libnvdimm/namespace: Fix reaping of invalidated block-window-namespac…

…e labels

A recent change to ndctl to attempt to reconfigure namespaces in place
uncovered a label accounting problem in block-window-type namespaces.
The ndctl "create.sh" test is able to trigger this signature:

 WARNING: CPU: 34 PID: 9167 at drivers/nvdimm/label.c:1100 __blk_label_update+0x9a3/0xbc0 [libnvdimm]
 [..]
 RIP: 0010:__blk_label_update+0x9a3/0xbc0 [libnvdimm]
 [..]
 Call Trace:
  uuid_store+0x21b/0x2f0 [libnvdimm]
  kernfs_fop_write+0xcf/0x1c0
  vfs_write+0xcc/0x380
  ksys_write+0x68/0xe0

When allocated capacity for a namespace is renamed (new UUID) the labels
with the old UUID need to be deleted. The ndctl behavior to always
destroy namespaces on reconfiguration hid this problem.

The immediate impact of this bug is limited since block-window-type
namespaces only seem to exist in the specification and not in any
shipping products. However, the label handling code is being reused for
other technologies like CXL region labels, so there is a benefit to
making sure both vertical labels sets (block-window) and horizontal
label sets (pmem) have a functional reference implementation in
libnvdimm.

Fixes: c4703ce ("libnvdimm/namespace: Fix label tracking error")
Cc: <stable@vger.kernel.org>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

drivers/nvdimm/label.c

@@ -980,6 +980,15 @@ static int __blk_label_update(struct nd_region *nd_region,
 		}
 	}
 
+	/* release slots associated with any invalidated UUIDs */
+	mutex_lock(&nd_mapping->lock);
+	list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list)
+		if (test_and_clear_bit(ND_LABEL_REAP, &label_ent->flags)) {
+			reap_victim(nd_mapping, label_ent);
+			list_move(&label_ent->list, &list);
+		}
+	mutex_unlock(&nd_mapping->lock);
+
 	/*
 	 * Find the resource associated with the first label in the set
 	 * per the v1.2 namespace specification.