fix(LinstorSR): ensure _is_master is always set

`_is_master` is not always initialized, and more precisely
in the case of detach where LinstorSR.load method is not called.

Still in this same situation, this can lead to the deletion of DRBD diskless
on the master while we try to always have one so as not to needlessly
recreate one later.

Signed-off-by: Ronan Abhamon <ronan.abhamon@vates.fr>

drivers/LinstorSR.py

@@ -362,9 +362,6 @@ def load(self, sr_uuid):
         self._linstor = None  # Ensure that LINSTOR attribute exists.
         self._journaler = None
 
-        self._is_master = False
-        if 'SRmaster' in self.dconf and self.dconf['SRmaster'] == 'true':
-            self._is_master = True
         self._group_name = self.dconf['group-name']
 
         self._vdi_shared_time = 0
@@ -437,7 +434,7 @@ def connect():
 
                 return wrapped_method(self, *args, **kwargs)
 
-            if not self._is_master:
+            if not self.is_master():
                 if self.cmd in [
                     'sr_create', 'sr_delete', 'sr_update', 'sr_probe',
                     'sr_scan', 'vdi_create', 'vdi_delete', 'vdi_resize',
@@ -472,7 +469,7 @@ def connect():
 
                 # Ensure we use a non-locked volume when vhdutil is called.
                 if (
-                    self._is_master and self.cmd.startswith('vdi_') and
+                    self.is_master() and self.cmd.startswith('vdi_') and
                     self.cmd != 'vdi_create'
                 ):
                     self._linstor.ensure_volume_is_not_locked(
@@ -487,7 +484,7 @@ def connect():
                     #
                     # If the command is a SR command we want at least to remove
                     # resourceless volumes.
-                    if self._is_master and self.cmd not in [
+                    if self.is_master() and self.cmd not in [
                         'vdi_attach', 'vdi_detach',
                         'vdi_activate', 'vdi_deactivate',
                         'vdi_epoch_begin', 'vdi_epoch_end',
@@ -783,6 +780,14 @@ def scan(self, uuid):
         self._kick_gc()
         return ret
 
+    def is_master(self):
+        if 'SRmaster' not in self.dconf:
+            self._is_master = self.session is not None and util.is_master(self.session)
+        else:
+            self._is_master = self.dconf['SRmaster'] == 'true'
+
+        return self._is_master
+
     @_locked_load
     def vdi(self, uuid):
         return LinstorVDI(self, uuid)
@@ -968,7 +973,7 @@ def _synchronize_metadata_and_xapi(self):
             )
 
     def _synchronize_metadata(self):
-        if not self._is_master:
+        if not self.is_master():
             return
 
         util.SMlog('Synchronize metadata...')
@@ -1015,7 +1020,7 @@ def _load_vdis(self):
         if self._vdis_loaded:
             return
 
-        assert self._is_master
+        assert self.is_master()
 
         # We use a cache to avoid repeated JSON parsing.
         # The performance gain is not big but we can still
@@ -1494,7 +1499,7 @@ def _reconnect(self):
             controller_uri,
             self._group_name,
             repair=(
-                self._is_master and
+                self.is_master() and
                 self.srcmd.cmd in self.ops_exclusive
             ),
             logger=util.SMlog
@@ -1796,7 +1801,7 @@ def attach(self, sr_uuid, vdi_uuid):
         writable = 'args' not in self.sr.srcmd.params or \
             self.sr.srcmd.params['args'][0] == 'true'
 
-        if not attach_from_config or self.sr._is_master:
+        if not attach_from_config or self.sr.is_master():
             # We need to inflate the volume if we don't have enough place
             # to mount the VHD image. I.e. the volume capacity must be greater
             # than the VHD size + bitmap size.
@@ -1878,7 +1883,7 @@ def detach(self, sr_uuid, vdi_uuid):
                 )
 
         # We remove only on slaves because the volume can be used by the GC.
-        if self.sr._is_master:
+        if self.sr.is_master():
             return
 
         while vdi_uuid:
@@ -1899,7 +1904,7 @@ def detach(self, sr_uuid, vdi_uuid):
 
     def resize(self, sr_uuid, vdi_uuid, size):
         util.SMlog('LinstorVDI.resize for {}'.format(self.uuid))
-        if not self.sr._is_master:
+        if not self.sr.is_master():
             raise xs_errors.XenError(
                 'VDISize',
                 opterr='resize on slave not allowed'
@@ -2158,7 +2163,7 @@ def update(self, sr_uuid, vdi_uuid):
     # --------------------------------------------------------------------------
 
     def _prepare_thin(self, attach):
-        if self.sr._is_master:
+        if self.sr.is_master():
             if attach:
                 attach_thin(
                     self.session, self.sr._journaler, self._linstor,
@@ -2747,7 +2752,7 @@ def _attach_using_http_nbd(self):
 
         # 0. Fetch drbd path.
         must_get_device_path = True
-        if not self.sr._is_master:
+        if not self.sr.is_master():
             # We are on a slave, we must try to find a diskful locally.
             try:
                 volume_info = self._linstor.get_volume_info(self.uuid)
@@ -2762,7 +2767,7 @@ def _attach_using_http_nbd(self):
             must_get_device_path = hostname in volume_info.diskful
 
         drbd_path = None
-        if must_get_device_path or self.sr._is_master:
+        if must_get_device_path or self.sr.is_master():
             # If we are master, we must ensure we have a diskless
             # or diskful available to init HA.
             # It also avoid this error in xensource.log