Merge pull request #278 from Obihoernchen/fix/configurable-endpoints

configurable sockapi path and ipmi port

confluent_server/aiohmi/ipmi/console.py

@@ -148,7 +148,11 @@ class Console(object):
         # some BMCs disagree on the endianness, so do both
         valid_ports = (self.port, struct.unpack(
             '<H', struct.pack('>H', self.port))[0])
-        if (data[8] + (data[9] << 8)) not in valid_ports:
+        solport = data[8] + (data[9] << 8)
+        # A bmc behind a port forward answers with the port it listens on
+        # rather than the one it was reached through; payloads ride the
+        # session, never the advertised port.
+        if solport not in valid_ports and self.port == 623:
             # TODO(jbjohnso): support atypical SOL port number
             raise NotImplementedError("Non-standard SOL Port Number")
         # ignore data[10:11] for now, the vlan detail, shouldn't matter to this

confluent_server/confluent/plugins/hardwaremanagement/ipmi.py

@@ -263,14 +263,37 @@ def get_conn_params(node, configdata):
         kg = configdata['secret.ipmikg']['value']
     else:
         kg = passphrase
-    # TODO(jbjohnso): check if the end has some number after a : without []
-    # for non default port
+    # Read a port off the address, as the redfish plugin does. Brackets come
+    # off, unlike there: this address goes to socket.getaddrinfo, which rejects
+    # a bracketed literal, rather than into a URL where one is required.
+    bmc = bmc.strip()
+    port = 623
+    if bmc.startswith('['):
+        bracket_end = bmc.find(']')
+        if bracket_end > 0:
+            if len(bmc) > bracket_end + 1 and bmc[bracket_end + 1] == ':':
+                try:
+                    port = int(bmc[bracket_end + 2:])
+                except (ValueError, TypeError):
+                    pass
+            bmc = bmc[1:bracket_end]
+    elif bmc.count(':') == 1:
+        hostpart, _, portstr = bmc.rpartition(':')
+        try:
+            port = int(portstr)
+        except (ValueError, TypeError):
+            pass
+        bmc = hostpart
+    if not 0 < port <= 65535:
+        # NOTE: Fallback to 623 if port read is not valid
+        port = 623
+
     return {
         'username': username,
         'passphrase': passphrase,
         'kg': kg,
         'bmc': bmc,
-        'port': 623,
+        'port': port,
     }
 
 
@@ -288,6 +311,11 @@ def _donothing(data):
 class IpmiConsole(conapi.Console):
     configattributes = frozenset(_configattributes)
     bmctonodemapping = {}
+    # Whether this instance is the one that put its endpoint in the mapping
+    # above. False until it does, so that an instance rejected as a duplicate,
+    # or one that failed earlier in __init__, does not remove on the way out an
+    # entry that belongs to the node holding the endpoint.
+    claimedbmc = False
 
     def __init__(self, node, config):
         self.error = None
@@ -304,20 +332,26 @@ class IpmiConsole(conapi.Console):
         self.kg = connparams['kg']
         self.bmc = connparams['bmc']
         self.port = connparams['port']
+        # The port is part of the identity: several bmcs may sit behind one
+        # address on different ports, and they are distinct devices.
+        self.bmckey = (self.bmc, self.port)
         self.connected = False
-        # ok, is self.bmc unique among nodes already
+        # ok, is this bmc unique among nodes already
         # Cannot actually create console until 'connect', when we get callback
-        if (self.bmc in self.bmctonodemapping and
-                self.bmctonodemapping[self.bmc] != node):
+        if (self.bmckey in self.bmctonodemapping and
+                self.bmctonodemapping[self.bmckey] != node):
             raise Exception(
                 "Duplicate hardwaremanagement.manager attribute for {0} and {1}".format(
-                    node, self.bmctonodemapping[self.bmc]))
-        self.bmctonodemapping[self.bmc] = node
+                    node, self.bmctonodemapping[self.bmckey]))
+        self.bmctonodemapping[self.bmckey] = node
+        self.claimedbmc = True
 
     def __del__(self):
         self.solconnection = None
+        if not self.claimedbmc:
+            return
         try:
-            del self.bmctonodemapping[self.bmc]
+            del self.bmctonodemapping[self.bmckey]
         except KeyError:
             pass
 

confluent_server/confluent/sockapi.py

@@ -438,31 +438,40 @@ async def _tlsstartup(cnn):
         raise Exception('Unable to find workable SSL support')
     tasks.spawn(sessionhdl(cnn, authname, cert=cert))
 
-def removesocket():
+default_socketpath = "/var/run/confluent/api.sock"
+
+
+def removesocket(socketpath=None):
+    if socketpath is None:
+        socketpath = default_socketpath
     try:
-        os.remove("/var/run/confluent/api.sock")
+        os.remove(socketpath)
     except OSError:
         pass
 
-async def _unixdomainhandler(bind_group=None, bind_perms=None):
+async def _unixdomainhandler(bind_group=None, bind_perms=None,
+                             socketpath=None):
     aloop = asyncio.get_running_loop()
     if not bind_perms:
         bind_perms = 0o666
+    if socketpath is None:
+        socketpath = default_socketpath
     unixsocket = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
     unixsocket.settimeout(0)
     try:
-        os.remove("/var/run/confluent/api.sock")
+        os.remove(socketpath)
     except OSError:  # if file does not exist, no big deal
         pass
-    if not os.path.isdir("/var/run/confluent"):
-        os.makedirs('/var/run/confluent', 0o755)
+    socketdir = os.path.dirname(socketpath)
+    if socketdir and not os.path.isdir(socketdir):
+        os.makedirs(socketdir, 0o755)
     oldumask = os.umask(0o777 - bind_perms)
-    unixsocket.bind("/var/run/confluent/api.sock")
-    os.chmod("/var/run/confluent/api.sock", bind_perms)
+    unixsocket.bind(socketpath)
+    os.chmod(socketpath, bind_perms)
     if bind_group:
-        shutil.chown("/var/run/confluent/api.sock", group=bind_group)
+        shutil.chown(socketpath, group=bind_group)
     os.umask(oldumask)
-    atexit.register(removesocket)
+    atexit.register(removesocket, socketpath)
     unixsocket.listen(5)
     while True:
         cnn, addr = await aloop.sock_accept(unixsocket)
@@ -491,13 +500,15 @@ async def _unixdomainhandler(bind_group=None, bind_perms=None):
 
 
 class SockApi(object):
-    def __init__(self, bindhost=None, bindport=None, bindgroup=None, bindperms=None):
+    def __init__(self, bindhost=None, bindport=None, bindgroup=None,
+                 bindperms=None, socketpath=None):
         self.tlsserver = None
         self.unixdomainserver = None
         self.bind_host = bindhost or '::'
         self.bind_port = bindport or 13001
         self.bind_group = bindgroup
         self.bind_perms = bindperms
+        self.socketpath = socketpath or default_socketpath
 
     async def start(self):
         global auditlog
@@ -509,7 +520,8 @@ class SockApi(object):
             self.start_remoteapi()
         else:
             tasks.spawn(self.watch_for_cert())
-        self.unixdomainserver = tasks.spawn_task(_unixdomainhandler(self.bind_group, self.bind_perms))
+        self.unixdomainserver = tasks.spawn_task(_unixdomainhandler(
+            self.bind_group, self.bind_perms, self.socketpath))
 
     async def watch_for_cert(self):
         watcher = libc.inotify_init1(os.O_NONBLOCK)