debian-mirror-gitlab/doc/development/sec/security_report_ingestion_overview.md
2023-06-20 00:43:36 +05:30

8.6 KiB

stage group info type
Secure Threat Insights To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#assignments concepts

Security report ingestion overview

WARNING: The Vulnerability::Feedback model is currently undergoing deprecation and should be actively avoided in all further development. It is currently maintained with feature parity to enable revert should any issues arise, but is intended to be removed in 16.0. Any interactions relating to the Feedback model are superseded by the StateTransition, IssueLink, and MergeRequestLink models. You can find out more on in this epic.

Commonly used terms

Feedback

An instance of Vulnerabilities::Feedback class. They are created to keep track of users' interactions with Vulnerability Findings before they are promoted to a Vulnerability. This model is deprecated and due to be removed by GitLab 16.0 as part of the Deprecate and remove Vulnerabilities::Feedback epic.

An instance of Vulnerabilities::IssueLink class. They are used to link Vulnerability records to Issue records.

An instance of Vulnerabilities::MergeRequestLink class. They are used to link Vulnerability records to MergeRequest records.

Security Finding

An instance of Security::Finding class. These serve as a meta-data store of a specific vulnerability detected in a specific Security::Scan. They currently store partial finding data to improve performance of the pipeline security report. This class has been extended to store almost all required scan information so we can stop relying on job artifacts and is due to be used in favor of Vulnerability::Findings soon.

Security Scan

An instance of the Security::Scan class. Security scans are representative of a Ci::Build which output a Job Artifact which has been output as a security scan result, which GitLab acknowledges and ingests the findings of as Security::Finding records.

State Transition

An instance of the Vulnerabilities::StateTransition class. This model represents a state change of a respecitve Vulnerability record, for example the dismissal of a vulnerability which has been determined to be safe.

Vulnerability

An instance of Vulnerability class. A Vulnerability is representative of a Vulnerability::Finding which has been detected in the default branch of the project, or if the present_on_default_branch flag is false, is representative of a finding which has been interacted with in some way outside of the default branch, such as if it is dismissed (State Transition), or linked to an Issue or Merge Request. They are created based on information available in Vulnerabilities::Finding class. Every Vulnerability must have a corresponding Vulnerabilities::Finding object to be valid, however this is not enforced at the database level.

Finding

An instance of Vulnerabilities::Finding class. A Vulnerability::Finding is a database only representation of a security finding which has been merged into the default branch of a project, as the same Vulnerability may be present in multiple places within a project. This class was previously called Vulnerabilities::Occurrence; after renaming the class, we kept the associated table name vulnerability_occurrences due to the effort involved in renaming large tables.

Identifier

An instance of the Vulnerabilities::Identifier class. Each vulnerability is given a unique identifier that can be derived from it's finding, enabling multiple Findings of the same Vulnerability to be correlated accordingly.

Vulnerability Read

An instance of the Vulnerabilities::Read class. This is a denormalised record of Vulnerability and Vulnerability::Finding data to improve performance of filtered querying of vulnerability data to the front end.

Remediation

An instance of the Vulnerabilities::Remediation class. A remediation is representative of a known solution to a detected Vulnerability. These enable GitLab to recommend a change to resolve a specific Vulnerability.

Vulnerability creation from Security Reports

Assumptions:

  • Project uses GitLab CI
  • Project uses security scanning tools
  • No Vulnerabilities are present in the database
  • All pipelines perform security scans

Scan runs in a pipeline for a non-default branch

  1. Code is pushed to the branch.
  2. GitLab CI runs a new pipeline for that branch.
  3. Pipeline status transitions to any of ::Ci::Pipeline.completed_statuses.
  4. Security::StoreScansWorker is called and it schedules Security::StoreScansService.
  5. Security::StoreScansService calls Security::StoreGroupedScansService and schedules ScanSecurityReportSecretsWorker.
  6. Security::StoreGroupedScansService calls Security::StoreScanService.
  7. Security::StoreScanService calls Security::StoreFindingsService.
  8. ScanSecurityReportSecretsWorker calls Security::TokenRevocationService to automatically revoke any leaked keys that were detected.
  9. At this point we only have Security::Finding records as these findings are not present in the default branch of the project.

At this point, the following things can happen to the Security::Finding which would result in its promotion to a Vulnerability::Finding with a respective Vulnerability record:

Scan runs in a pipeline for the default branch

If the pipeline ran on the default branch then the following steps, in addition to the steps in Scan runs in a pipeline for a non-default branch, are executed:

  1. Security::StoreScansService gets called and schedules StoreSecurityReportsWorker.
  2. StoreSecurityReportsWorker executes Security::Ingestion::IngestReportsService.
  3. Security::Ingestion::IngestReportsService takes all reports from a given Pipeline and calls Security::Ingestion::IngestReportService and then calls Security::Ingestion::MarkAsResolvedService.
  4. Security::Ingestion::IngestReportService calls Security::Ingestion::IngestReportSliceService which executes a number of tasks for a report slice.

Dismissal

If you change the state of a vulnerability, such as selecting Dismiss vulnerability the following things currently happen:

  • A Feedback record of dismissal type is created to record the current state.
  • If they do not already exist, a Vulnerability Finding and a Vulnerability with present_on_default_branch: false attribute get created, to which a State Transition reflecting the state change is related.

You can optionally add a comment to the state change which is recorded on both the Feedback and the State Transition.

Issue or Merge Request creation

If you select Create issue or Create merge request the following things currently happen:

  • A Vulnerabilities::Feedback record is created. The Feedback will have a feedback_type of issue or merge request and an issue_id or merge_request_id that's not NULL respective to the attachment.
  • If they do not already exist, a Vulnerability Finding and a Vulnerability with present_on_default_branch: false attribute get created, to which a Issue Link or Merge Request Link will be related respective to the action taken.

Vulnerabilities in the Default Branch

Security Findings detected in scan run on the default branch are saved as Vulnerabilities with the present_on_default_branch: true attribute and respective Vulnerability Finding records. Vulnerability records that already exist from interactions outside of the default branch will be updated to present_on_default_branch: true

Vulnerabilities which have already been interacted with will retain all existing State Transitions, Merge Request Links and Issue Links, as well as a corresponding Vulnerability Feedback.

Vulnerability Read Creation

Vulnerability::Read records are created via postgres database trigger upon the creation of a Vulnerability::Finding record and as such are part of our ingestion process, though they have no impact on it bar it's denormalization performance benefits on the report pages.

This style of creation was intended to be fast and seamless, but has proven difficult to debug and maintain and may be migrated to the application layer later.