fix(drbd): create metadata during toggle-disk from diskless to diskful

[kvaps]

Summary

Fix DRBD metadata creation during toggle-disk operation from diskless to diskful resource.

Problem

When converting a diskless resource to diskful via toggle-disk -s <storage-pool>, several conditions prevented DRBD metadata from being created:

1. processChild() didn't process child layers for diskless resources during disk addition
2. detachVolumesIfNecessary() returned empty checkMetaData list for diskless resources
3. Metadata creation block was skipped for diskless resources
4. hasMetaData() assumed metadata exists when hasDisk() returns true, even during retry when storage exists but no metadata

This caused toggle-disk to fail with "No valid meta data found" error from drbdadm adjust, especially for LUKS encrypted resources.

Solution

Add DISK_ADD_REQUESTED/DISK_ADDING flag checks to:

• Process child layers during toggle-disk
• Include volumes in metadata check list
• Enter metadata creation block
• Force metadata verification during retry

Test plan

• Create diskless DRBD resource with LUKS encryption
• Run toggle-disk -s <storage-pool> to convert to diskful
• Verify DRBD metadata is created before drbdadm adjust
• Verify toggle-disk completes successfully
• Test retry scenario: if first attempt fails, cleanup storage and retry

[ghernadi]

Hi there!

For this PR my questions / failed reproducer attempt is basically the same as in my comment in PR 473.

I admit it might look a bit confusing why the code checks for boolean processChildren = !isDiskless || isDiskRemoving; but not || isDiskAdding. The point here is simply that during the "disk adding" state, the resource is simply not marked diskless. That is also the reason why I do not really see how your patch changes anything. Can you elaborate here?

[kvaps]

Hi Gabor!

I believe there's a misunderstanding. Let me show the issue with code references.

Important context: This is about RETRY scenarios

Both PR #473 and #474 are related to PR #472 (LUKS data offset fix). The problem appears in retry scenarios:

1. First toggle-disk attempt fails (e.g., due to LUKS offset mismatch - PR fix(luks): use fixed data offset for consistent LUKS header size #472)
2. Storage layer and LUKS device are already created
3. User retries toggle-disk
4. On retry: storage exists, but metadata doesn't exist yet
5. Without our patches: metadata is never created → "No valid meta data found"

In your test, you're testing the happy path where everything succeeds on the first attempt. That works because the two devMgrRuns happen quickly in sequence. But in retry scenarios, the state is different.

DRBD_DISKLESS flag is NOT removed during toggle-disk

From CtrlRscToggleDiskApiCallHandler.java documentation (lines 99-100):

* <li>DISKLESS, DISK_ADD_REQUESTED - the transition has been requested but not yet started</li>
* <li>DISKLESS, DISK_ADD_REQUESTED, DISK_ADDING - the peers should prepare for resource to gain disks</li>

And markDiskAdded() (lines 1453-1458) which removes DRBD_DISKLESS is called AFTER satellite completes - see line 1068 in finishOperationInTransaction().

So during satellite processing: isDrbdDiskless() == true

The actual problem: checkMetaData is empty

Look at detachVolumesIfNecessary() (DrbdLayer.java:1117):

if (!rsc.isDrbdDiskless(workerCtx) ||          // FALSE during toggle-disk!
    rsc.getStateFlags().isSet(workerCtx, Resource.Flags.DISK_REMOVING)  // FALSE
)
{
    // volumes added to checkMetaData
}
return checkMetaData;  // EMPTY LIST!

During toggle-disk: !true || false = false → block skipped → checkMetaData is empty!

Then at line 703:

for (DrbdVlmData<Resource> drbdVlmData : checkMetaData)  // empty list!
{
    if (!hasMetaData(drbdVlmData) ...) { createMetaData.add(...); }
}

Loop doesn't execute → metadata is never created!

Why your test works (but retry fails)

Your test succeeds because:

1. First devMgrRun: Storage/LUKS created, checkMetaData empty → no metadata
2. Controller receives "success", calls markDiskAdded() → removes DRBD_DISKLESS
3. Second devMgrRun: isDrbdDiskless=false → checkMetaData populated → metadata created
4. Everything happens quickly → success

But in retry scenario:

1. First attempt: Storage/LUKS created, then fails (e.g., LUKS offset issue)
2. DRBD_DISKLESS flag is NOT removed (operation didn't complete)
3. Retry: Storage already exists, but isDrbdDiskless() still true
4. checkMetaData is empty → metadata never created → "No valid meta data found"

How to reproduce

1. Create a diskless DRBD+LUKS resource
2. Start toggle-disk, but make it fail somehow (e.g., kill satellite mid-operation, or use PR fix(luks): use fixed data offset for consistent LUKS header size #472 issue with different LUKS offsets between nodes)
3. Retry toggle-disk
4. Observe "No valid meta data found" error

Our patches ensure metadata is created even during retry when storage already exists but the DRBD_DISKLESS flag is still set.

Best regards

[kvaps]

Decided not to use these changes for now, closing. Will reopen if the issues arise again.