Implement Block PVCs in our storage backend
This implements full support for Block PVCs in our Kubernetes storage backend.
The block PVCs are backed by files made available to the pods using loop devices and
have read-only and online expansion support.
This also requires a Kubernetes patch because they call losetup if block PVCs are used
with CSI to establish a form of lock on the backing block device. This lock is not
exclusive and does absolutely nothing for our use case and could get very expensive
on dense machines so I removed it.
Test Plan: Comes with E2E tests
X-Origin-Diff: phab/D746
GitOrigin-RevId: 430d3f445286c0d3498b2153df333a19f3fcab89
diff --git a/third_party/go/patches/k8s-removed-block-device-pseudo-locks.patch b/third_party/go/patches/k8s-removed-block-device-pseudo-locks.patch
new file mode 100644
index 0000000..40a9ac6
--- /dev/null
+++ b/third_party/go/patches/k8s-removed-block-device-pseudo-locks.patch
@@ -0,0 +1,73 @@
+Copyright 2020 The Monogon Project Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+
+
+From f0c9fc613c15e30453295beb25bf73763a4a82df Mon Sep 17 00:00:00 2001
+From: Lorenz Brun <lorenz@brun.one>
+Date: Mon, 29 Mar 2021 13:56:06 +0200
+Subject: [PATCH] Removed questionable block device pseudo-locks
+
+This code implements a weird sort of lock using Linux loop devices. It's not exclusive (multiple loop devices per
+backing device are supported) and the devices themselves are not accessed. The lock is also not exclusive as multiple
+loop devices per backing device are allowed. The only mention of why any of this is done is in the comment below, stating
+that it is possible that a device can be removed and another device can be attached with the same name. In Metropolis
+this is not a problem as our CSI/Provisioner setup exercises tight control over devices, thus silent removals are not
+possible.
+
+Because this code calls the losetup utility which Metropolis doesn't want to ship and also repeatedly lists loop devices
+in an inefficient manner which will cause problems with the anticipated density of VM-hosting Metropolis nodes and
+does nothing for us, let's just nuke it instead of rewriting it against //metropolis/pkg/loop.
+---
+ pkg/volume/util/util.go | 18 ------------------
+ 1 file changed, 18 deletions(-)
+
+diff --git a/pkg/volume/util/util.go b/pkg/volume/util/util.go
+index 367e05a0d1a..f53f1dcbb4e 100644
+--- a/pkg/volume/util/util.go
++++ b/pkg/volume/util/util.go
+@@ -527,17 +527,6 @@ func MapBlockVolume(
+ return fmt.Errorf("blkUtil.MapDevice failed. devicePath: %s, podVolumeMapPath:%s, volumeMapName: %s, bindMount: %v: %v",
+ devicePath, podVolumeMapPath, volumeMapName, false, mapErr)
+ }
+-
+- // Take file descriptor lock to keep a block device opened. Otherwise, there is a case
+- // that the block device is silently removed and attached another device with the same name.
+- // Container runtime can't handle this problem. To avoid unexpected condition fd lock
+- // for the block device is required.
+- _, mapErr = blkUtil.AttachFileDevice(filepath.Join(globalMapPath, string(podUID)))
+- if mapErr != nil {
+- return fmt.Errorf("blkUtil.AttachFileDevice failed. globalMapPath:%s, podUID: %s: %v",
+- globalMapPath, string(podUID), mapErr)
+- }
+-
+ return nil
+ }
+
+@@ -551,13 +540,6 @@ func UnmapBlockVolume(
+ volumeMapName string,
+ podUID utypes.UID,
+ ) error {
+- // Release file descriptor lock.
+- err := blkUtil.DetachFileDevice(filepath.Join(globalUnmapPath, string(podUID)))
+- if err != nil {
+- return fmt.Errorf("blkUtil.DetachFileDevice failed. globalUnmapPath:%s, podUID: %s: %v",
+- globalUnmapPath, string(podUID), err)
+- }
+-
+ // unmap devicePath from pod volume path
+ unmapDeviceErr := blkUtil.UnmapDevice(podDeviceUnmapPath, volumeMapName, false /* bindMount */)
+ if unmapDeviceErr != nil {
+--
+2.25.1
+