fix: stop SMGC service on SR detach to prevent orphaned systemd units

Signed-off-by: Mathieu Labourier <mathieu.labourier@vates.tech>

Author: Millefeuille42

libs/sm/cleanup.py

@@ -3278,6 +3278,18 @@ def start_gc(session, sr_uuid):
     subprocess.run([__file__, '-b', '-u', sr_uuid, '-g'],
                    stdout=subprocess.PIPE, stderr=subprocess.PIPE, close_fds=True)
 
+def _gc_service_cmd(sr_uuid, action, extra_args=None):
+    """
+    Build and run the systemctl command for the GC service using util.doexec.
+    """
+    sr_uuid_esc = sr_uuid.replace("-", "\\x2d")
+    cmd=["/usr/bin/systemctl", "--quiet"]
+    if extra_args:
+        cmd.extend(extra_args)
+    cmd += [action, f"SMGC@{sr_uuid_esc}"]
+    return util.doexec(cmd)
+
+
 def start_gc_service(sr_uuid, wait=False):
     """
     This starts the templated systemd service which runs GC on the given SR UUID.
@@ -3288,13 +3300,18 @@ def start_gc_service(sr_uuid, wait=False):
     run has finished. This is used to force a run of the GC instead of just kicking it
     in the background.
     """
-    sr_uuid_esc = sr_uuid.replace("-", "\\x2d")
     util.SMlog(f"Kicking SMGC@{sr_uuid}...")
-    cmd=[ "/usr/bin/systemctl", "--quiet" ]
-    if not wait:
-        cmd.append("--no-block")
-    cmd += ["start", f"SMGC@{sr_uuid_esc}"]
-    subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, close_fds=True)
+    _gc_service_cmd(sr_uuid, "start", extra_args=None if wait else ["--no-block"])
+
+
+def stop_gc_service(sr_uuid):
+    """
+    Stops the templated systemd service which runs GC on the given SR UUID.
+    """
+    util.SMlog(f"Stopping SMGC@{sr_uuid}...")
+    (rc, _stdout, stderr) = _gc_service_cmd(sr_uuid, "stop")
+    if rc != 0:
+        util.SMlog(f"Failed to stop gc service `SMGC@{sr_uuid}`: `{stderr}`")
 
 
 def gc_force(session, srUuid, force=False, dryRun=False, lockSR=False):

libs/sm/drivers/FileSR.py

@@ -635,6 +635,7 @@ def attach(self, sr_uuid, vdi_uuid):
 
     def detach(self, sr_uuid, vdi_uuid):
         self.attached = False
+        cleanup.stop_gc_service(sr_uuid)
 
     def resize(self, sr_uuid, vdi_uuid, size):
         if not self.exists:

libs/sm/drivers/LVHDSR.py

@@ -641,6 +641,7 @@ def detach(self, uuid):
         # However, we should still delete lock files on slaves as it is the
         # only place to do so.
         self._cleanup(self.isMaster)
+        cleanup.stop_gc_service(sr_uuid)
 
     def forget_vdi(self, uuid):
         if not self.legacyMode:

libs/sm/drivers/NFSSR.py

@@ -192,6 +192,7 @@ def detach(self, sr_uuid):
             raise xs_errors.XenError('NFSUnMount', opterr=exc.errstr)
 
         self.attached = False
+        cleanup.stop_gc_service(sr_uuid)
 
     def create(self, sr_uuid, size):
         util._testHost(self.dconf['server'], NFSPORT, 'NFSTarget')

libs/sm/drivers/SMBSR.py

@@ -243,6 +243,7 @@ def detach(self, sr_uuid):
             raise xs_errors.XenError('SMBUnMount', opterr=exc.errstr)
 
         self.attached = False
+        cleanup.stop_gc_service(sr_uuid)
 
     def create(self, sr_uuid, size):
         self.__check_license()