6.5
Medium
CVSS 3.1 base
What to do
If you run kestra 1.3.19 and earlier, upgrade to the fixed release in your line: 1.0.45 or 1.3.21.
Anyone who can reach it over the network, with a low-privilege account and no user interaction, can read sensitive data.Read from the CVSS 3.1 vector
To exploit it
What it gets
My analysis of this finding: root cause, the vulnerable code, the impact and the fix, in the site’s format. The official advisory of record is linked above.
| Package | Ecosystem | Vulnerable | Fixed in |
|---|---|---|---|
io.kestra:kestra | maven | <= 1.3.19 | 1.0.45, 1.3.21 |
Kestra’s execution file endpoints share one helper, validateFile(), whose job is to decide whether a requested kestra:// storage URI actually belongs to the execution named in the path. When the URI points at a different execution, the helper returns an HttpResponse.redirect() pointing at the correct execution’s endpoint, and the redirect is what forces the per-execution access check to run again against the right owner. Two of the three endpoints that call it honor that return value. The preview endpoint, previewFileFromExecution, throws it away. The result is an IDOR: any authenticated user who owns a single execution can read the output files of any other execution in the tenant, across namespace boundaries, by pointing path at the victim’s storage URI. I confirmed this against a live Kestra OSS v1.3.19 instance.
webserver/src/main/java/io/kestra/webserver/controllers/api/ExecutionController.java. The preview handler calls the validator as a statement and never looks at what comes back:
// previewFileFromExecution, line 1863this.validateFile(execution.get(), path, "/api/v1/" + this.getTenant() + "executions/{executionId}/file?path=" + path); String extension = FilenameUtils.getExtension(path.toString());// ... proceeds straight to storageInterface.get(...) using execution.get()'s tenant + namespaceThe bugcalls the validator but throws away the redirect it returns
The two sibling methods in the same class get it right. downloadFileFromExecution (line 913) and getFileMetadatasFromExecution (line 950) both capture and propagate the redirect:
HttpResponse<StreamedFile> httpResponse = this.validateFile(execution.get(), path, "...");
if (httpResponse != null) {
return httpResponse; // redirect to the owning execution, re-triggers the check
}
validateFile itself (line 843) is the control being defeated. When the path carries a foreign execution id, it does not read the file, it hands back a redirect so the request is re-dispatched against the execution that owns the data:
Optional<String> redirectedExecution = StorageContext.extractExecutionId(path);
if (redirectedExecution.isPresent()) {
return HttpResponse.redirect(/* .../executions/{foreignExecutionId}/file?path=... */);
}
By discarding that value, the preview path reads the victim’s file directly with the attacker execution’s tenant and namespace, which is exactly the context the redirect was supposed to force a recheck of.
With a single execution id the attacker legitimately owns and a kestra:// URI naming a victim execution, GET /executions/{attackerExecId}/file/preview?path=... returns HTTP 200 with the victim file’s content. The equivalent download request returns 301 and goes nowhere useful, which is the tell that the control works everywhere except here. In OSS this is cross-namespace exfiltration between teams sharing one instance; in EE it sidesteps namespace-level RBAC on execution outputs. My PoC pulled a task output file containing database and cloud credentials from a victim execution. Anything a workflow writes to internal storage, credentials, PII, financial records, is readable.
Make the preview endpoint match its siblings: capture the validateFile return and return it when non-null, so a foreign-execution path is redirected and rechecked instead of read.
HttpResponse<?> httpResponse = this.validateFile(execution.get(), path, "...");if (httpResponse != null) { return httpResponse;}The fixcaptures the validator's redirect instead of discarding itThe fixreturns the redirect so the owning execution is rechecked
Sanaan Fayaz Wani (GitHub sfwani) reported this vulnerability to the kestra maintainers under coordinated disclosure and is credited as a reporter in GHSA-r6v3-xxwj-9h42, published 2026-06-03.
All published findings: advisory index.