From e6292d8f90b801eb65a4d30b7852c8785f099ad9 Mon Sep 17 00:00:00 2001 From: Madhu Rajanna Date: Wed, 20 Nov 2024 10:12:59 +0100 Subject: [PATCH] cephfs: take lock on targetpath on node operation We should not be dependent on the CO to ensure that it will serialize the request instead of that we need to have own internal locks to ensure that we dont do concurrent operations for same request. Signed-off-by: Madhu Rajanna (cherry picked from commit 7cfeae579f0dfdc19c115e44dd0b268cebe04633) --- internal/cephfs/nodeserver.go | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-) diff --git a/internal/cephfs/nodeserver.go b/internal/cephfs/nodeserver.go index 44c068601..a97cc0919 100644 --- a/internal/cephfs/nodeserver.go +++ b/internal/cephfs/nodeserver.go @@ -494,6 +494,13 @@ func (ns *NodeServer) NodePublishVolume( targetPath := req.GetTargetPath() volID := fsutil.VolumeID(req.GetVolumeId()) + if acquired := ns.VolumeLocks.TryAcquire(targetPath); !acquired { + log.ErrorLog(ctx, util.TargetPathOperationAlreadyExistsFmt, targetPath) + + return nil, status.Errorf(codes.Aborted, util.TargetPathOperationAlreadyExistsFmt, targetPath) + } + defer ns.VolumeLocks.Release(targetPath) + volOptions := &store.VolumeOptions{} defer volOptions.Destroy() @@ -506,9 +513,6 @@ func (ns *NodeServer) NodePublishVolume( return nil, status.Errorf(codes.Internal, "failed to create mounter for volume %s: %v", volID, err.Error()) } - // Considering kubelet make sure the stage and publish operations - // are serialized, we dont need any extra locking in nodePublish - if err = util.CreateMountPoint(targetPath); err != nil { log.ErrorLog(ctx, "failed to create mount point at %s: %v", targetPath, err) @@ -599,12 +603,17 @@ func (ns *NodeServer) NodeUnpublishVolume( return nil, err } - // considering kubelet make sure node operations like unpublish/unstage...etc can not be called - // at same time, an explicit locking at time of nodeunpublish is not required. targetPath := req.GetTargetPath() + volID := req.GetVolumeId() + if acquired := ns.VolumeLocks.TryAcquire(targetPath); !acquired { + log.ErrorLog(ctx, util.TargetPathOperationAlreadyExistsFmt, targetPath) + + return nil, status.Errorf(codes.Aborted, util.TargetPathOperationAlreadyExistsFmt, targetPath) + } + defer ns.VolumeLocks.Release(targetPath) // stop the health-checker that may have been started in NodeGetVolumeStats() - ns.healthChecker.StopChecker(req.GetVolumeId(), targetPath) + ns.healthChecker.StopChecker(volID, targetPath) isMnt, err := util.IsMountPoint(ns.Mounter, targetPath) if err != nil {