OSV 1.4.0 · github-reviewed · 修改于 2026-07-30 22:25
发布时间
2026-07-30 22:25
GitHub 审查时间
2026-07-30 22:25
NVD 发布时间
2026-07-30 05:17
源文件
advisories/github-reviewed/2026/07/GHSA-jm28-2wcr-qf3h/GHSA-jm28-2wcr-qf3h.json
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.
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:
GetLogsGetActionLogsExecutionStatusEventStreamHowever, 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.
I verified this locally with a one-off test against the real StartActionAndWait handler.
Configuration used:
secret_action with shell echo SECRET_FROM_ACTIONlow matched by ACL:
exec: truelogs: falseview: falsekill: falseThen 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.
...AndWait endpoints break that separation.Enforce logs permission before returning LogEntry content from synchronous execution endpoints.
Two reasonable fixes:
logs:falseif !acl.IsAllowedLogs(api.cfg, user, binding.Action) {
return nil, connect.NewError(connect.CodePermissionDenied, fmt.Errorf("permission denied to view action output"))
}
logs:falsepb := 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:
StartActionAndWaitStartActionByGetAndWaitservice/internal/api/api.goservice/internal/config/config.goservice/internal/acl/acl.go