S3-antivirus (code named: Salve)

ClamAV for your S3.

S3-antivirus runs a small node process that polls SQS for S3 notifications. It then fetches the S3 object and scans it, reporting the results on SNS topics.

getting started (development)

Install the dependencies.

# OS X
$> brew update && brew install clamav

# Ubuntu
$> apt-get install libclamav-dev clamav-freshclam

Install the virus database

$> freshclam

Then clone the repo and install the other dependencies.

$> git clone https://github.com/execonline-inc/antivirus.git
$> cd antivirus
$> npm install

The AWS configuration uses environment variables. The following variables will need to be set.

AWS_ACCESS_KEY_ID      # AWS access key
AWS_SECRET_ACCESS_KEY  # AWS secret key... shhhh!
AWS_REGION             # AWS region (us-east-1, for example)
AWS_AV_QUEUE           # SQS queue subscribed to S3 notifications (a url)
AWS_AV_CLEAN_TOPIC     # SNS topic arn. For clean file notifications.
AWS_AV_INFECTED_TOPIC  # SNS topic arn. For infected file notifications.

Then run the script.

$> node index.js

the file on it's journey...

Here's a diagram of how S3-antivirus fits into your infrastructure.

When S3 adds an object it sends out a notification event. We push these notifications to an SNS topic. The benefit of SNS is that we can have many services subscribed to the same topic.

SNS topics offer no guarantees of delivery. They are fire and forget. If the virus scanner is down when an event arrives, then that S3 object isn't scanned. This just won't do.

Instead of subscribing to the SNS topic, we create an SQS queue. We subscribe the queue to the S3 events topic. SQS queues offer message persistence (up to four days, by default). They also have configurable dead letter fail over. Our virus scanner long polls the SQS queue. This offers us increased confidence that we will not miss scanning a file.

When the scanner receives an S3 event from the queue, it downloads the object from S3 and scans it.

If a scanned file is clean (no virus detected), we push a notification out on a "clean" SNS topic. If we detect a virus, we push a notification out on an "infected" SNS topic.

The infected topic and the clean topic don't have to be different topics in practice. We've separated them so that it is easier to configure the virus scanner to suit your needs.

Anything that can subscribe to SNS topics can process infected (or clean!) files.

deployment stories

S3-antivirus is just node, so deploy it anyway you would deploy node applications.

For convenience, we've included a Dockerfile. Docker images provide a convenient deployment artifact. You can build and test your environment on your computer. Then deploy that exact environment. This works on any platform that supports docker images.

To use the Dockerfile, you first need to install Docker

Build the docker image:

$> docker build -t salve .

You can run the docker build as a container.

$> docker run --env-file some/path/to/salve/environment/file salve

Note: Running S3-antivirus this way requires an env file. The env-file format is this:

AWS_ACCESS_KEY_ID=<AWS access key>
AWS_SECRET_ACCESS_KEY=<AWS secret key... shhhh!>
AWS_REGION=us-east-1
AWS_AV_QUEUE=https://some-sqs-queue/url
AWS_AV_CLEAN_TOPIC=some-topic-arn
AWS_AV_INFECTED_TOPIC=another-or-possibly-the-same-topic-arn

S3-antivirus was first deployed using Elastic Beanstalk. Elastic Beanstalk is a low management computing environment offered by AWS. We've included a command to build a package suitable for deploying in this environment.

$> npm run package
# generates `deploy.zip`

deployment checklist

• Create SNS topics (first time)
  • S3 events
  • Infected files
  • Clean files
• Create SQS queue and subscribe to S3 events topic (first time)
• Upload deploy.zip to Elastic Beanstalk
  • Configure environment variables (first time only)

You configure your environment once. Then deploy updates by uploading an new deploy.zip.

Elastic Beanstalk is already connected to CloudWatch. You can set whatever alerts are appropriate. ElasticBeanstalk autoscales. Changing the configuration will support a larger volume of file scans.

a word about coding paradigms

You'll note two key design decisions in this code:

• The use of a declarative, functional style. We used the excellent Ramda library.
• All asynchronous code uses Promises.

The functional style allowed us to compose the necessary behaviors from primitives provided by Ramda. This left us writing little code ourselves.

Promises allowed us to compose and sequence asynchronous operations without falling into nested callbacks.