|
| 1 | +# KEP-4940: Add Pod Security Admission (PSA) to block setting `.host` field from ProbeHandler and LifecycleHandler |
| 2 | + |
| 3 | +<!-- toc --> |
| 4 | +- [Release Signoff Checklist](#release-signoff-checklist) |
| 5 | +- [Summary](#summary) |
| 6 | + - [Goals](#goals) |
| 7 | + - [Non-Goals](#non-goals) |
| 8 | +- [Proposal](#proposal) |
| 9 | + - [Risks and Mitigations](#risks-and-mitigations) |
| 10 | +- [Design Details](#design-details) |
| 11 | + - [Test Plan](#test-plan) |
| 12 | + - [Prerequisite testing updates](#prerequisite-testing-updates) |
| 13 | + - [Unit tests](#unit-tests) |
| 14 | + - [Integration tests](#integration-tests) |
| 15 | + - [e2e tests](#e2e-tests) |
| 16 | + - [Graduation Criteria](#graduation-criteria) |
| 17 | + - [Upgrade / Downgrade Strategy](#upgrade--downgrade-strategy) |
| 18 | + - [Version Skew Strategy](#version-skew-strategy) |
| 19 | +- [Production Readiness Review Questionnaire](#production-readiness-review-questionnaire) |
| 20 | + - [Feature Enablement and Rollback](#feature-enablement-and-rollback) |
| 21 | + - [Rollout, Upgrade and Rollback Planning](#rollout-upgrade-and-rollback-planning) |
| 22 | + - [Monitoring Requirements](#monitoring-requirements) |
| 23 | + - [Dependencies](#dependencies) |
| 24 | + - [Scalability](#scalability) |
| 25 | + - [Troubleshooting](#troubleshooting) |
| 26 | +- [Implementation History](#implementation-history) |
| 27 | +- [Drawbacks](#drawbacks) |
| 28 | +- [Alternatives](#alternatives) |
| 29 | +<!-- /toc --> |
| 30 | + |
| 31 | +## Release Signoff Checklist |
| 32 | + |
| 33 | +<!-- |
| 34 | +**ACTION REQUIRED:** In order to merge code into a release, there must be an |
| 35 | +issue in [kubernetes/enhancements] referencing this KEP and targeting a release |
| 36 | +milestone **before the [Enhancement Freeze](https://git.k8s.io/sig-release/releases) |
| 37 | +of the targeted release**. |
| 38 | +
|
| 39 | +For enhancements that make changes to code or processes/procedures in core |
| 40 | +Kubernetes—i.e., [kubernetes/kubernetes], we require the following Release |
| 41 | +Signoff checklist to be completed. |
| 42 | +
|
| 43 | +Check these off as they are completed for the Release Team to track. These |
| 44 | +checklist items _must_ be updated for the enhancement to be released. |
| 45 | +--> |
| 46 | + |
| 47 | +Items marked with (R) are required *prior to targeting to a milestone / release*. |
| 48 | + |
| 49 | +- [ ] (R) Enhancement issue in release milestone, which links to KEP dir in [kubernetes/enhancements] (not the initial KEP PR) |
| 50 | +- [ ] (R) KEP approvers have approved the KEP status as `implementable` |
| 51 | +- [ ] (R) Design details are appropriately documented |
| 52 | +- [ ] (R) Test plan is in place, giving consideration to SIG Architecture and SIG Testing input (including test refactors) |
| 53 | + - [ ] e2e Tests for all Beta API Operations (endpoints) |
| 54 | + - [ ] (R) Ensure GA e2e tests meet requirements for [Conformance Tests](https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/conformance-tests.md) |
| 55 | + - [ ] (R) Minimum Two Week Window for GA e2e tests to prove flake free |
| 56 | +- [ ] (R) Graduation criteria is in place |
| 57 | + - [ ] (R) [all GA Endpoints](https://github.com/kubernetes/community/pull/1806) must be hit by [Conformance Tests](https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/conformance-tests.md) |
| 58 | +- [ ] (R) Production readiness review completed |
| 59 | +- [ ] (R) Production readiness review approved |
| 60 | +- [ ] "Implementation History" section is up-to-date for milestone |
| 61 | +- [ ] User-facing documentation has been created in [kubernetes/website], for publication to [kubernetes.io] |
| 62 | +- [ ] Supporting documentation—e.g., additional design documents, links to mailing list discussions/SIG meetings, relevant PRs/issues, release notes |
| 63 | + |
| 64 | +<!-- |
| 65 | +**Note:** This checklist is iterative and should be reviewed and updated every time this enhancement is being considered for a milestone. |
