fix(smoke): conditional role drill, healthy-gate for drills, exit reaping

scripts/smoke.sh

@@ -18,14 +18,20 @@
 #   8. Bad-value drill: feeds --reconcile a wrong bands value and an
 #      oversized SSID pool (25 dummy entries), verifies both are
 #      repaired, restores originals.
-#   9. Rollback watchdog drill: runs mk8-watchdog.sh against a config
+#   9. RF role drill (only when SMOKE_UPLINK_SSID is set): switches
+#      radio1 to the uplink role against the named lab AP, expects an
+#      association, then back to attack with hopping resumed.
+#  10. Rollback watchdog drill: runs mk8-watchdog.sh against a config
 #      profile while the web UI is up (expects clean promote exit),
 #      then STOPS the pagerwebui service and expects the watchdog to
 #      roll back and journal a 'rollback' event, then restarts webui.
 #      Preceded by a 5-second warning countdown; brief web outage.
+#      Drills run only when every read-only check has passed.
 #
 # Environment:
 #   PASS              webui password used for POST /api/login (check 7).
+#   SMOKE_UPLINK_SSID lab AP SSID; enables the --write RF role drill.
+#   SMOKE_UPLINK_PSK  optional PSK for the lab AP.
 #
 # Exit status: 0 when every executed check passes, 1 otherwise.
 
@@ -42,9 +48,12 @@ JAR=/tmp/mk8-smoke-cookies.$$
 
 WRITE=0
 PASS="${PASS:-}"
+UPLINK_SSID="${SMOKE_UPLINK_SSID:-}"
+UPLINK_PSK="${SMOKE_UPLINK_PSK:-}"
 WEB_STOPPED=0
 FAILED=0
 WAIT_RC=0
+wp=""
 
 PY="$(command -v python3 2>/dev/null || true)"
 [ -n "$PY" ] || PY=/usr/bin/python3
@@ -53,14 +62,20 @@ PY="$(command -v python3 2>/dev/null || true)"
 usage() {
     printf 'Usage: smoke.sh [--write]\n'
     printf '\n'
-    printf 'Read-only checks run by default. --write adds two destructive\n'
-    printf 'drills that briefly toggle UCI config and stop/start the\n'
-    printf 'pagerwebui service; original values are restored automatically.\n'
+    printf 'Read-only checks run by default. --write adds destructive drills\n'
+    printf '(run only if every read-only check passed) that briefly toggle\n'
+    printf 'UCI config and stop/start the pagerwebui service; original values\n'
+    printf 'are restored automatically.\n'
     printf 'Set PASS=<webui password> to enable the authenticated API check.\n'
+    printf 'Set SMOKE_UPLINK_SSID=[<PSK via SMOKE_UPLINK_PSK>] to enable the\n'
+    printf '--write RF role drill against a lab AP.\n'
 }
 
 on_exit() {
     rm -f "$JAR" 2>/dev/null
+    if [ -n "$wp" ]; then
+        kill "$wp" 2>/dev/null
+    fi
     if [ "$WEB_STOPPED" = "1" ]; then
         info 'restoring pagerwebui service'
         "$WEBUI_INIT" start >/dev/null 2>&1
@@ -291,6 +306,67 @@ drill_bad_values() {
     "$PY" "$REL/server.py" --reconcile >/dev/null 2>&1
 }
 
+drill_role() {
+    if [ -z "$UPLINK_SSID" ]; then
+        skip 'role drill: set SMOKE_UPLINK_SSID to enable'
+        return
+    fi
+    info "role drill: uplink '$UPLINK_SSID' then attack"
+    out="$("$PY" - "$REL" "$LEGACY" "$UPLINK_SSID" "$UPLINK_PSK" <<'PYEOF' 2>&1
+import json
+import os
+import sys
+
+rel, legacy, ssid, psk = sys.argv[1:5]
+mk8_rfplan = None
+for d in (rel, legacy):
+    if os.path.isfile(os.path.join(d, 'mk8_rfplan.py')):
+        sys.path.insert(0, d)
+        try:
+            import mk8_rfplan
+            break
+        except Exception:
+            sys.path.remove(d)
+if mk8_rfplan is None:
+    print('FAIL mk8_rfplan not importable from release or legacy dir')
+    raise SystemExit(1)
+
+r1 = mk8_rfplan.set_role(
+    'uplink', ssid=ssid or None, psk=psk or None)
+if not isinstance(r1, dict) or not r1.get('ok'):
+    print('FAIL uplink set_role failed: %s' % json.dumps(r1))
+    raise SystemExit(1)
+
+assoc = mk8_rfplan.associated()
+if not assoc:
+    print('FAIL uplink associated() returned nothing after set_role')
+else:
+    print('associated as %s' % assoc)
+
+try:
+    r2 = mk8_rfplan.set_role('attack')
+except Exception as exc:
+    r2 = {'ok': False, 'error': str(exc)}
+if not isinstance(r2, dict) or not r2.get('ok'):
+    print('FAIL attack set_role failed: %s' % json.dumps(r2))
+    raise SystemExit(1)
+raise SystemExit(0 if assoc else 1)
+PYEOF
+)"
+    rc=$?
+    printf '%s\n' "$out" | sed 's/^/     /'
+    if [ "$rc" -ne 0 ]; then
+        fail 'role drill: uplink association failed (see detail above)'
+        return
+    fi
+    hop="$(uci_get pineapd.wlan1mon.hop)"
+    if [ "$hop" = "1" ]; then
+        pass 'role drill: uplink assoc + attack role + hopping resumed'
+    else
+        fail "role drill: pineapd.wlan1mon.hop=$hop after attack role (want 1)"
+    fi
+}
+
 drill_watchdog() {
     WDOG="$REL/mk8-watchdog.sh"
     [ -f "$WDOG" ] || WDOG="$LEGACY/mk8-watchdog.sh"
@@ -399,9 +475,12 @@ main() {
     if [ "$WRITE" = "1" ]; then
         if [ -z "$PY" ]; then
             fail 'drills: python3 required but not found'
+        elif [ "$FAILED" -ne 0 ]; then
+            skip 'drills: read-only checks failed; refusing drills'
         else
             printf -- '--- --write drills ---\n'
             drill_bad_values
+            drill_role
             drill_watchdog
         fi
     fi