Distributed Cuckoo

As mentioned in Submit an Analysis, Cuckoo provides a REST API for Distributed Cuckoo usage. Distributed Cuckoo allows one to setup a single REST API point to which samples and URLs can be submitted which will then, in turn, be submitted to one of the configured Cuckoo nodes.

A typical setup thus includes a machine on which Distributed Cuckoo is run and one or more machines running an instance of the Cuckoo daemon and the Cuckoo REST API.

A few notes:

  • Using Distributed Cuckoo only makes sense when running at least two cuckoo nodes.
  • Distributed Cuckoo can be run on a machine that also runs a Cuckoo daemon and REST API, however, make sure it has enough disk space if the intention is to submit a lot of samples.

Starting the Distributed REST API

The Distributed REST API has the following command line options:

$ cuckoo distributed server --help
Usage: cuckoo distributed server [OPTIONS]

Options:
  -H, --host TEXT     Host to bind the Distributed Cuckoo server on
  -p, --port INTEGER  Port to bind the Distributed Cuckoo server on
  --uwsgi             Dump uWSGI configuration
  --nginx             Dump nginx configuration
  --help              Show this message and exit.

As may be derived from the help output, starting Distributed Cuckoo may be as simple as running cuckoo distributed server.

The various configuration options are described in the configuration file, but following we have more in-depth descriptions as well. More advanced usage naturally includes deployment using uWSGI and nginx.

Distributed Cuckoo Configuration

Report Formats

The reporting formats denote which reports you’d like to retrieve later on. Note that all task-related data will be removed from the Cuckoo nodes once the related reports have been fetched so that the machines are not running out of disk space. This does, however, force you to specify all the report formats that you’re interested in, because otherwise that information will be lost.

Reporting formats include, but are not limited to and may also include your own reporting formats, report.json, report.html, etc.

Samples Directory

The samples directory denotes the directory where the submitted samples will be stored temporarily, until the associated task has been deleted.

Reports Directory

Much like the Samples Directory the Reports Directory defines the directory where reports will be stored until they’re fetched and deleted from the Distributed REST API.

RESTful resources

Following are all RESTful resources. Also make sure to check out the Quick usage section which documents the most commonly used commands.

Resource Description
GET GET /api/node Get a list of all enabled Cuckoo nodes.
POST POST /api/node Register a new Cuckoo node.
GET GET /api/node/<name> Get basic information about a node.
PUT PUT /api/node/<name> Update basic information of a node.
POST POST /api/node/<name>/refresh Refresh a Cuckoo nodes metadata.
DELETE DELETE /api/node/<name> Disable (not completely remove!) a node.
GET GET /api/task Get a list of all (or a part) of the tasks in the database.
POST POST /api/task Create a new analysis task.
GET GET /api/task/<id> Get basic information about a task.
DELETE DELETE /api/task/<id> Delete all associated information of a task.
GET GET /api/report/<id>/<format> Fetch an analysis report.
GET GET /api/pcap/<id> Fetches the PCAP of an analysis.

GET /api/node

Returns all enabled nodes. For each node the information includes the associated name, its API URL, and machines:

$ curl http://localhost:9003/api/node
{
    "success": true,
    "nodes": {
        "localhost": {
            "machines": [
                {
                    "name": "cuckoo1",
                    "platform": "windows",
                    "tags": []
                }
            ],
            "name": "localhost",
            "url": "http://localhost:8090/"
        }
    }
}

POST /api/node

Register a new Cuckoo node by providing the name and the URL:

$ curl http://localhost:9003/api/node -F name=localhost \
    -F url=http://localhost:8090/
{
    "success": true
}

GET /api/node/<name>

Get basic information about a particular Cuckoo node:

$ curl http://localhost:9003/api/node/localhost
{
    "success": true,
    "nodes": [
        {
            "name": "localhost",
            "url": "http://localhost:8090/"
            "machines": [
                {
                    "name": "cuckoo1",
                    "platform": "windows",
                    "tags": []
                }
            ]
        }
    ]
}

PUT /api/node/<name>

Update basic information of a Cuckoo node:

$ curl -XPUT http://localhost:9003/api/node/localhost -F name=newhost \
    -F url=http://1.2.3.4:8090/
{
    "success": true
}

POST /api/node/<name>/refresh

Refreshes metadata associated by a Cuckoo node, in particular, its machines:

$ curl -XPOST http://localhost:9003/api/node/localhost/refresh
{
    "success": true,
    "machines": [
        {
            "name": "cuckoo1",
            "platform": "windows",
            "tags": []
        },
        {
            "name": "cuckoo2",
            "platform": "windows",
            "tags": []
        }
    ]
}

DELETE /api/node/<name>

Disable a Cuckoo node, therefore not having it process any new tasks, but keeping its history in the Distributed Cuckoo database:

$ curl -XDELETE http://localhost:9003/api/node/localhost
{
    "success": true
}

GET /api/task

Get a list of all tasks in the database. In order to limit the amount of results, there’s an offset, limit, finished, and owner field available:

$ curl http://localhost:9003/api/task?limit=1
{
    "success": true,
    "tasks": {
        "1": {
            "clock": null,
            "custom": null,
            "owner": "",
            "enforce_timeout": null,
            "machine": null,
            "memory": null,
            "options": null,
            "package": null,
            "path": "/tmp/dist-samples/tmphal8mS",
            "platform": "windows",
            "priority": 1,
            "tags": null,
            "task_id": 1,
            "timeout": null
        }
    }
}

POST /api/task

Submit a new file or URL to be analyzed:

$ curl http://localhost:9003/api/task -F file=@sample.exe
{
    "success": true,
    "task_id": 2
}

GET /api/task/<id>

Get basic information about a particular task:

$ curl http://localhost:9003/api/task/2
{
    "success": true,
    "tasks": {
        "2": {
            "id": 2,
            "clock": null,
            "custom": null,
            "owner": "",
            "enforce_timeout": null,
            "machine": null,
            "memory": null,
            "options": null,
            "package": null,
            "path": "/tmp/tmpPwUeXm",
            "platform": "windows",
            "priority": 1,
            "tags": null,
            "timeout": null,
            "task_id": 1,
            "node_id": 2,
            "finished": false
        }
    }
}

DELETE /api/task/<id>

Delete all associated data of a task, namely the binary, the PCAP, and the reports:

$ curl -XDELETE http://localhost:9003/api/task/2
{
    "success": true
}

GET /api/report/<id>/<format>

Fetch a report for the given task in the specified format:

# Defaults to the JSON report.
$ curl http://localhost:9003/api/report/2
...

GET /api/pcap/<id>

Fetches the PCAP for the given task:

$ curl http://localhost:9003/api/pcap/2
...

Proposed setup

The following description depicts a Distributed Cuckoo setup with two Cuckoo machines, cuckoo0 and cuckoo1. In this setup the first machine, cuckoo0, also hosts the Distributed Cuckoo REST API.

Configuration settings

Our setup will require a couple of updates with regards to the configuration files.

conf/cuckoo.conf

Update process_results to off as we will be running our own results processing script (for performance reasons).

Update tmppath to something that holds enough storage to store a few hundred binaries. On some servers or setups /tmp may have a limited amount of space and thus this wouldn’t suffice.

Update connection to use something not sqlite3. Preferably PostgreSQL or MySQL. SQLite3 doesn’t support multi-threaded applications and as such is not a good choice for systems such as Cuckoo (as-is).

You should create a database specifically for the distributed cuckoo setup. Do not be tempted to use any existing cuckoo database in order to avoid update problems with the DB scripts. In the configuration use the new database name. The remaining configuration such as usernames, servers, etc can be the same as for your cuckoo install. Don’t forget to use one DB per node and one for the machine running Distributed Cuckoo (the “management machine” or “controller”).

conf/processing.conf

You may want to disable some processing modules, such as virustotal.

conf/reporting.conf

Depending on which report(s) are required for integration with your system it might make sense to only make those report(s) that you’re going to use. Thus disabling the other ones.

conf/virtualbox.conf

Assuming VirtualBox is the Virtual Machine manager of choice, the mode will have to be changed to headless or you will have some restless nights (this is the default nowadays).

Setup Cuckoo

On each machine you will have to run the Cuckoo Daemon, the Cuckoo API, and one or more Cuckoo Process instances. For more information on setting that up, please refer to Starting Cuckoo.

Setup Distributed Cuckoo

On the Distributed Cuckoo machine you’ll have to setup the Distributed Cuckoo REST API and the Distributed Cuckoo Worker.

As stated earlier, Distributed Cuckoo REST API may be started by running cuckoo distributed server or by deploying it properly with uWSGI and nginx.

The Distributed Cuckoo Worker may be started by running supervisorctl start distributed in the CWD (make sure to start supervisord first as per Cuckoo in the background). This will automatically start the Worker with the correct configuration and arguments, etc.

