contest: hw: mctrl: add support for sysrq

Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Author: kuba-moo

contest/hw/lib/mc_client.py

@@ -96,6 +96,17 @@ def health_check(self, machine_id):
         r.raise_for_status()
         return r.json()
 
+    def send_sysrq(self, machine_id, key):
+        """Send a SysRq key to a machine via SOL."""
+        data = {
+            'caller': self.caller,
+            'machine_id': machine_id,
+            'key': key,
+        }
+        r = requests.post(f'{self.base_url}/send_sysrq', json=data, timeout=30)
+        r.raise_for_status()
+        return r.json()
+
 
 def resolve_nic_id(nic_info_list, vendor, model):
     """Resolve a NIC id from vendor and model strings."""

contest/hw/lib/sol_listener.py

@@ -34,6 +34,8 @@ def __init__(self, db_pool, bmc_map):
         self.bmc_map = bmc_map
         self._stop_event = threading.Event()
         self._threads = []
+        self._master_fds = {}  # machine_id -> master_fd
+        self._fds_lock = threading.Lock()
 
     def start(self):
         """Start a reader thread for each machine."""
@@ -48,6 +50,22 @@ def stop(self):
         """Signal all sessions to stop."""
         self._stop_event.set()
 
+    def send_sysrq(self, machine_id, key):
+        """Send a SysRq key via the SOL session.
+
+        Writes ipmitool's break escape sequence (~B) followed by the
+        key character to the active SOL pty.  Returns True on success.
+        """
+        with self._fds_lock:
+            fd = self._master_fds.get(machine_id)
+        if fd is None:
+            return False
+        try:
+            os.write(fd, b'~B' + key.encode('ascii'))
+            return True
+        except OSError:
+            return False
+
     def _insert_chunk(self, machine_id, line, eol):
         ts = datetime.datetime.now(datetime.UTC)
         conn = self.db_pool.getconn()
@@ -114,6 +132,8 @@ def _run_session(self, machine_id, bmc):
             print(f"SOL: session started for machine {machine_id} "
                   f"(BMC {bmc.bmc_ipaddr})")
 
+            with self._fds_lock:
+                self._master_fds[machine_id] = master_fd
             try:
                 while not self._stop_event.is_set():
                     try:
@@ -129,6 +149,8 @@ def _run_session(self, machine_id, bmc):
                         except Exception as e:
                             print(f"SOL: DB error for machine {machine_id}: {e}")
             finally:
+                with self._fds_lock:
+                    self._master_fds.pop(machine_id, None)
                 os.close(master_fd)
                 proc.terminate()
                 try:

contest/hw/machine_control.py

@@ -302,6 +302,33 @@ def health_check():
     })
 
 
+@app.route('/send_sysrq', methods=['POST'])
+def send_sysrq():
+    """Send a SysRq key to a machine via the active SOL session."""
+    data = request.get_json() or {}
+    machine_id = data.get('machine_id')
+    key = data.get('key', '')
+
+    if machine_id is None:
+        return jsonify({'error': 'machine_id required'}), 400
+
+    if len(key) != 1 or not key.isascii():
+        return jsonify({'error': 'key must be a single ASCII character'}), 400
+
+    if not _check_auth(machine_id):
+        return jsonify({'error': 'unauthorized'}), 403
+
+    if sol_listener is None:
+        return jsonify({'error': 'SOL not running'}), 503
+
+    ok = sol_listener.send_sysrq(machine_id, key)
+    if not ok:
+        return jsonify({'error': f'No active SOL session for machine {machine_id}'}), 404
+
+    print(f"SysRq: sent '{key}' to machine {machine_id}")
+    return jsonify({'ok': True, 'machine_id': machine_id, 'key': key})
+
+
 @app.route('/reserve', methods=['POST'])
 def reserve():
     """Atomically reserve a group of machines."""
@@ -383,7 +410,7 @@ def reservation_close():
 
 def main():
     """Initialize services and run Flask app."""
-    global db_pool, health_checker, res_mgr  # pylint: disable=global-statement
+    global db_pool, health_checker, res_mgr, sol_listener  # pylint: disable=global-statement
 
     config = configparser.ConfigParser()
     cfg_paths = ['hw.config', 'machine_control.config']
@@ -426,6 +453,7 @@ def main():
 
     def _start_threads():
         """Start background threads — called after gunicorn fork."""
+        global sol_listener  # pylint: disable=global-statement
         sol_listener = SOLCollector(db_pool, bmc_map)
         sol_listener.start()
         health_checker.start()

contest/hw/mc_cli.py

@@ -193,6 +193,16 @@ def cmd_health_check(args, mc):
     return 0
 
 
+def cmd_sysrq(args, mc):
+    """Send a SysRq key to a machine via SOL."""
+    result = mc.send_sysrq(args.machine_id, args.key)
+    if args.json:
+        print(json.dumps(result, indent=2))
+        return 0
+    print(f"Sent SysRq '{args.key}' to machine {args.machine_id}")
+    return 0
+
+
 def main(argv=None):
     """Entry point: parse args and dispatch subcommand."""
     parser = argparse.ArgumentParser(
@@ -254,6 +264,13 @@ def main(argv=None):
     p_hc.add_argument('--machine-id', type=int, required=True,
                       help='machine ID')
 
+    p_sysrq = sub.add_parser('sysrq',
+                             help='send SysRq key to a machine via SOL')
+    p_sysrq.add_argument('--machine-id', type=int, required=True,
+                         help='machine ID')
+    p_sysrq.add_argument('--key', required=True,
+                         help='SysRq key (e.g. c=crashdump, b=reboot, e=SIGTERM)')
+
     args = parser.parse_args(argv)
 
     if not args.command:
@@ -275,6 +292,7 @@ def main(argv=None):
         'close': cmd_close,
         'power-cycle': cmd_power_cycle,
         'health-check': cmd_health_check,
+        'sysrq': cmd_sysrq,
     }
     try:
         return commands[args.command](args, mc)