CA-390937: don't clean caches until we have our locks

Doing the cleanupCache before acquiring the locks can result in a failure, in particular as part of XenCert where an sr_detach operation attempts to abort GC but the GC is not registered as running and then the umount of NFS fails as the cache cleanup still has the directory open. Also check for the SR still being attached after the active lock has been acquired. Signed-off-by: Mark Syms <[email protected]>
MarkSymsCtx · May 22, 2024 · 32562bb · 32562bb
1 parent 468de2e
commit 32562bb
Showing 1 changed file with 9 additions and 1 deletion.
diff --git a/drivers/cleanup.py b/drivers/cleanup.py
@@ -2937,6 +2937,15 @@ def _gcLoop(sr, dryRun=False, immediate=False):
     if not lockActive.acquireNoblock():
         Util.log("Another GC instance already active, exiting")
         return
+
+    # Check we're still attached after acquiring locks
+    if not sr.xapi.isPluggedHere():
+        Util.log("SR no longer attached, exiting")
+        return
+
+    # Clean up Intellicache files
+    sr.cleanupCache()
+
     # Track how many we do
     coalesced = 0
     task_status = "success"
@@ -3045,7 +3054,6 @@ def _gc(session, srUuid, dryRun=False, immediate=False):
     if not sr.gcEnabled(False):
         return
 
-    sr.cleanupCache()
     try:
         _gcLoop(sr, dryRun, immediate=immediate)
     finally: