OliveTin: StartActionAndWait Endpoints Bypass `logs` Permission and Return Action Output

Summary

The synchronous execution RPCs StartActionAndWait and StartActionByGetAndWait return the full LogEntry for the just-executed action without checking whether the caller is allowed to read that action's logs.

OliveTin's ACL model separates exec from logs. A deployment can intentionally allow a user to run an action while denying access to its historical or live output. That separation is enforced in GetLogs, GetActionLogs, ExecutionStatus, and EventStream, but it is not enforced in the synchronous ...AndWait endpoints.

As a result, any user who can execute an action through these endpoints can read the action output immediately even when the action's ACL explicitly sets logs:false.

Details

OliveTin defines separate per-action permissions:

// service/internal/config/config.go
type PermissionsList struct {
    View bool `koanf:"view"`
    Exec bool `koanf:"exec"`
    Logs bool `koanf:"logs"`
    Kill bool `koanf:"kill"`
}

The normal log and streaming paths correctly enforce logs permission:

// service/internal/api/api.go
func (api *oliveTinAPI) isLogEntryAllowed(e *executor.InternalLogEntry, user *authpublic.AuthenticatedUser) bool {
    if user == nil || !isValidLogEntry(e) {
        return false
    }
    return acl.IsAllowedLogs(api.cfg, user, e.Binding.Action)
}

That check is used by:

• GetLogs
• GetActionLogs
• ExecutionStatus
• EventStream

However, the synchronous execution endpoints directly return the created LogEntry without any logs ACL check:

// service/internal/api/api.go
func (api *oliveTinAPI) StartActionAndWait(ctx ctx.Context, req *connect.Request[apiv1.StartActionAndWaitRequest]) (*connect.Response[apiv1.StartActionAndWaitResponse], error) {
    ...
    internalLogEntry, ok := api.startActionAndWaitRun(binding, args, user)
    if !ok {
        return nil, connect.NewError(connect.CodeNotFound, fmt.Errorf("execution not found"))
    }
    return connect.NewResponse(&apiv1.StartActionAndWaitResponse{
        LogEntry: api.internalLogEntryToPb(internalLogEntry, user),
    }), nil
}

func (api *oliveTinAPI) StartActionByGetAndWait(ctx ctx.Context, req *connect.Request[apiv1.StartActionByGetAndWaitRequest]) (*connect.Response[apiv1.StartActionByGetAndWaitResponse], error) {
    ...
    internalLogEntry, ok := api.executor.GetLog(execReq.TrackingID)
    if ok {
        return connect.NewResponse(&apiv1.StartActionByGetAndWaitResponse{
            LogEntry: api.internalLogEntryToPb(internalLogEntry, user),
        }), nil
    }
    return nil, connect.NewError(connect.CodeNotFound, fmt.Errorf("execution not found"))
}

And internalLogEntryToPb() includes the full output:

// service/internal/api/api.go
func (api *oliveTinAPI) internalLogEntryToPb(logEntry *executor.InternalLogEntry, authenticatedUser *authpublic.AuthenticatedUser) *apiv1.LogEntry {
    pble := &apiv1.LogEntry{
        ActionTitle:         logEntry.ActionTitle,
        Output:              logEntry.Output,
        ExitCode:            logEntry.ExitCode,
        ExecutionTrackingId: logEntry.ExecutionTrackingID,
        ...
    }
    ...
    return pble
}

The executor separately enforces exec permission, but there is no subsequent check that the response should omit or deny Output when logs:false.

Local Reproduction

I verified this locally with a one-off test against the real StartActionAndWait handler.

Configuration used:

• action secret_action with shell echo SECRET_FROM_ACTION
• user low matched by ACL:
  • exec: true
  • logs: false
  • view: false
  • kill: false

Then I invoked StartActionAndWait as low through the real handler path using header-based auth.

Observed output from the test run:

=== RUN   TestTempStartActionAndWaitLeaksOutputWithoutLogsPermission
time="2026-03-12T23:36:33+01:00" level=info msg="Action requested" actionTitle=secret-action tags="[]"
time="2026-03-12T23:36:33+01:00" level=info msg="Action parse args - Before" actionTitle=secret-action cmd="echo SECRET_FROM_ACTION"
time="2026-03-12T23:36:33+01:00" level=info msg="Action parse args - After" actionTitle=secret-action cmd="echo SECRET_FROM_ACTION"
time="2026-03-12T23:36:33+01:00" level=info msg="Action started" actionTitle=secret-action timeout=5
time="2026-03-12T23:36:33+01:00" level=info msg="Action finished" actionTitle=secret-action exit=0 outputLength=40 timedOut=false
    temp_acl_bypass_test.go:56: output="S\x00E\x00C\x00R\x00E\x00T\x00_\x00F\x00R\x00O\x00M\x00_\x00A\x00C\x00T\x00I\x00O\x00N\x00\r\x00\n\x00" blocked=false action="secret-action"
--- PASS: TestTempStartActionAndWaitLeaksOutputWithoutLogsPermission (0.03s)

The UTF-16LE formatting is from Windows echo, but the key result is that the response returned the real command output even though the caller's ACL explicitly denied log access.

Impact

• Bypass of log confidentiality controls: Operators may configure actions to be executable but not log-readable. The ...AndWait endpoints break that separation.
• Disclosure of secrets printed by actions: Many OliveTin actions wrap administrative scripts and commands whose stdout/stderr may contain credentials, internal paths, hostnames, tokens, or sensitive operational data.
• Adjacent to previous output-leak bugs: OliveTin already had an EventStream output disclosure bug. This is a separate response-path authorization gap on the synchronous execution APIs.

Recommended Fix

Enforce logs permission before returning LogEntry content from synchronous execution endpoints.

Two reasonable fixes:

1. Deny the endpoint response when logs:false

if !acl.IsAllowedLogs(api.cfg, user, binding.Action) {
    return nil, connect.NewError(connect.CodePermissionDenied, fmt.Errorf("permission denied to view action output"))
}

2. Or redact log-bearing fields when logs:false

pb := api.internalLogEntryToPb(internalLogEntry, user)
if !acl.IsAllowedLogs(api.cfg, user, binding.Action) {
    pb.Output = ""
    pb.ExitCode = 0
}
return connect.NewResponse(&apiv1.StartActionAndWaitResponse{LogEntry: pb}), nil

The same fix should be applied to both:

• StartActionAndWait
• StartActionByGetAndWait

References

• service/internal/api/api.go
• service/internal/config/config.go
• service/internal/acl/acl.go