Merge L3 default-deny-failclosed (R-NET fail-closed green)

conformance/tests/l3-fault-injection/default-deny-failclosed/chainsaw-test.yaml

@@ -1,20 +1,15 @@
 # =============================================================================
-# L3 FAULT-INJECTION SCAFFOLD — Default-deny fail-closed  (R-NET-01, R-NET-02)
+# L3 FAULT-INJECTION — Default-deny fail-closed  (R-NET-01, R-NET-02)  *SAFETY-CRITICAL*  [RUNNABLE]
-#                               *SAFETY-CRITICAL*
 # Catalog/spec: docs/14 R-TEST-05 lists default-deny among the behaviors that MUST
-# be proven by fault-injection (not happy-path). This is the L3 partner of the L2
+# be proven by FAULT INJECTION (not happy-path). L3 partner of tests/l2-behavioral/default-deny.
-# behavioral default-deny test (tests/l2-behavioral/default-deny).
 #
-# GIVEN a source that CAN reach a dest only because of an explicit allow rule,
+# GIVEN a source that can reach dest:8080 ONLY because of an explicit allow rule,
-# WHEN  the allow rule is REVOKED (fault injection: delete the CiliumNetworkPolicy),
+# WHEN  the allow rule is REVOKED (fault: delete both halves of the NetworkPolicy),
-# THEN  connectivity drops to fail-closed within a bounded window (revocation takes
+# THEN  connectivity drops to fail-closed within a bounded window (the baseline denies
-#       effect immediately — the baseline denies once the allow is gone), while DNS
+#       once the allow is gone), while DNS still resolves.
-#       still resolves.
 #
-# This proves the deny is enforced by the live datapath, not merely declared, and
+# Proves the deny is enforced by the live datapath, not merely declared, and that
-# that removing access fails closed rather than lingering open.
+# removing access fails CLOSED rather than lingering open.
-#
-# STATUS: scaffold — needs T-2.2 (baseline) + T-2.3 (rendered allow) + T-3.0 harness.
 # =============================================================================
 apiVersion: chainsaw.kyverno.io/v1alpha1
 kind: Test
@@ -27,43 +22,73 @@ metadata:
     conformance.home.k3s/requirements: "R-NET-01, R-NET-02"
     conformance.home.k3s/layer: "L3"
     conformance.home.k3s/safety-critical: "true"
+    conformance.home.k3s/executable: "true"
     conformance.home.k3s/intent: >-
       Revoking an allow rule drops connectivity to fail-closed immediately; DNS
       still works — the datapath enforces default-deny, not just declares it.
 spec:
+  timeouts:
+    apply: 90s
+    assert: 120s
+    exec: 150s
   steps:
     - name: given-source-allowed-to-dest
       try:
         - apply:
-            file: fixtures/allowed-pair.yaml   # TODO(T-2.3): source+dest+allow policy
+            file: fixtures/allowed-pair.yaml
+        # No readinessProbe (default-deny drops the kubelet probe) → Ready once Running.
+        - assert:
+            resource:
+              apiVersion: v1
+              kind: Pod
+              metadata: { name: source }
+              status: { phase: Running }
+        - assert:
+            resource:
+              apiVersion: v1
+              kind: Pod
+              metadata: { name: dest }
+              status: { phase: Running }
         - script:
+            # The explicit allow is the SOLE reason this succeeds (baseline denies all else).
             content: |
-              set -euo pipefail
+              set -eu
-              kubectl -n "$NAMESPACE" exec source -- nc -z -w 3 dest 80 && echo ALLOWED_OK
+              kubectl -n "$NAMESPACE" exec source -- nc -z -w 5 dest 8080 && echo ALLOWED_OK
             check:
               (contains($stdout, 'ALLOWED_OK')): true
     - name: when-revoke-allow-then-fail-closed
       try:
-        # FAULT INJECTION: delete the allow policy; baseline default-deny remains.
+        # -- FAULT INJECTION: revoke the allow (both halves); baseline default-deny remains. --
         - script:
             content: |
-              set -euo pipefail
+              set -eu
-              echo "TODO(T-3.0): delete the source->dest CiliumNetworkPolicy allow"
+              kubectl -n "$NAMESPACE" delete networkpolicy \
+                allow-source-to-dest-egress allow-source-to-dest-ingress --wait=true
+              echo REVOKED
         - script:
-            # Bounded poll: connection must start failing (fail-closed on revocation).
+            # Connectivity must fail CLOSED. Poll briefly for the datapath to converge; a
+            # persistent success is a LEAK (fails the test).
             content: |
-              set -euo pipefail
+              set -u
-              if kubectl -n "$NAMESPACE" exec source -- nc -z -w 3 dest 80; then
+              for i in $(seq 1 10); do
-                echo "LEAK: still reachable after allow revoked" >&2; exit 1
+                if kubectl -n "$NAMESPACE" exec source -- nc -z -w 5 dest 8080 2>/dev/null; then
+                  echo "attempt $i: still reachable, waiting for revocation to converge..."; sleep 3
+                else
+                  echo FAIL_CLOSED_OK; exit 0
                 fi
-              echo FAIL_CLOSED_OK
+              done
+              echo "LEAK: dest still reachable after allow revoked" >&2; exit 1
             check:
               (contains($stdout, 'FAIL_CLOSED_OK')): true
         - script:
-            # DNS must still work (default-deny permits DNS egress).
+            # DNS must STILL resolve (baseline permits DNS egress) — the deny is scoped, not total.
+            # busybox nslookup can exit non-zero on a successful forward lookup (server-PTR quirk),
+            # so assert on the RESOLVED ANSWER in the output, not the exit code.
             content: |
-              set -euo pipefail
+              set -u
-              kubectl -n "$NAMESPACE" exec source -- nslookup kubernetes.default && echo DNS_OK
+              out="$(kubectl -n "$NAMESPACE" exec source -- nslookup kubernetes.default 2>&1 || true)"
+              echo "$out"
+              echo "$out" | grep -qiE 'kubernetes\.default|10\.43\.' && echo DNS_OK || { echo "DNS_BLOCKED" >&2; exit 1; }
             check:
               (contains($stdout, 'DNS_OK')): true
       finally:

