fix(cleanup): remove running file before relink

Sometime, the VDI activation code in blktap2.py would wait on coalesce
being stopped while the GC code would be waiting for the activation
being done to continue with the relink, causing a deadlock.
We remove the coalesce running file before the relink so that it
continue as before, the activation will proceed then the relink will
run and refresh the tapdisk with the new parent.
We let the relink run everytime, it should be almost a noop if tapdisk
has made the parent change itself.

Signed-off-by: Damien Thenot <damien.thenot@vates.tech>

Author: Nambrok

drivers/cleanup.py

@@ -2480,7 +2480,6 @@ def _delete_running_file(self, vdi: VDI):
         os.unlink(self._gc_running_file(vdi))
 
     def _coalesce(self, vdi: VDI):
-        skipRelink = False
         if self.journaler.get(vdi.JRN_RELINK, vdi.uuid):
             # this means we had done the actual coalescing already and just
             # need to finish relinking and/or refreshing the children
@@ -2508,42 +2507,36 @@ def _coalesce(self, vdi: VDI):
                     #Leaf opened on another host, we need to call online coalesce
                     util.SMlog("Remote coalesce for {}".format(vdi.path))
                     vdi._doCoalesceOnHost(list(host_refs)[0])
-                    skipRelink = True
                 else:
                     util.SMlog("Offline coalesce for {}".format(vdi.path))
                     vdi._doCoalesce()
             except Exception as e:
                 util.SMlog("EXCEPTION while coalescing: {}".format(e))
                 self._delete_running_file(vdi)
                 raise
-            """
-            vdi._doCoalesce will call vdi._coalesceCowImage (after doing other things).
-            It will then call VDI._doCoalesceCowImage in a runAbortable context
-            """
+
             self.journaler.remove(vdi.JRN_COALESCE, vdi.uuid)
+            self._delete_running_file(vdi)
 
             util.fistpoint.activate("LVHDRT_before_create_relink_journal", self.uuid)
 
             # we now need to relink the children: lock the SR to prevent ops
             # like SM.clone from manipulating the VDIs we'll be relinking and
             # rescan the SR first in case the children changed since the last
             # scan
-            if not skipRelink:
-                self.journaler.create(vdi.JRN_RELINK, vdi.uuid, "1")
+            self.journaler.create(vdi.JRN_RELINK, vdi.uuid, "1")
 
-        if not skipRelink: #TODO: we might want to let relink happen for VDI not currently in use
-            self.lock()
-            try:
-                vdi.parent._tagChildrenForRelink()
-                self.scan()
-                vdi._relinkSkip()
-            finally:
-                self.unlock()
-                # Reload the children to leave things consistent
-                vdi.parent._reloadChildren(vdi)
-            self.journaler.remove(vdi.JRN_RELINK, vdi.uuid)
+        self.lock()
+        try:
+            vdi.parent._tagChildrenForRelink()
+            self.scan()
+            vdi._relinkSkip() #TODO: We could check if the parent is already the right one before doing the relink, or we could do the relink a second time, it doesn't seem to cause issues
+        finally:
+            self.unlock()
+            # Reload the children to leave things consistent
+            vdi.parent._reloadChildren(vdi)
+        self.journaler.remove(vdi.JRN_RELINK, vdi.uuid)
 
-        self._delete_running_file(vdi)
         self.deleteVDI(vdi)
 
     class CoalesceTracker: