kibana/x-pack/plugins/security
2020-12-02 23:40:06 +01:00
..
common Make all providers to preserve original URL when session expires. (#84229) 2020-12-02 11:32:22 +01:00
public Upgrade Node.js to version 14 (#83425) 2020-12-02 23:40:06 +01:00
server Migrate security routes to a new Elasticsearch client. (#84528) 2020-12-02 15:00:01 +01:00
jest.config.js Jest multi-project configuration (#77894) 2020-12-02 11:42:23 -08:00
kibana.json Removing circular dependency between spaces and security (#81891) 2020-11-19 13:41:13 -05:00
README.md ECS audit logging (#74640) 2020-10-16 20:40:38 +01:00

Kibana Security Plugin

See Configuring security in Kibana.

Audit logging

Example

const auditLogger = securitySetup.audit.asScoped(request);
auditLogger.log({
  message: 'User is updating dashboard [id=123]',
  event: {
    action: 'saved_object_update',
    category: EventCategory.DATABASE,
    type: EventType.CHANGE,
    outcome: EventOutcome.UNKNOWN,
  },
  kibana: {
    saved_object: { type: 'dashboard', id: '123' },
  },
});

What events should be logged?

The purpose of an audit log is to support compliance, accountability and security by capturing who performed an action, what action was performed and when it occurred. It is not the purpose of an audit log to aid with debugging the system or provide usage statistics.

Kibana guidelines:

Each API call to Kibana will result in a record in the audit log that captures general information about the request (http_request event).

In addition to that, any operation that is performed on a resource owned by Kibana (e.g. saved objects) and that falls in the following categories, should be included in the audit log:

  • System access (incl. failed attempts due to authentication errors)
  • Data reads (incl. failed attempts due to authorisation errors)
  • Data writes (incl. failed attempts due to authorisation errors)

If Kibana does not own the resource (e.g. when running queries against user indices), then auditing responsibilities are deferred to Elasticsearch and no additional events will be logged.

Examples:

For a list of audit events that Kibana currently logs see: docs/user/security/audit-logging.asciidoc

When should an event be logged?

Due to the asynchronous nature of most operations in Kibana, there is an inherent tradeoff between the following logging approaches:

  • Logging the intention before performing an operation, leading to false positives if the operation fails downstream.
  • Logging the outcome after completing an operation, leading to missing records if Kibana crashes before the response is received.
  • Logging both, intention and outcome, leading to unnecessary duplication and noisy/difficult to analyse logs.

Kibana guidelines:

  • Write operations should be logged immediately after all authorisation checks have passed, but before the response is received (logging the intention). This ensures that a record of every operation is persisted even in case of an unexpected error.
  • Read operations, on the other hand, should be logged after the operation completed (logging the outcome) since we won't know what resources were accessed before receiving the response.
  • Be explicit about the timing and outcome of an action in your messaging. (e.g. "User has logged in" vs. "User is creating dashboard")

Can an action trigger multiple events?

  • A request to Kibana can perform a combination of different operations, each of which should be captured as separate events.
  • Operations that are performed on multiple resources (bulk operations) should be logged as separate events, one for each resource.
  • Actions that kick off background tasks should be logged as separate events, one for creating the task and another one for executing it.
  • Internal checks, which have been carried out in order to perform an operation, or errors that occured as a result of an operation should be logged as an outcome of the operation itself, using the ECS event.outcome and error fields, instead of logging a separate event.
  • Multiple events that were part of the same request can be correlated in the audit log using the ECS trace.id property.