| 66 | +--> |
| 67 | + |
| 68 | +[kubernetes.io]: https://kubernetes.io/ |
| 69 | +[kubernetes/enhancements]: https://git.k8s.io/enhancements |
| 70 | +[kubernetes/kubernetes]: https://git.k8s.io/kubernetes |
| 71 | +[kubernetes/website]: https://git.k8s.io/website |
| 72 | + |
| 73 | +## Summary |
| 74 | + |
| 75 | +We have a `Host` field that can be set from `TCPSocketAction` and |
| 76 | +`HTTPGetAction` fields which are part of the `ProbeHandler` and |
| 77 | +`LifecycleHandler` structs in Kubernetes that are used in |
| 78 | +`InitContainers` and `Containers` structs of `PodSpec`. |
| 79 | +The `Host` field is used for allowing users to specify |
| 80 | +another entity other than the podIP (which is the default value) to |
| 81 | +which Kubelet should perform probes to. |
| 82 | +However this opens it up for security attacks since the `Host` |
| 83 | +field can be set to pretty much any value in the system including |
| 84 | +security sensitive external hosts or localhost on the node. |
| 85 | +Kubelet will be probing this set `Host` value which can |
| 86 | +lead to blind SSRF attacks. |
| 87 | + |
| 88 | +### Goals |
| 89 | + |
| 90 | +* Add Pod Security Admission (PSA) to enable admins to restrict |
| 91 | + users from creating probes with the `Host` field set. |
| 92 | +* The Baseline Pod Security Standard (PSS) will be updated to enforce |
| 93 | + blocking this field so that it helps with easier adoption for |
| 94 | + workload operators given this is a known issue we want to prevent. |
| 95 | + |
| 96 | +### Non-Goals |
| 97 | + |
| 98 | +* Removing `.Host` field from the API and dropping support (It is |
| 99 | + unsaid rule that nothing can get removed from core Kubernetes API) |
| 100 | + |
| 101 | +## Proposal |
| 102 | + |
| 103 | +There is a long term plan to deprecate the existing TCP and HTTP probe |
| 104 | +types in the API to replace them with ones with slightly different semantics. |
| 105 | +See [KEP-4559](https://github.com/kubernetes/enhancements/pull/4558) for more |
| 106 | +details. Given the unsolvable security problems with the Host field, |
| 107 | +we do not plan to offer it in the new types. |
| 108 | + |
| 109 | +Meanwhile, the older API is never going to go away. So we also want to |
| 110 | +add PSA to allow admins to be able to restrict users from creating |
| 111 | +probes with the Host field set when using the (about to be deprecated) API. |
| 112 | +This is implemented by [kubernetes PR 125271](https://github.com/kubernetes/kubernetes/pull/125271) |
| 113 | +that does exactly that. |
| 114 | + |
| 115 | +### Risks and Mitigations |
| 116 | + |
| 117 | +There might be users who depend on the `Host` field in |
| 118 | +their existing probes which will continue to work and if |
| 119 | +newly created probes also need the `Host` field to point |
| 120 | +to an external destination then the admin can avoid enforcing |
| 121 | +the PSA to block it. |
| 122 | + |
| 123 | +## Design Details |
| 124 | + |
| 125 | +Add a Baseline APILevel Pod Security Admission policy to allow admins of the |
| 126 | +cluster to block users from setting `.host` field in: |
| 127 | + |
| 128 | +* spec.containers[*].LivenessProbe.ProbeHandler.HTTPGet.Host |
| 129 | +* spec.containers[*].ReadinessProbe.ProbeHandler.HTTPGet.Host |
| 130 | +* spec.containers[*].StartupProbe.ProbeHandler.HTTPGet.Host |
| 131 | +* spec.containers[*].LivenessProbe.ProbeHandler.TCPSocket.Host |
| 132 | +* spec.containers[*].ReadinessProbe.ProbeHandler.TCPSocket.Host |
| 133 | +* spec.containers[*].StartupProbe.ProbeHandler.TCPSocket.Host |
| 134 | +* spec.containers[*].Lifecycle.PostStart.TCPSocket.Host // Deprecated. TCPSocket is NOT supported as a LifecycleHandler and kept for backward compatibility. |
| 135 | +* spec.containers[*].Lifecycle.PreStop.TCPSocket.Host // Deprecated. TCPSocket is NOT supported as a LifecycleHandler and kept for backward compatibility. |
| 136 | +* spec.containers[*].Lifecycle.PostStart.HTTPGet.Host |
| 137 | +* spec.containers[*].Lifecycle.PreStop.HTTPGet.Host |
