github/kubernetes
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix descriptor lock release logic for block volume unmapDevice

Browse Source 
Fixes: #69114
This commit is contained in:
Masaki Kimura
2018-09-26 19:56:58 +00:00
parent c1ad6e961a
commit 3d808540df
4 changed files with 21 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
pkg/volume/rbd/rbd.go
View File

@@ -959,29 +959,6 @@ func (rbd *rbdDiskUnmapper) TearDownDevice(mapPath, _ string) error {
if err != nil {
return fmt.Errorf("rbd: failed to get loopback for device: %v, err: %v", device, err)
}
// Get loopback device which takes fd lock for device beofore detaching a volume from node.
// TODO: This is a workaround for issue #54108
// Currently local attach plugins such as FC, iSCSI, RBD can't obtain devicePath during
// GenerateUnmapDeviceFunc() in operation_generator. As a result, these plugins fail to get
// and remove loopback device then it will be remained on kubelet node. To avoid the problem,
// local attach plugins needs to remove loopback device during TearDownDevice().
blkUtil := volumepathhandler.NewBlockVolumePathHandler()
loop, err := volumepathhandler.BlockVolumePathHandler.GetLoopDevice(blkUtil, device)
if err != nil {
if err.Error() != volumepathhandler.ErrDeviceNotFound {
return fmt.Errorf("rbd: failed to get loopback for device: %v, err: %v", device, err)
}
glog.Warning("rbd: loopback for device: % not found", device)
} else {
if len(loop) != 0 {
// Remove loop device before detaching volume since volume detach operation gets busy if volume is opened by loopback.
err = volumepathhandler.BlockVolumePathHandler.RemoveLoopDevice(blkUtil, loop)
if err != nil {
return fmt.Errorf("rbd: failed to remove loopback :%v, err: %v", loop, err)
}
glog.V(4).Infof("rbd: successfully removed loop device: %s", loop)
}
}
err = rbd.manager.DetachBlockDisk(*rbd, mapPath)
if err != nil {