openstack/cinder
Code Issues Proposed changes

Fix remotefs clone volume locking

Browse Source 
Any method in the remotefs/nfs code that manipulates
the qcow2 snapshot chain must be run separately
from other methods that may touch this snapshot chain.

This code intended to do this with a lock on the
volume id, but it unintentionally locked only on
the destination volume id rather than the source
volume id where the snapshots are.

This causes concurrent clone operations to fail in
the NFS driver.  This patch fixes this in the
RemoteFSSnapDriverDistributed class which fixes the
NFS driver and a handful of others.

A follow up patch should be applied for the
RemoteFSSnapDriver class with a similar change, but as
that class is only used by one driver (which I can't
test), this patch only adds a TODO for that change.

Related-Bug: #1840712
Related-Bug: #1852449
Closes-Bug: #1851512

Change-Id: I64e041feaeb50c95808da46a34f334a9985018a8
This commit is contained in:
Eric Harney 2019-11-06 09:22:27 -05:00
parent bd80bc7270
commit 7cc2e402f9
2 changed files with 15 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
cinder/volume/drivers/remotefs.py
View File

@ -1654,6 +1654,7 @@ class RemoteFSSnapDriver(RemoteFSSnapDriverBase):
def create_volume_from_snapshot(self, volume, snapshot):
return self._create_volume_from_snapshot(volume, snapshot)
# TODO: should be locking on src_vref id -- bug #1852449
@locked_volume_id_operation
def create_cloned_volume(self, volume, src_vref):
"""Creates a clone of the specified volume."""
@ -1696,6 +1697,8 @@ class RemoteFSSnapDriverDistributed(RemoteFSSnapDriverBase):
def create_volume_from_snapshot(self, volume, snapshot):
return self._create_volume_from_snapshot(volume, snapshot)
# lock the source volume id first
@coordination.synchronized('{self.driver_prefix}-{src_vref.id}')
@coordination.synchronized('{self.driver_prefix}-{volume.id}')
def create_cloned_volume(self, volume, src_vref):
"""Creates a clone of the specified volume."""

12
releasenotes/notes/fix-remotefs-clone-volume-locking-385e740d4a5a813b.yaml Normal file
View File

@ -0,0 +1,12 @@
---
fixes:
- |
An incorrect lock in the remotefs code, which is used for the NFS driver,
and other similar drivers, resulted in concurrent clone volume operations
failing. create_cloned_volume now locks on the source volume id, meaning
multiple clone operations from the same source volume are serialized.
A lock in the volume manager flow generally prevents this on normal clone
volume operations, but this clone method in the driver is called for
operations such as cloning from the cinder image-volume cache or cloning
from a cinder backend used as a glance store.