conformance/tests/l3-fault-injection/default-deny-failclosed/fixtures/allowed-pair.yaml

@@ -0,0 +1,108 @@
+# =============================================================================
+# Fixtures for the default-deny FAIL-CLOSED fault-injection test (R-NET-01/02).
+# Applied into the per-test namespace chainsaw creates (hermetic, self-cleaned).
+#
+# Models: a namespace default-deny baseline + ONE explicit allow that is the SOLE
+# reason source can reach dest:8080. The test later DELETES the allow (fault) and
+# proves connectivity drops to fail-closed while DNS still resolves.
+#
+# Standard Kubernetes NetworkPolicy (Cilium enforces it). No readinessProbe on the
+# pods: under default-deny the kubelet TCP probe (node->pod) is dropped, so a
+# probe-bearing pod would never go Ready — without a probe the container is Ready
+# once running, which is all the test needs (K2 finding).
+# =============================================================================
+---
+# --- Default-deny baseline: deny all ingress+egress, allow only DNS egress. -----------
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default-deny-baseline
+spec:
+  podSelector: {}
+  policyTypes: [Ingress, Egress]
+  ingress: []
+  egress:
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+---
+# --- The ONE explicit allow (source -> dest tcp/8080): the two matching halves a
+#     real rendered dependency produces (egress permit on source, ingress on dest).
+#     Deleting BOTH is the fault injection.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: allow-source-to-dest-egress
+spec:
+  podSelector:
+    matchLabels: { app: source }
+  policyTypes: [Egress]
+  egress:
+    - to:
+        - podSelector:
+            matchLabels: { app: dest }
+      ports:
+        - protocol: TCP
+          port: 8080
+---
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: allow-source-to-dest-ingress
+spec:
+  podSelector:
+    matchLabels: { app: dest }
+  policyTypes: [Ingress]
+  ingress:
+    - from:
+        - podSelector:
+            matchLabels: { app: source }
+      ports:
+        - protocol: TCP
+          port: 8080
+---
+# --- source: an idle probe pod (nc is issued via kubectl exec). ------------------------
+apiVersion: v1
+kind: Pod
+metadata:
+  name: source
+  labels: { app: source }
+spec:
+  containers:
+    - name: app
+      image: busybox:1.36.1
+      command: ["sh", "-c", "sleep 3600"]
+---
+# --- dest: LISTENS on 8080 so a block is caused by POLICY, not a missing listener. -----
+apiVersion: v1
+kind: Pod
+metadata:
+  name: dest
+  labels: { app: dest }
+spec:
+  containers:
+    - name: app
+      image: busybox:1.36.1
+      command: ["httpd", "-f", "-p", "8080", "-h", "/tmp"]
+      ports:
+        - containerPort: 8080
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: dest
+spec:
+  selector: { app: dest }
+  ports:
+    - name: p8080
+      port: 8080
+      targetPort: 8080