| 138 | +* spec.initContainers[*].LivenessProbe.ProbeHandler.HTTPGet.Host |
| 139 | +* spec.initContainers[*].ReadinessProbe.ProbeHandler.HTTPGet.Host |
| 140 | +* spec.initContainers[*].StartupProbe.ProbeHandler.HTTPGet.Host |
| 141 | +* spec.initContainers[*].LivenessProbe.ProbeHandler.TCPSocket.Host |
| 142 | +* spec.initContainers[*].ReadinessProbe.ProbeHandler.TCPSocket.Host |
| 143 | +* spec.initContainers[*].StartupProbe.ProbeHandler.TCPSocket.Host |
| 144 | +* spec.initContainers[*].Lifecycle.PostStart.TCPSocket.Host // Deprecated. TCPSocket is NOT supported as a LifecycleHandler and kept for backward compatibility. |
| 145 | +* spec.initContainers[*].Lifecycle.PreStop.TCPSocket.Host // Deprecated. TCPSocket is NOT supported as a LifecycleHandler and kept for backward compatibility. |
| 146 | +* spec.initContainers[*].Lifecycle.PostStart.HTTPGet.Host |
| 147 | +* spec.initContainers[*].Lifecycle.PreStop.HTTPGet.Host |
| 148 | + |
| 149 | +### Test Plan |
| 150 | + |
| 151 | +* Unit and E2E tests will be added to ensure the PSA works as expected |
| 152 | + |
| 153 | +##### Prerequisite testing updates |
| 154 | + |
| 155 | +None |
| 156 | + |
| 157 | +##### Unit tests |
| 158 | + |
| 159 | +Necessary unit tests will be added to the [PSA package] for |
| 160 | +testing the new code. |
| 161 | +Current test coverage status for the package is: |
| 162 | +- `k8s.io/pod-security-admission/policy`: `2025-05-06` - `89.9%` |
| 163 | +- `k8s.io/pod-security-admission/test`: `TBD` - `TBD` |
| 164 | + |
| 165 | +[PSA package]: https://github.com/kubernetes/kubernetes/tree/master/staging/src/k8s.io/pod-security-admission/policy |
| 166 | + |
| 167 | +##### Integration tests |
| 168 | + |
| 169 | +The following integration tests will be added to verify the PSA validation logic: |
| 170 | + |
| 171 | +1. Test that pods with `.host` field set in probes are rejected when PSA is enabled with baseline level |
| 172 | +2. Test that pods without `.host` field set in probes are allowed when PSA is enabled with baseline level |
| 173 | +3. Test that existing pods with `.host` field set continue to work when PSA is enabled |
| 174 | +4. Test that pods with `.host` field set are allowed when PSA is disabled or using an older version |
| 175 | + |
| 176 | +These tests will be added to: |
| 177 | +- `test/integration/auth/podsecurity_test.go` |
| 178 | +https://storage.googleapis.com/k8s-triage/index.html?test=TestPodSecurity |
| 179 | + |
| 180 | +The integration tests will verify the PSA policy validation logic by: |
| 181 | +- Creating test cases for each probe type (HTTPGet, TCPSocket) in a pod |
| 182 | +- Testing each probe location (LivenessProbe, ReadinessProbe, StartupProbe, LifecycleHandler) |
| 183 | +- Verifying the PSA policy enforcement at the baseline level |
| 184 | +- Testing the behavior with different PSA configurations |
| 185 | + |
| 186 | +##### e2e tests |
| 187 | + |
| 188 | +There are no Pod Security specific E2E tests (we rely on integration test coverage instead), |
| 189 | +but the Pod Security admission controller is enabled in E2E clusters, |
| 190 | +and all E2E test namespaces are labeled with the enforcement label for Pod Security. |
| 191 | + |
| 192 | +### Graduation Criteria |
| 193 | + |
| 194 | +The PSA added will be done within a single release |
| 195 | +and given there will be no feature gates for that, |
| 196 | +there is no need for multi-release graduation criteria. |
| 197 | +All related code will land within the same single release |
| 198 | + |
| 199 | +### Upgrade / Downgrade Strategy |
| 200 | + |
| 201 | +Any older pods with this field set should not be affected |
| 202 | +with the above solution. Only newer pods getting created |
| 203 | +with the field will be alerted. |
| 204 | + |
| 205 | +Users who are using this field can switch to using exec |
| 206 | +probes moving forward which should unblock them given exec |
| 207 | +probes can provide the same functionality. |
| 208 | + |
| 209 | + |
| 210 | +### Version Skew Strategy |
| 211 | + |
| 212 | +N/A since its only within a single component: pod-security-admission |
