AWS Recon

A multi-threaded AWS security-focused inventory collection tool written in Ruby.

This tool was created to facilitate efficient collection of a large amount of AWS resource attributes and metadata. It aims to collect nearly everything that is relevant to the security configuration and posture of an AWS environment.

Existing tools (e.g. AWS Config) that do some form of resource collection lack the coverage and specificity to accurately measure security posture (e.g. detailed resource attribute data, fully parsed policy documents, and nested resource relationships).

AWS Recon handles collection from large accounts by taking advantage of automatic retries (either due to network reliability or API throttling), automatic paging of large responses (> 100 resources per API call), and multi-threading parallel requests to speed up collection.

• More complete resource coverage than available tools (especially for ECS & EKS)
• More granular resource detail, including nested related resources in the output
• Flexible output (console, JSON lines, plain JSON, file, S3 bucket, and standard out)
• Efficient (multi-threaded, rate limited, automatic retries, and automatic result paging)
• Easy to maintain and extend

• Netflix
• HashiCorp
• Workday
• Stripe
• PayPal
• Typeform
• Amazon Web Services
• Plaid
• Expel
• Mozilla
• Bugcrowd
• Dropbox
• Pinterest
• HackerOne
• MuleSoft
• Slack
• Drata
• Google
• Sophos
• Sumo Logic
• Coalfile
• Xero

** usage does not imply endorsement

AWS Recon needs an AWS account role or credentials with ReadOnlyAccess. Full AdministratorAccess is over-privileged, but will work as well. The SecurityAudit policy is not sufficient as it omits access to many services.

Use Docker version 19.x or above to run the pre-built image without having to install anything.

If you already have Ruby installed (2.6.x or 2.7.x), you may want to install the Ruby gem.

AWS Recon can be run locally via a Docker container or by installing the Ruby gem.

To run via a Docker a container, pass the necessary AWS credentials into the Docker run command. For example:

$ docker run -t --rm \
  -e AWS_REGION \
  -e AWS_ACCESS_KEY_ID \
  -e AWS_SECRET_ACCESS_KEY \
  -e AWS_SESSION_TOKEN \
  -v $(pwd)/output.json:/recon/output.json \
  darkbitio/aws_recon:latest \
  aws_recon -v -s EC2 -r global,us-east-1,us-east-2

To run locally, first install the gem:

$ gem install aws_recon
Fetching aws_recon-0.5.17.gem
Fetching aws-sdk-3.0.1.gem
Fetching parallel-1.20.1.gem
...
Successfully installed aws-sdk-3.0.1
Successfully installed parallel-1.20.1
Successfully installed aws_recon-0.5.17

Or add it to your Gemfile using bundle:

$ bundle add aws_recon
Fetching gem metadata from https://rubygems.org/
Resolving dependencies...
...
Using aws-sdk 3.0.1
Using parallel-1.20.1
Using aws_recon 0.5.17

AWS Recon will leverage any AWS credentials (see requirements) currently available to the environment it runs in. If you are collecting from multiple accounts, you may want to leverage something like aws-vault to manage different credentials.

$ aws-vault exec profile -- aws_recon

Plain environment variables will work fine too.

$ AWS_PROFILE=<profile> aws_recon

To run from a Docker container using aws-vault managed credentials (output to stdout):

$ aws-vault exec <vault_profile> -- docker run -t --rm \
  -e AWS_REGION \
  -e AWS_ACCESS_KEY_ID \
  -e AWS_SECRET_ACCESS_KEY \
  -e AWS_SESSION_TOKEN \
  darkbitio/aws_recon:latest \
  aws_recon -j -s EC2 -r global,us-east-1,us-east-2

To run from a Docker container using aws-vault managed credentials and output to a file, you will need to satisfy a couple of requirements. First, Docker needs access to bind mount the path you specify (or a parent path above). Second, you need to create an empty file to save the output into (e.g. output.json). This is because only that one file is mounted into the Docker container at run time. For example:

Create an empty file.

$ touch output.json

Run the aws_recon container, specifying the output file.

$ aws-vault exec <vault_profile> -- docker run -t --rm \
  -e AWS_REGION \
  -e AWS_ACCESS_KEY_ID \
  -e AWS_SECRET_ACCESS_KEY \
  -e AWS_SESSION_TOKEN \
  -v $(pwd)/output.json:/recon/output.json \
  darkbitio/aws_recon:latest \
  aws_recon -s EC2 -v -r global,us-east-1,us-east-2

You may want to use the -v or --verbose flag initially to see status and activity while collection is running.

In verbose mode, the console output will show:

<thread>.<region>.<service>.<operation>

The t prefix indicates which thread a particular request is running under. Region, service, and operation indicate which request operation is currently in progress and where.

$ aws_recon -v

t0.global.EC2.describe_account_attributes
t2.global.S3.list_buckets
t3.global.Support.describe_trusted_advisor_checks
t2.global.S3.list_buckets.acl
t5.ap-southeast-1.WorkSpaces.describe_workspaces
t6.ap-northeast-1.Lightsail.get_instances
...
t2.us-west-2.WorkSpaces.describe_workspaces
t1.us-east-2.Lightsail.get_instances
t4.ap-southeast-1.Firehose.list_delivery_streams
t7.ap-southeast-1.Lightsail.get_instances
t0.ap-south-1.Lightsail.get_instances
t1.us-east-2.Lightsail.get_load_balancers
t7.ap-southeast-2.WorkSpaces.describe_workspaces
t2.eu-west-3.SageMaker.list_notebook_instances
t3.eu-west-2.SageMaker.list_notebook_instances

Finished in 46 seconds. Saving resources to output.json.

# collect S3 and EC2 global resources, as well as us-east-1 and us-east-2

$ AWS_PROFILE=<profile> aws_recon -s S3,EC2 -r global,us-east-1,us-east-2

# collect S3 and EC2 global resources, as well as us-east-1 and us-east-2

$ AWS_PROFILE=<profile> aws_recon --services S3,EC2 --regions global,us-east-1,us-east-2

# save output to S3 bucket

$ AWS_PROFILE=<profile> aws_recon \
  --services S3,EC2 \
  --regions global,us-east-1,us-east-2 \
  --verbose \
  --s3-bucket my-recon-bucket

# save output to S3 bucket with a home region other than us-east-1

$ AWS_PROFILE=<profile> aws_recon \
  --services S3,EC2 \
  --regions global,us-east-1,us-east-2 \
  --verbose \
  --s3-bucket my-recon-bucket:us-west-2

Example OpenCSPM formatted (NDJSON) output.

$ AWS_PROFILE=<profile> aws_recon -l \
  -s S3,EC2 \
  -r global,us-east-1,us-east-2 \
  -f custom

or

$ AWS_PROFILE=<profile> aws_recon -j \
  -s S3,EC2 \
  -r global,us-east-1,us-east-2 \
  -f custom > output.json

API exceptions related to permissions are silently ignored in most cases. These errors are usually due to one of these cases:

• using a role without sufficient permissions
• querying an account with SCPs in place that prevent usage of certain services
• trying to query a service that isn't enabled/available in your region/account

In verbose mode, you will see exception logs in the output:

t2.us-east-1.EC2.describe_subnets.0
t4.us-east-1.SSM.describe_instance_information.0
t6.us-east-1.SecurityHub.InvalidAccessException   <-----
t2.us-east-1.EC2.describe_addresses.0
t4.us-east-1.SSM.describe_parameters.0
t1.us-east-1.GuardDuty.list_detectors.0

Use the -q command line option to re-raise these exceptions so troubleshooting access issues is easier.

Traceback (most recent call last):
arn:aws:sts::1234567890:assumed-role/role/my-audit-role is not authorized to perform:
 codepipeline:GetPipeline on resource: arn:aws:codepipeline:us-west-2:1234567890:pipeline
 (Aws::CodePipeline::Errors::AccessDeniedException)

The exact API operation that triggered the exception is indicated on the last line of the stack trace. If you can't resolve the necessary access, you should exclude those services with -x or --not-services, or leave off the -q option so the collection can continue.

AWS Recon uses multiple threads to try to overcome some of the I/O challenges of performing many API calls to endpoints all over the world.

For global services like IAM, Shield, and Support, requests are not multi-threaded. The S3 module is multi-threaded since each bucket requires several additional calls to collect complete metadata.

For regional services, a thread (up to the thread limit) is spawned for each service in a region. By default, up to 8 threads will be used. If your account has resources spread across many regions, you may see a speed improvement by increasing threads with -t X, where X is the number of threads.

AWS Recon will make a minimum of ~2,000 API calls in a new/empty account, just to query the supported services in all 20 standard (non-GovCloud, non-China) regions. It is very likely to encounter API rate-limiting (throttling) on large accounts if you enable more threads than the default (8).

Recon will automatically backoff and respect the retry limits in the API response. If you observe long pauses during collection, this is likely what is happening. Retry collection with the -d or --debug option to observe the wire trace and see if you're being throttled. Consider using fewer threads or requesting higher rate limits from AWS if you are regularly getting rate-limited.

Most users will want to limit collection to relevant services and regions. Running without any exclusions will attempt to collect all resources from all regions enabled for the account.

