Cisco Threat Intel API

A Pragmatic, Operationalized Threat Intel Service and Data Model

For full documentation see doc/

We also think the Use Cases document is a good starting point.

Interactive, Swagger docs for the API are available once you have it running, at:

http://localhost:3000/index.html

Goals

Sharing actionable threat intel
Simple, pragmatic data model
Ease of integration and exploration
Extremely fast Verdict lookups
Hypertextual integration with other services

This is not a full STIX/TAXII service. Its intent is to help Analysts know what is important, and for detection and prevention tools to know what to look for.

In addition to the RESTful HTTP API, it also has a GraphQL API and many event handlers.

The data model is defined in the CTIM project, although it's quite easy to see the API and the models it handles using the built-in Swagger UI once you have it running.

CTIA is implemented in Clojure

Usage

Data Stores and External Dependencies

CTIA uses Leiningen as its "build" tool, you can install it by following the instructions here: http://leiningen.org/#install

By default, CTIA uses Elasticsearch 5.x as its data store. Assuming you have it running on 127.0.0.1:9200 you can simply start CTIA.

You can jump to the Development section to see instructions on how to run elasticsearch and other optional supporting tools using Docker. CTIA may use Kafka, Redis and ES to push events.

Purging ES Stores

Using an uberjar build you can purge all the ES Stores with this command:

java -cp ctia.jar:resources:. clojure.main -m ctia.task.purge-es-stores

Using lein use this one:

lein run -m ctia.task.purge-es-stores

Run the application locally

Running from a cloned repository:

lein run -m ctia.main

Packaging and running as standalone jar

This is the proper way to run this in production.

lein do clean, uberjar
java -Xmx4g -Djava.awt.headless=true -Dlog.console.threshold=INFO -jar target/ctia.jar

You may tweak the java arguments as per your requirements.

Development

The easiest way to get running is to use docker.

On Mac OS X, you should use Docker for Mac which includes all the dependencies you need to run Docker containers.

With Kafka and Zookeeper now part of the dev cluster, you will need to increase the memory you allocate to Docker. You can do this thru your Docker preferences. This has been tested with a 4GB allocation.

We provide a default containers/dev/docker-compose.yml which will bring up the dependencies you need in containers.

You can bring up a development environment:

docker-compose -f containers/dev/docker-compose.yml up
# for M1 macs
docker-compose -f containers/dev/m1-docker-compose.yml up

Using docker for mac, this will bind the following ports on your development machine to the services running in the containers:

Redis - 6379
elasticsearch - 9200 and 9300
kibana - 5601
zookeeper - 2181
kafka - 9092
riemann - 5555-5557
riemann-dash - 4567

If you ever need to reset your entire dev environment, just kill the docker-compose process and run:

docker-compose -f containers/dev/docker-compose.yml down
docker-compose -f containers/dev/docker-compose.yml up --force-recreate --remove-orphans

In particular, this resets ElasticSearch indices, which cannot be created more than once.

Local workflow

To start CTIA locally, use ./scripts/run for ES5 and ./scripts/run7 for ES7.

To run tests, use ./scripts/test for all non-integration tests, and ./scripts/test7 for just ES7 non-integration tests.

For a REPL workflow, run lein repl. Use (start) to start CTIA, (stop) to stop it, and (go) to restart it.

Testing and CI

All PRs must pass lein test with no fails for PRs to be accepted. Any new code functionality/change should have tests accompanying it.

PRs are built and deployed using GitHub Actions.

To skip CI on a PR commit, use [skip ci] in the commit message. Due to branch protection rules, once you are ready to merge your PR, a full build must be completed. When you choose the final commit message for your squashed PR, please be aware that Travis also supports skipping CI on certain commit messages--in other words, certain commit messages may skip deployment.

To run the more rigorous cron job (multi-JVM/Clojure matrix) on a PR, start commit messages with {:test-suite :cron}.

Test balancing

Continuous integration test runs are parallelized using prior timing information in dev-resources/ctia_test_timings.edn.

For GitHub Actions Pull Requests, tests are automatically balanced by accessing the cache of previous builds.

To experiment with this feature locally, you need the timings in your local repository. There are 2 ways to generate them, described in the below subsections:

locally (slow)
by downloading a GitHub Actions artifact for a recent build (fast).

Rebalance tests locally (slow)

Run ./build/run-tests.sh to get the latest test timings.
Run ./scripts/summarize-tests.clj to collect timings
Run cp target/test-results/all-test-timings.edn dev-resources/ctia_test_timings.edn to update the latest timings.
Commit this change and push.

Rebalance tests via GitHub Actions (fast)

You will need a completed GitHub Actions Pull Request build to base it on.

