DEV Community

facilishz
facilishz

Posted on

52,068 SonarQube pages on 9000, 12,052 fingerprints, and the tokens that sit in the project settings

52,068 SonarQube pages on 9000, 12,052 fingerprints, and the tokens that sit in the project settings

SonarQube is a code analysis platform. Teams connect it to a source repository, it pulls the code, analyses it and reports on quality and security findings. To do that it stores credentials: repository tokens, CI webhook secrets, and in many installs the integration tokens that let the pipeline push analysis results.
Port 9000 is SonarQube's default and is also one of the most crowded ports on the internet, because every tool with a web interface that needed a port picked a round number. A count on that port is a count of a neighbourhood.

The measurements

Requested from ZoomEye through the vendor SDK on 26 October 2026, one record per query, scope all asset types. Totals only.
| Query | Reported total |
| --- | --- |
| port=9000 && service="http" | 7,869,772 |
| port=9000 && title="SonarQube" | 52,068 |
| app="SonarQube" | 12,052 |
| app="SonarQube" && country="US" | 3,948 |
| title="SonarQube" && country="US" | 36,298 |
The HTTP count on 9000 is close to eight million and is not worth quoting as anything except a demonstration that a port is not a product. Requiring the SonarQube title cuts it to 52,068, and the application fingerprint gives 12,052.
The direction of the gap between title and fingerprint is the interesting part. Titles outnumber fingerprints by more than four to one. SonarQube's web application sets a page title that names the product on its login page, so a title match is easy. The fingerprint depends on a distinctive response that survives the authentication redirect, and a scanner that only receives a login page has less to work with.
The title count is therefore the better upper bound for reachable instances, and the fingerprint count is the conservative floor. Quoting only one of them, without saying which, produces a number a reader cannot interpret.
The regional figures continue the pattern and follow the same field logic: a country filter applied to a title count inherits the title's looseness, so 36,298 should not be compared with 3,948 as if they measured the same thing on different bases.

Why a reachable SonarQube is worth a ticket

The platform holds the analysis results for every project it scans, which is a list of every known weakness in the organisation's source code with file and line references. It is a map of where to look, and more useful to an attacker than a generic version fingerprint.
It also holds the tokens. A project's analysis token lets a caller submit analysis results for that project, which can influence the quality gate. A global token or an administrator session does considerably more: user management, project configuration, and the ability to create new tokens.
SonarQube's documentation covers the authentication model and the token permissions. The default for a self-hosted install is form authentication, and the administrator account is created during setup with a password that operators have been known to leave at the default value in a lab and then promote to production.

Checks for your own estate

Confirm the instance is reachable only from the network that runs the analysis and the developers who read the results. A public instance is a design decision worth recording, and most estates find at least one that nobody intended.
Confirm the default administrator password was changed at setup. This is a one-line check against the identity API and it catches the case the documentation warns about.
Review token permissions and revoke tokens for projects that no longer exist. Stale tokens are the ones that outlive the employee who created them.
Check whether the instance is configured with a forced user authentication default, which determines whether anonymous users can browse project dashboards. Anonymous read is a deliberate feature on public instances and an unnecessary disclosure on internal ones.

Limits

The counts describe pages and fingerprints observed on one day, and a SonarQube behind an SSO proxy that returns a generic redirect will appear in neither. The numbers do not establish that authentication is weak, only that the interface is answering. Regional splits follow address geolocation.

References

Top comments (0)