$ aws_recon -h

AWS Recon - AWS Inventory Collector (0.5.17)

Usage: aws_recon [options]
    -r, --regions [REGIONS]          Regions to scan, separated by comma (default: all)
    -n, --not-regions [REGIONS]      Regions to skip, separated by comma (default: none)
    -s, --services [SERVICES]        Services to scan, separated by comma (default: all)
    -x, --not-services [SERVICES]    Services to skip, separated by comma (default: none)
    -c, --config [CONFIG]            Specify config file for services & regions (e.g. config.yaml)
    -b, --s3-bucket [BUCKET:REGION]  Write output file to S3 bucket (default: '')
    -o, --output [OUTPUT]            Specify output file (default: output.json)
    -f, --format [FORMAT]            Specify output format (default: aws)
    -t, --threads [THREADS]          Specify max threads (default: 8, max: 128)
    -l, --json-lines                 Output NDJSON/JSONL format (default: false)
    -u, --user-data                  Collect EC2 instance user data (default: false)
    -z, --skip-slow                  Skip slow operations (default: false)
    -g, --skip-credential-report     Skip generating IAM credential report (default: false)
    -j, --stream-output              Stream JSON lines to stdout (default: false)
    -v, --verbose                    Output client progress and current operation
    -q, --quit-on-exception          Stop collection if an API error is encountered (default: false)
    -d, --debug                      Output debug with wire trace info
    -h, --help                       Print this help information

Output is always some form of JSON - either JSON lines or plain JSON. The output is either written to a file (the default), or written to stdout (with -j).

When writing to an S3 bucket, the JSON output is automatically compressed with gzip.

If you have enabled manually enabled regions:

• me-south-1 - Middle East (Bahrain)
• af-south-1 - Africa (Cape Town)
• ap-east-1 - Asia Pacific (Hong Kong)
• eu-south-1 - Europe (Milan)

and you are using STS to assume a role into an account, you will need to enable v2 STS tokens in the account you are assuming the role from to be able to run AWS Recon against those regions.

Version 1 tokens are valid only in AWS Regions that are available by default. These tokens do not work in manually enabled Regions, such as Asia Pacific (Hong Kong). Version 2 tokens are valid in all Regions. However, version 2 tokens are longer and might affect systems where you temporarily store tokens.

If you are using a static access key/secret, you can collect from these regions regardless of STS token version.

Current "coverage" by service is listed below. The services without coverage will eventually be added. PRs are certainly welcome. :)

AWS Recon aims to collect all resources and metadata that are relevant in determining the security posture of your AWS account(s). However, it does not actually examine the resources for security posture - that is the job of other tools that take the output of AWS Recon as input.

• AccessAnalyzer
• AdvancedShield
• ApplicationAutoScaling
• Athena
• Backup
• GuardDuty
• Macie
• Systems Manager
• Trusted Advisor
• ACM
• API Gateway
• AutoScaling
• CodePipeline
• CodeBuild
• CloudFormation
• CloudFront
• CloudWatch
• CloudWatch Logs
• CloudTrail
• Config
• DirectoryService
• DirectConnect
• DMS
• DynamoDB
• EC2
• ECR
• ECRPublic
• ECS
• EFS
• EKS
• ELB
• EMR
• Elasticsearch
• ElastiCache
• Firehose
• FMS
• Glacier
• Glue
• IAM
• KMS
• Kafka
• Kinesis
• Lambda
• Lightsail
• Organizations
• RDS
• Redshift
• Route53
• Route53Domains
• S3
• SageMaker
• SES
• SecretsManager
• SecurityHub
• ServiceQuotas
• Shield
• SNS
• SQS
• Transfer
• VPC
• WAF
• WAFv2
• Workspaces
• Xray

One of the primary motivations for AWS Recon was to build a tool that is easy to maintain and extend. If you feel like coverage could be improved for a particular service, we would welcome PRs to that effect. Anyone with a moderate familiarity with Ruby will be able to mimic the pattern used by the existing collectors to query a specific service and add the results to the resource collection.

Clone this repository:

$ git clone git@github.com:darkbitio/aws-recon.git
$ cd aws-recon

Create a sticky gemset if using RVM:

$ rvm use 2.7.2@aws_recon_dev --create --ruby-version

Run bin/setup to install dependencies. Then, run rake test to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and tags, and push the .gem file to rubygems.org.

• Test coverage with AWS SDK stubbed resources

AWS Recon was inspired by the excellent work of the people and teams behind these tools:

• CloudMapper https://github.com/duo-labs/cloudmapper
• Prowler https://github.com/toniblyx/prowler
• CloudSploit https://github.com/cloudsploit/scans