Send it to ask@stokaro.com, not to the issue tracker. An issue is public from
the moment it is filed, and a report there tells everybody who reads it how to
reproduce the problem before there is a release that fixes it.
Write security in the subject so it is not read as a commercial enquiry.
Include what somebody needs to reproduce it:
- the chart version, and the manager, executor, and runner image digests;
- the Kubernetes version and distribution;
- the database engine and server version, where the report involves one;
- the resources involved, with credentials and connection strings removed;
- what the operator did, what you expected instead, and what an attacker gains.
You will get a reply saying whether the report is understood and what happens next. A fix ships in a release that says what it changed, on Release notes, and the advisory is published with it.
The boundary this policy covers lies between the operator and the application
namespaces it serves: the people who write PtahSchema, PtahMigration and
approval resources there, the database credentials those namespaces hold, and
the operation Pods that run with them. The operator's job at that boundary is
to hold five authorities apart -- who writes desired state, who approves a
plan, what the controller may write, what credentials a Job receives, and
which namespaces may manage a database another namespace manages too -- and to
bind evidence to the exact artifact it was produced from. A way to cross one
of those boundaries is what this policy is for:
- running SQL a plan does not contain, or applying a plan nobody approved;
- stopping or delaying another namespace's resources by claiming its database
without a
PtahRealmthat admits the claim; - reaching a credential from a resource or a Job that should not have it;
- making the operator write an object its admission policies refuse, or writing one on its behalf;
- passing evidence, a plan, an approval, or an image off as one it is not;
- making the operator report an Apply as settled when it is not, or lose the record that it is unresolved.
So is a way past the checks that keep a stale or buggy previous release from doing harm: one that narrows a newer CRD schema, starts over state a newer controller wrote, runs beside the current manager, or writes an object outside the shape admission allows the controller.
Security model says how those boundaries are drawn and which component holds each one.
- Anything that starts from the release namespace, or from
coordination.namespacewhen it is a separate one. Both are part of the operator's trusted computing base: Kubernetes lets whoever can create a Pod in a namespace run it as any ServiceAccount there, so a principal that can create or modify workloads, exec into Pods, request ServiceAccount tokens or write Leases in either namespace is a Ptah administrator. What it does there is administration. That includes a previous release that was tampered with rather than merely stale. The release namespace contract says so, and it is a deployment's job to keep both namespaces to Ptah administrators. - A cluster where the reporter already holds the privileges in question. An
account that may edit a
PtahSchemamay selectspec.policy.apply: Always, and that is the resource's own field rather than a bypass; an account that may write the controller's objects may do what the controller does. - The database's own privileges. The operator runs what the credentials it is given allow, and granting a cluster-wide administrative account is a deployment decision the security model asks you not to make.
- The deployment responsibilities that page lists as yours: NetworkPolicies, image pinning, one release per cluster.
- Ptah itself. Schema rendering, migration files, and dialect support belong in stokaro/ptah.
There is one supported version: the newest published release. Before v1 there
are no deprecations and no backports, and
API compatibility
says what that means for objects you have already stored. Pin the chart version
you deploy and read what a release changed before you move to it.