Visit the GitHub Actions Pull Request build you would like to use to rebalance
- eg., https://github.com/threatgrid/ctia/actions/runs/342562872
Go to the all-pr-checks job (should be the last job) and download the all-test-timings artifact
Unzip the downloaded file and locate the inflated all-test-timings.edn file.
Copy the inflated all-test-timings.edn file to dev-resources/ctia_test_timings.edn
Commit this change and push.

Note: Unfortunately, even public artifacts require an API key to download them via the API, so this is not straightforward to automate (see more).

Data Access Control

Document Access control is defined at the document level, rules are defined using TLP combined with the max-record-visibility property (Traffic Light Protocol) by default:

Everyone Max Record visibility

Green/White TLP

Identity	Read	Write
Owner	✔	✔
Group/Org	✔	✔
Others	✔

Amber TLP

Identity	Read	Write
Owner	✔	✔
Group/Org	✔	✔
Others

Red TLP

Identity	Read	Write
Owner	✔	✔
Group/Org
Others

Group Max Record visibility

Green/White TLP

Identity	Read	Write
Owner	✔	✔
Group/Org	✔	✔
Others

Amber TLP

Identity	Read	Write
Owner	✔	✔
Group/Org	✔	✔
Others

Red TLP

Identity	Read	Write
Owner	✔	✔
Group/Org
Others

Custom Access Rules

it is possible to grant additional access to any user/group using either authorized_users or authorized_groups document fields, when an identity is marked in one of these fields, it gets full R/W access to the documents.

Please note that the authorized_groups property may work only if max record visibility is set to everyone

Examples:

The following actor Entity is marked as Red, thus allowing only its owner RW access, since "foo" and "bar" are marked as authorized_users the owners of those identites also have RW access.

  {"id": "actor-5023697b-3857-4652-9b53-ccda297f9c3e",
   "type": "actor",
   "schema_version": "0.4.2",
   "actor_type": "Hacker",
   "confidence": "High",
   "source": "a source",
   "tlp": "red",
   "valid_time": {},
   "authorized_users": ["foo" "bar"]}

The following actor Entity is marked as Amber, thus allowing only its owner or group RW access, since "foogroup" and "bargroup" are marked as authorized_groups identities in these groups also get full RW access.

  {"id": "actor-5023697b-3857-4652-9b53-ccda297f9c3e",
   "type": "actor",
   "schema_version": "0.4.2",
   "actor_type": "Hacker",
   "confidence": "High",
   "source": "a source",
   "tlp": "red",
   "valid_time": {},
   "authorized_groups": ["foogroup" "bargroup"]}

Bulk and Bundle

CTIA provides Bulk and Bundle routes to help processing multiple entities. see Bulk and Bundle documentation

Bundle import

The /bundle API endpoint allows users with the correct permissions to POST a CTIM bundle object.

The ability to post bundles is controlled by the import-bundle capability.

When a bundle is submitted:

All entities that have already been imported with the external ID whose prefix has been configured with the ctia.store.external-key-prefixes property are searched.
If they are identified by transient IDs, a mapping table between transient and stored IDs is built.
Only new entities are created in the same way as the /bulk API endpoint with transient IDs resolutions. Existing entities are not modified.

If more than one entity is referenced by the same external ID, the first one which is retrieved is picked.

Response of the bundle API endpoint:

{:results [{:id "http://example.com/ctia/entity-type/entity-type-991d8dfb-b54e-4435-ac58-2297b4d886c1"
            :tempid "transient:1f48f48c-4130-47f1-92dc-a6df8ab110b6"
            :action "create"
            :external_id "indicator-abuse-ch-077d653844d95d8cd8e4e51cb1f9215feae50426"
            :error "An error occurs"}]

Field	Description
`:id`	The real ID
`:original_id`	Provided ID if different from real ID (ex: transient ID)
`:result`	`error`, `created` or `exists`
`:external_id`	External ID used to identify the entity
`:error`	Error message

Feeds

CTIA allows generating Feeds as public urls meant to easily exchange data through views skipping authentication. You may use this functionality to generate blocklists easily consumable by simple systems.

The first kind of Feed is the Indicator one, you may create it posting a Feed document specifying an indicator_id and an output type.

CTIA will then return the realized Feed document including two new fields: feed_view_url and feed_view_url_csv

both of those urls will be publicly available without authentication so they must be shared carefully.
both urls return the Judgements associated with the provided indicator_id through their relationships
depending on the selected output it will either extract and return the observables only or the full Judgements
the CSV view output either the full Judgement as CSV or the Observable values only depending on the output.

Elasticsearch Store management

see CTIA Elasticsearch Stores: managing big Indices

see Migration procedure

see CTIA Elasticsearch CRUD details

Store Checks

There is a dedicated task to check all stores of a configured CTIA instance. this task will run through all configured stores and validate each document in bulk.

Launch the task with:

java -cp ctia.jar:resources:. clojure.main -m ctia.task.check-es-stores <batch-size>

or from source with leiningen:

lein run -m ctia.task.check-es-stores <batch-size>

Task arguments

argument	description	example
batch-size	how many documents to validate at once	1000

API

List Pagination

HTTP routes providing a list use a default limit of 100 records. An API client can change this parameter up to 10 0000 records.

when a limit is applied to the response, pagination headers are returned:

header	description	example
X-TOTAL	total number of hits in the data store	5000
X-OFFSET	the current pagination offset	200
X-NEXT	ready made parameters to fetch the next results page	limit=100&offset=100&search_after=foo
X-PREVIOUS	ready made parameters to fetch the previous results page	limit=100&offset=0
X-SORT	the sort parameter for use with `search_after`, the id of the last result page	["actor-77b01a42-6d2b-4081-8fd0-c887bf54140c"]

To easily scroll through all results of a list, just iterate, appending X-Next to your base query URL. if no X-Next header is present, you have reached the last page.

Offset Pagination

To be used for simple matters, when the result window is inferior to 10 000 (offset + limit) use a combination of offset and limit parameters to paginate results.

Stateless Cursor Pagination

To be used when the result window is superior to 10 000, allows to easily loop across all pages of a query response. use limit and offset along with search_after filled with the value from the X-Sort response header to get the next page.

Rate limit

Requests may be rate limited by enabling the middleware using the ctia.http.rate-limit.enabled property.

It rate limits how many HTTP requests a CTIA group can make in an hour. The group is identified with the property :identity of the current Ring request.

Before the rate limit is reached, the header X-Ratelimit-Group-Limit is returned in the response:

HTTP/1.1 200 OK
Content-Type: application/json;charset=utf-8
Date: Wed, 31 Oct 2018 14:05:30 GMT
Server: Jetty(9.4.z-SNAPSHOT)
Strict-Transport-Security: max-age=31536000; includeSubdomains
Vary: Accept-Encoding, User-Agent
X-Ctim-Version: 1.0.6
X-Ctia-Config: b9b3477528d9616ed85221f2827bf1da443e8f00
X-Ctia-Version: 70323eb3b72da558e7f056e418533402f65d335a
X-Ratelimit-Group-Limit: 8000

If the rate limit is exceeded:

The client receives a response with the 429 HTTP status, a retry-after header and the JSON message {"error": "Too Many Requests"}. The retry-after header indicates the number of seconds to wait before making a new request.

HTTP/1.1 429 Too Many Requests
Content-Length: 30
Content-Type: application/json
Date: Wed, 31 Oct 2018 14:05:30 GMT
Retry-After: 3557
Server: Jetty(9.4.z-SNAPSHOT)
Strict-Transport-Security: max-age=31536000; includeSubdomains
X-Ctim-Version: 1.0.6
X-Ctia-Config: b9b3477528d9616ed85221f2827bf1da443e8f00
X-Ctia-Version: 70323eb3b72da558e7f056e418533402f65d335a

A message is logged with the :info level

Update index state task

There is a dedicated task for updating ES index state on startup. It performs idempotent changes to ES mappings and settings that are needed before CTIA can start.

In theory, this task could be performed automatically when CTIA starts, but in the case where CTIA is load-balanced across several instances this causes performance problems as all instances will update ES simulateneously. It has been separated out for this reason.

It must be successfully completed before starting CTIA via:

java -cp ctia.jar:resources:. clojure.main -m ctia.task.update-index-state

or from source with leiningen:

lein run -m ctia.task.update-index-state

License

Eclipse Public License v1.0

Data Model

The data model of CTIA is closely based on STIX, with a few simplifications. See Cisco Threat Intel Model for details.

yogsototh/ctia

Cisco Threat Intel API

Goals

Usage

Data Stores and External Dependencies

Purging ES Stores

Run the application locally

Packaging and running as standalone jar

Development

Local workflow

Testing and CI

Test balancing

Rebalance tests locally (slow)

Rebalance tests via GitHub Actions (fast)

Data Access Control

Everyone Max Record visibility

Green/White TLP

Amber TLP

Red TLP

Group Max Record visibility

Green/White TLP

Amber TLP

Red TLP

Custom Access Rules

Bulk and Bundle

Bundle import

Feeds

Elasticsearch Store management

Store Checks

Task arguments

API

List Pagination

Offset Pagination

Stateless Cursor Pagination

Rate limit

Update index state task

License

Data Model