| 213 | +and doesn't cross multiple components. |
| 214 | + |
| 215 | +## Production Readiness Review Questionnaire |
| 216 | + |
| 217 | +### Feature Enablement and Rollback |
| 218 | + |
| 219 | +###### How can this feature be enabled / disabled in a live cluster? |
| 220 | + |
| 221 | +We decided to not go with feature gates and use PSA versioning. |
| 222 | +So if the admin sets pod-security.kubernetes.io/enforce-version: v1.34 |
| 223 | +along with pod-security.kubernetes.io/enforce: <LEVEL> |
| 224 | +on a namespace this feature will get enabled. |
| 225 | + |
| 226 | +###### Does enabling the feature change any default behavior? |
| 227 | + |
| 228 | +* There is no effect on clusters where PSA is not enabled OR an older |
| 229 | +PSA version is used. |
| 230 | + |
| 231 | +* There is no effect on clusters where `.Host` probes are not used |
| 232 | + |
| 233 | +* There is no effect on clusters where an older PSA versioning is being |
| 234 | +used |
| 235 | + |
| 236 | +* If users create new pod with `.Host` probes field set and the admin |
| 237 | +has set baseline PSA level to `enforce` mode then the request will be |
| 238 | +actively blocked and rejected. Existing pods with `.Host` probes |
| 239 | +that are upgrading will not be impacted unless PSA level is set to |
| 240 | +`enforce` mode. |
| 241 | + |
| 242 | + |
| 243 | +###### Can the feature be disabled once it has been enabled (i.e. can we roll back the enablement)? |
| 244 | + |
| 245 | +No |
| 246 | + |
| 247 | +###### What happens if we reenable the feature if it was previously rolled back? |
| 248 | + |
| 249 | +N/A |
| 250 | + |
| 251 | +###### Are there any tests for feature enablement/disablement? |
| 252 | + |
| 253 | +N/A since there is no feature gate |
| 254 | + |
| 255 | +### Rollout, Upgrade and Rollback Planning |
| 256 | + |
| 257 | + |
| 258 | +###### How can a rollout or rollback fail? Can it impact already running workloads? |
| 259 | + |
| 260 | +* Running workloads/deployments that have `.Host` probes set when upgraded to |
| 261 | + the latest version where they get rolled-out, if the PSA enforce label is |
| 262 | + placed on the namespace of the workload, then the workload will fail to get created. |
| 263 | +* If pod security label is not enabled on the namespace, then there is no |
| 264 | + impact on running workloads |
| 265 | + |
| 266 | +###### What specific metrics should inform a rollback? |
| 267 | + |
| 268 | +If your workloads are not rolling out due to the policy rejecting the request, |
| 269 | +then cluster admins can use the [PSA denial metrics]. Example, the `pod_security_evaluations_total` |
| 270 | +can indicate how many "deny" decisions were done based on number of policy evaluations that |
| 271 | +occurred. |
| 272 | + |
| 273 | +[PSA denial metrics]: https://kubernetes.io/docs/concepts/security/pod-security-admission/#metrics |
| 274 | + |
| 275 | +###### Were upgrade and rollback tested? Was the upgrade->downgrade->upgrade path tested? |
| 276 | + |
| 277 | +N/A |
| 278 | + |
| 279 | +###### Is the rollout accompanied by any deprecations and/or removals of features, APIs, fields of API types, flags, etc.? |
| 280 | + |
| 281 | +N/A |
| 282 | + |
| 283 | +### Monitoring Requirements |
| 284 | + |
| 285 | +N/A |
| 286 | + |
| 287 | +###### How can an operator determine if the feature is in use by workloads? |
| 288 | + |
| 289 | +If pods have probes with `.Host` field set and PSA label is set on that pod's namespace |
| 290 | +to a version where the new admission has been added, then it means the feature is enabled. |
| 291 | + |
| 292 | +###### How can someone using this feature know that it is working for their instance? |
| 293 | + |
| 294 | +Trying to create a pod with `.Host` field set in the probes will fail |
| 295 | +like this: |
| 296 | +``` |
| 297 | +Error from server (Forbidden): error when creating "psa/fail-case-pod.yaml": pods "liveness-http-pass" is forbidden: violates PodSecurity "restricted:latest": probeHost (container "liveness" uses probeHost 135.45.63.4) |
| 298 | +``` |
| 299 | +Trying to rollout a deployment with `.Host` field set in probes will fail with the following status: |
| 300 | +``` |
| 301 | + - lastTransitionTime: "2025-06-17T06:17:36Z" |
| 302 | + lastUpdateTime: "2025-06-17T06:17:36Z" |
| 303 | + message: 'pods "hello-world-577c86d6dd-bs7nt" is forbidden: violates PodSecurity |
| 304 | + "restricted:latest": probeHost (container "hello-world" uses probeHost 135.45.63.4)' |
| 305 | + reason: FailedCreate |
| 306 | + status: "True" |
| 307 | + type: ReplicaFailure |
| 308 | + observedGeneration: 1 |
| 309 | + unavailableReplicas: 1 |
| 310 | +``` |
| 311 | + |
| 312 | +###### What are the reasonable SLOs (Service Level Objectives) for the enhancement? |
| 313 | +N/A |
| 314 | + |
| 315 | +###### What are the SLIs (Service Level Indicators) an operator can use to determine the health of the service? |
| 316 | + |
| 317 | +cluster admins can use the [PSA denial metrics] to determine if something is |
| 318 | +wrong with their workloads and services are not serving properly due to policy |
| 319 | +enforcement. |
| 320 | + |
| 321 | +[PSA denial metrics]: https://kubernetes.io/docs/concepts/security/pod-security-admission/#metrics |
| 322 | + |
| 323 | +###### Are there any missing metrics that would be useful to have to improve observability of this feature? |
| 324 | + |
| 325 | +N/A |
| 326 | + |
| 327 | +### Dependencies |
| 328 | + |
| 329 | +None |
| 330 | + |
| 331 | +###### Does this feature depend on any specific services running in the cluster? |
| 332 | + |
| 333 | +No |
| 334 | + |
| 335 | +### Scalability |
| 336 | + |
| 337 | +N/A |
| 338 | + |
| 339 | +###### Will enabling / using this feature result in any new API calls? |
| 340 | + |
| 341 | +No |
| 342 | + |
| 343 | +###### Will enabling / using this feature result in introducing new API types? |
| 344 | + |
| 345 | +No |
| 346 | + |
| 347 | +###### Will enabling / using this feature result in any new calls to the cloud provider? |
| 348 | + |
| 349 | +No |
| 350 | + |
| 351 | +###### Will enabling / using this feature result in increasing size or count of the existing API objects? |
| 352 | + |
| 353 | +No |
| 354 | + |
| 355 | +###### Will enabling / using this feature result in increasing time taken by any operations covered by existing SLIs/SLOs? |
| 356 | + |
| 357 | +No |
| 358 | + |
| 359 | +###### Will enabling / using this feature result in non-negligible increase of resource usage (CPU, RAM, disk, IO, ...) in any components? |
| 360 | + |
| 361 | +No |
| 362 | + |
| 363 | +###### Can enabling / using this feature result in resource exhaustion of some node resources (PIDs, sockets, inodes, etc.)? |
| 364 | + |
| 365 | +No |
| 366 | + |
| 367 | +### Troubleshooting |
| 368 | + |
| 369 | +###### How does this feature react if the API server and/or etcd is unavailable? |
| 370 | + |
| 371 | +N/A |
| 372 | + |
| 373 | +###### What are other known failure modes? |
| 374 | + |
| 375 | +N/A |
| 376 | + |
| 377 | +###### What steps should be taken if SLOs are not being met to determine the problem? |
| 378 | + |
| 379 | +So if the admin sets `pod-security.kubernetes.io/enforce-version: v1.34` |
| 380 | +on a namespace this feature will get enabled and workloads rolling out |
| 381 | +with `.Host` probes set will be impacted. One of the remediation procedures to |
| 382 | +get workloads into a healthy state would be: |
| 383 | + |
| 384 | +* To pin the the [PSA namespace label] to a version prior to the version where this |
| 385 | + field is introduced (example set it to v1.33) |
| 386 | +* Restart your workloads. |
| 387 | + |
| 388 | +[PSA namespace label]: https://kubernetes.io/docs/concepts/security/pod-security-admission/#pod-security-admission-labels-for-namespaces |
| 389 | + |
| 390 | +## Implementation History |
| 391 | + |
| 392 | +## Drawbacks |
| 393 | + |
| 394 | +N/A |
| 395 | + |
| 396 | +## Alternatives |
| 397 | + |
| 398 | +The alternative is to remove this field from the API after |
| 399 | +its deprecated, but that's not a supported API action. |
0 commit comments