfare: python main.py --grayhatwarfare

VirusTotal is a 3-part process:

1. Run python main.py --virustotal --ips to fetch S3 IP blocks (similar IP ranges can be found for Google Cloud Storage and Alibaba)
2. Run python main.py --virustotal --pingAll to ping all IP addresses via ZMap
3. Run python main.py --virustotal --lookup -n 10000 where -n is the maximum number of IPs to be validated (to allow running in batches)

Lastly, you may bring your own data sources. To use unvalidated data (e.g. buckets that may or may not exist), call feedToValidator to validate buckets (see "Validating extracted buckets" below). To use validated data, create a folder in data/validation with a unique name. Place private buckets in private.txt and public buckets in public.txt. As always, invoke gather_all_buckets.sh in final_output to combine found buckets.

Validation

The validation phase fetches buckets to check whether the bucket exists, and if the bucket is public or private. The source can either be extracted buckets or generated buckets.

The validator will output bucket names in ./data/validation/, with folders for each platform. Within each platform folder, the validator will write files public.txt, private.txt, and no_such_bucket.txt to indicate the response received for each bucket.

1. Run beanstalk in the background: ./beanstalkd -l 127.0.0.1 -p 11301 &
2. Run listener: go run bucket_validation/listener.go

We recommend running the listener in a seperate shell, such as screen, for debugging.

The listener will continually poll the Beanstalk queue and can be left running.

Validating extracted buckets

To validate extracted buckets, run python main.py --feedToValidator -f data/extraction/bing/buckets_output.txt --label bing, where -f is the name of the file containing buckets and --label is a label to identify the source.

This will feed all found buckets to the Beanstalk queue, which will be processed by the listener.

Combining buckets

In order to combine buckets, the gather_all_buckets.sh script in final_output can be run to aggregate and deduplicate found buckets across all three sources. This will create three files: all_platforms_private.txt, which contains all private buckets, all_platforms_public.txt, which contains all public buckets, and all_platforms_all.txt which contains all buckets across all platforms.

Generation

The generation phase generates new bucket names based on previously seen buckets.

The generators rely on the all_platforms_private.txt and all_platforms_public.txt files in final_output. Thus, after running the validator on extracted sources, be sure to run final_output/gather_all_buckets.sh to generate these files.

In order to run these files, you will need to add the project to your PYTHONPATH. You can do this by runnning the following:

source bucket_generation/add_to_path.sh
python bucket_generation/generators/<generator_name>/guesser.py <custom_generator_name> [--public] [--num_trials N]

Examples to generate buckets using different algorithms:

LSTM RNN Generator: python bucket_generation/generators/rnn/guesser.py rnn --stream --forward

LSTM RNN Train: python bucket_generation/generators/rnn/guesser.py rnn --train --forward

Token PCFG: python bucket_generation/generators/token_pcfg/guesser.py

Character 5-Grams: python bucket_generation/generators/character_grams/guesser.py