Register Cuckoo nodes

As outlined in Quick usage the Cuckoo nodes have to be registered with the Distributed Cuckoo REST API:

$ curl http://localhost:9003/api/node -F name=cuckoo0 -F url=http://localhost:8090/
$ curl http://localhost:9003/api/node -F name=cuckoo1 -F url=http://1.2.3.4:8090/

Having registered the Cuckoo nodes all that’s left to do now is to submit tasks and fetch reports once finished. Documentation on these commands can be found in the Quick usage section. In case your Cuckoo node is not on localhost, replace localhost with the IP address of the node where the Cuckoo REST API is running.

If you want to experiment with load balancing between the nodes you may want to try using a lower value for the threshold parameter in the $CWD/distributed/settings.py file as the default value is 500 (meaning tasks are assigned to Cuckoo nodes in batches of 500).

Quick usage

For practical usage the following few commands will be most interesting.

Register a Cuckoo node, in this case a Cuckoo API running on the same machine in this case:

$ curl http://localhost:9003/api/node -F name=localhost -F ip=127.0.0.1

Disable a Cuckoo node:

$ curl -XDELETE http://localhost:9003/api/node/localhost

Submit a new analysis task without any special requirements (e.g., using Cuckoo tags, a particular machine, etc):

$ curl http://localhost:9003/api/task -F file=@/path/to/sample.exe

Get the report of a task has been finished (if it hasn’t finished you’ll get an error with code 420). Following example will default to the JSON report:

$ curl http://localhost:9003/api/report/1

If a Cuckoo node gets stuck and needs a reset, the following steps could be performed to restart it cleanly. Note that this requires usage of our SaltStack configuration and some manual SQL commands (and preferably the Distributed Cuckoo Worker is temporary disabled, i.e., supervisorctl stop distributed):

$ psql -c "UPDATE task SET status = 'pending' WHERE status = 'processing' AND node_id = 123"
$ salt cuckoo1 state.apply cuckoo.clean
$ salt cuckoo1 state.apply cuckoo.start

If the entire Cuckoo cluster was somehow locked up, i.e., all tasks have been ‘assigned’, are ‘processing’, or have the ‘finished’ status while none of the Cuckoo nodes are currently working on said analyses (e.g., due to numerous resets etc), then the following steps may be used to reset the entire state:

$ supervisorctl -c ~/.cuckoo/supervisord.conf stop distributed
$ salt '*' state.apply cuckoo.stop
$ salt '*' state.apply cuckoo.clean
$ psql -c "UPDATE task SET status = 'pending', node_id = null WHERE status IN ('assigned', 'processing', 'finished')"
$ salt '*' state.apply cuckoo.start
$ supervisorctl -c ~/.cuckoo/supervisord.conf start distributed

If a Cuckoo node has a number of tasks that failed to process, therefore locking up the Cuckoo node altogether, then upgrading the Cuckoo instances with a bugfixed version and re-processing all analyses may do the trick:

$ salt cuckoo1 state.apply cuckoo.update  # Upgrade Cuckoo.
# To make sure there are failed analyses in the first place.
$ salt cuckoo1 cmd.run "sudo -u cuckoo psql -c \"SELECT * FROM tasks WHERE status = 'failed_processing'\"
# Reset each analyses to be re-processed.
$ salt cuckoo1 cmd.run "sudo -u cuckoo psql -c \"UPDATE tasks SET status = 'completed', processing = null WHERE status = 'failed_processing'\""

In order to upgrade the Distributed Cuckoo master, one may want to perform the following steps:

$ /etc/init.d/uwsgi stop
$ supervisorctl -c ~/.cuckoo/supervisord.conf stop distributed
$ pip uninstall -y cuckoo
$ pip install cuckoo==2.0.0         # Specify your version here.
$ pip install Cuckoo-2.0.0.tar.gz   # Or use a locally archived build.
$ cuckoo distributed migrate
$ supervisorctl -c ~/.cuckoo/supervisord.conf start distributed
$ /etc/init.d/uwsgi start
$ /etc/init.d/nginx restart

In order to test your entire Cuckoo cluster, i.e., every machine on every Cuckoo node, one may take the stuff/distributed/cluster-test.py script as an example. As-is it allows one to check for an active internet connection in each and every configured machine in the cluster. This script may be used to identify machines that are incorrect or have been corrupted in one way or another. Example usage may look as follows:

# Assuming Distributed Cuckoo listens on localhost and that you want to
# run the 'internet' script (see also the source of cluster-test.py).
$ python stuff/distributed/cluster-test.py localhost -s internet