Merge PR ceph#59676 into main

* refs/pull/59676/head:

Reviewed-by: Rishabh Dave <[email protected]>
Reviewed-by: Kotresh Hiremath Ravishankar <[email protected]>
Reviewed-by: Venky Shankar <[email protected]>

Author: vshankar

qa/suites/fs/full/tasks/mgr-osd-full.yaml

@@ -34,3 +34,8 @@ tasks:
     clients:
       client.0:
         - fs/full/subvolume_ls.sh
+- workunit:
+    cleanup: true
+    clients:
+      client.0:
+        - fs/full/subvol_rm_retained_snap_when_cluster_is_full.sh

qa/workunits/fs/full/subvol_rm_retained_snap_when_cluster_is_full.sh

@@ -0,0 +1,67 @@
+#!/usr/bin/env bash
+set -ex
+
+# Test that command 'ceph fs subvolume rm --retained-snapshots' fails when the
+# OSD is full.
+#
+# A subvolume is created on a cluser with OSD size 2GB and a 1GB file is written on
+# the subvolume. The OSD size is then set to below 500MB so that OSD is treated as
+# full. Now the subvolume is removed but snapshots are retained. 
+
+expect_failure() {
+	if "$@"; then return 1; else return 0; fi
+}
+
+ceph fs subvolume create cephfs sub_0
+subvol_path=$(ceph fs subvolume getpath cephfs sub_0 2>/dev/null)
+
+echo "Printing system disk usages for host as well Ceph before writing on subvolume"
+df -h
+ceph osd df
+
+sudo dd if=/dev/urandom of=$CEPH_MNT$subvol_path/1GB_file-1 status=progress bs=1M count=1000
+
+ceph osd set-full-ratio 0.2
+ceph osd set-nearfull-ratio 0.16
+ceph osd set-backfillfull-ratio 0.18
+
+timeout=30
+while [ $timeout -gt 0 ]
+do
+  health=$(ceph health detail)
+  [[ $health = *"OSD_FULL"* ]] && echo "OSD is full" && break
+  echo "Wating for osd to be full: $timeout"
+  sleep 1
+  let "timeout-=1"
+done
+
+echo "Printing disk usage for host as well as Ceph after OSD ratios have been set"
+df -h
+ceph osd df
+
+#Take snapshot
+ceph fs subvolume snapshot create cephfs sub_0 snap_0
+
+#Delete subvolume with retain snapshot fails
+expect_failure ceph fs subvolume rm cephfs sub_0 --retain-snapshots
+
+#Validate subvolume is not deleted
+ceph fs subvolume info cephfs sub_0
+
+# Validate config file is not truncated and GLOBAL section exists
+sudo grep "GLOBAL" $CEPH_MNT/volumes/_nogroup/sub_0/.meta
+
+# Hard cleanup
+sudo rmdir $CEPH_MNT/volumes/_nogroup/sub_0/.snap/snap_0
+sudo rm -rf $CEPH_MNT/volumes/_nogroup/sub_0
+
+#Reset the ratios to original values for the sake of rest of tests
+ceph osd set-full-ratio 0.95
+ceph osd set-nearfull-ratio 0.95
+ceph osd set-backfillfull-ratio 0.95
+
+echo "Printing disk usage for host as well as Ceph since test has been finished"
+df -h
+ceph osd df
+
+echo OK

src/pybind/mgr/volumes/fs/operations/versions/subvolume_v2.py

@@ -341,14 +341,14 @@ def open(self, op_type):
         except cephfs.Error as e:
             raise VolumeException(-e.args[0], e.args[1])
 
-    def trash_incarnation_dir(self):
+    def trash_incarnation_dir(self, subvol_path):
         """rename subvolume (uuid component) to trash"""
         self.create_trashcan()
         try:
-            bname = os.path.basename(self.path)
+            bname = os.path.basename(subvol_path)
             tpath = os.path.join(self.trash_dir, bname)
-            log.debug("trash: {0} -> {1}".format(self.path, tpath))
-            self.fs.rename(self.path, tpath)
+            log.debug(f"trash: {subvol_path} -> {tpath}")
+            self.fs.rename(subvol_path, tpath)
             self._link_dir(tpath, bname)
         except cephfs.Error as e:
             raise VolumeException(-e.args[0], e.args[1])
@@ -378,13 +378,20 @@ def remove(self, retainsnaps=False, internal_cleanup=False):
                 self.auth_mdata_mgr.delete_subvolume_metadata_file(self.group.groupname, self.subvolname)
                 return
         if self.state != SubvolumeStates.STATE_RETAINED:
-            self.trash_incarnation_dir()
-            self.metadata_mgr.remove_section(MetadataManager.USER_METADATA_SECTION)
-            self.metadata_mgr.update_global_section(MetadataManager.GLOBAL_META_KEY_PATH, "")
-            self.metadata_mgr.update_global_section(MetadataManager.GLOBAL_META_KEY_STATE, SubvolumeStates.STATE_RETAINED.value)
-            self.metadata_mgr.flush()
-            # Delete the volume meta file, if it's not already deleted
-            self.auth_mdata_mgr.delete_subvolume_metadata_file(self.group.groupname, self.subvolname)
+            try:
+                # save subvol path for later use(renaming subvolume to trash) before deleting path section from .meta
+                subvol_path = self.path
+                self.metadata_mgr.update_global_section(MetadataManager.GLOBAL_META_KEY_PATH, "")
+                self.metadata_mgr.update_global_section(MetadataManager.GLOBAL_META_KEY_STATE, SubvolumeStates.STATE_RETAINED.value)
+                self.metadata_mgr.remove_section(MetadataManager.USER_METADATA_SECTION)
+                self.metadata_mgr.flush()
+                self.trash_incarnation_dir(subvol_path)
+                # Delete the volume meta file, if it's not already deleted
+                self.auth_mdata_mgr.delete_subvolume_metadata_file(self.group.groupname, self.subvolname)
+            except MetadataMgrException as e:
+                log.error(f"failed to write config: {e}")
+                raise VolumeException(e.args[0], e.args[1])
+
 
     def info(self):
         if self.state != SubvolumeStates.STATE_RETAINED: