mgr/volumes: make clone state transition to fail when UUID dir is missing...

for the source subvolume. Also add tests for the same.

Fixes: https://tracker.ceph.com/issues/72957
Signed-off-by: Rishabh Dave <ridave@redhat.com>

diff --git a/qa/tasks/cephfs/test_volumes.py b/qa/tasks/cephfs/test_volumes.py
index ad7636af75e175743a7143ec6df2b4c4e8124be4..3bbd58ec99f8a692fb54fb7c412e4efabf7836c5 100644 (file)
--- a/qa/tasks/cephfs/test_volumes.py
+++ b/qa/tasks/cephfs/test_volumes.py
@@ -10839,3 +10839,26 @@ class TestCorruptedSubvolumes(TestVolumesHelper):
 
         # cleanup
         self.run_ceph_cmd(f'fs subvolume rm {self.volname} {sv1} --force')
+
+    def test_clone_when_src_subvol_has_missing_UUID_dir(self):
+        sv1 = 'sv1'
+        ss1 = 'ss1'
+        c1 = 'c1'
+
+        self.run_ceph_cmd(f'fs subvolume create {self.volname} {sv1}')
+        sv_path = self.get_ceph_cmd_stdout('fs subvolume getpath '
+                                           f'{self.volname} {sv1}').strip()
+        sv_path = sv_path[1:]
+        self.run_ceph_cmd(f'fs subvolume snapshot create {self.volname} {sv1} {ss1}')
+        self.run_ceph_cmd('config set mgr mgr/volumes/snapshot_clone_delay 2')
+        self.run_ceph_cmd(f'fs subvolume snapshot clone {self.volname} {sv1} {ss1} {c1}')
+
+        self.mount_a.run_shell(f'sudo rmdir {sv_path}', omit_sudo=False)
+
+        time.sleep(2)
+        self._wait_for_clone_to_fail(c1, timo=20)
+
+        # cleanup
+        self.run_ceph_cmd(f'fs subvolume snapshot rm {self.volname} {sv1} {ss1} '
+                           '--force')
+        self.run_ceph_cmd(f'fs subvolume rm {self.volname} {sv1} --force')

diff --git a/src/pybind/mgr/volumes/fs/async_cloner.py b/src/pybind/mgr/volumes/fs/async_cloner.py
index c22c6ee481634e88ff6d736a3f2e4c7908b07d8f..3522c67d443b3616f8bd9323d89f3cf2de2ca1b9 100644 (file)
--- a/src/pybind/mgr/volumes/fs/async_cloner.py
+++ b/src/pybind/mgr/volumes/fs/async_cloner.py
@@ -228,12 +228,12 @@ def do_clone(fs_client, volspec, volname, groupname, subvolname, should_cancel):
             set_quota_on_clone(fs_handle, (subvol0, subvol1, subvol2))
 
 def update_clone_failure_status(fs_client, volspec, volname, groupname, subvolname, ve):
-    with open_clone_subvol_pair_in_vol(fs_client, volspec, volname, groupname,
-            subvolname, lockless=False) as (subvol0, subvol1, subvol2):
+    with open_at_volume(fs_client, volspec, volname, groupname, subvolname,
+                        SubvolumeOpType.CLONE_INTERNAL) as clone:
         if ve.errno == -errno.EINTR:
-            subvol0.add_clone_failure(-ve.errno, "user interrupted clone operation")
+            clone.add_clone_failure(-ve.errno, "user interrupted clone operation")
         else:
-            subvol0.add_clone_failure(-ve.errno, ve.error_str)
+            clone.add_clone_failure(-ve.errno, ve.error_str)
 
 def log_clone_failure(volname, groupname, subvolname, ve):
     if ve.errno == -errno.EINTR:
@@ -261,7 +261,7 @@ def handle_clone_failed(fs_client, volspec, volname, index, groupname, subvolnam
     try:
         # detach source but leave the clone section intact for later inspection
         with open_clone_subvol_pair_in_vol(fs_client, volspec, volname, groupname,
-                subvolname) as (subvol0, subvol1, subvol2):
+                subvolname, failed=True) as (subvol0, subvol1, subvol2):
             subvol1.detach_snapshot(subvol2, index)
     except (MetadataMgrException, VolumeException) as e:
         log.error("failed to detach clone from snapshot: {0}".format(e))

diff --git a/src/pybind/mgr/volumes/fs/operations/subvolume.py b/src/pybind/mgr/volumes/fs/operations/subvolume.py
index 4182e4f75714d7c41a98c53ef29f24cbc511c4ba..885eec16e3fb6f0e6ef097643c2f80f5d29a6657 100644 (file)
--- a/src/pybind/mgr/volumes/fs/operations/subvolume.py
+++ b/src/pybind/mgr/volumes/fs/operations/subvolume.py
@@ -102,7 +102,8 @@ def open_subvol_in_group(mgr, vol_handle, vol_spec, group_name, subvol_name,
 
 @contextmanager
 def open_clone_subvol_pair_in_vol(vc, vol_spec, vol_name, group_name,
-                                  subvol_name, lockless=False):
+                                  subvol_name, lockless=False, failed=False):
+
     with open_subvol_in_vol(vc, vol_spec, vol_name, group_name, subvol_name,
                             SubvolumeOpType.CLONE_INTERNAL, lockless) \
                             as (vol_handle, _, dst_subvol):
@@ -113,9 +114,14 @@ def open_clone_subvol_pair_in_vol(vc, vol_spec, vol_name, group_name,
             # use the same subvolume to avoid metadata overwrites
             yield (dst_subvol, dst_subvol, src_snap_name)
         else:
+            if failed:
+                op_type = SubvolumeOpType.CLONE_FAILED
+            else:
+                op_type = SubvolumeOpType.CLONE_SOURCE
+
             with open_subvol_in_group(vc.mgr, vol_handle, vol_spec,
                                       src_group_name, src_subvol_name,
-                                      SubvolumeOpType.CLONE_SOURCE) \
+                                      op_type) \
                                       as src_subvol:
                 yield (dst_subvol, src_subvol, src_snap_name)
 

diff --git a/src/pybind/mgr/volumes/fs/operations/template.py b/src/pybind/mgr/volumes/fs/operations/template.py
index 2189e20ce13eb5fba47b81fb736987ad6ba2cce4..7b1cef1907cfa58242e44c6619b6677bd40e3d09 100644 (file)
--- a/src/pybind/mgr/volumes/fs/operations/template.py
+++ b/src/pybind/mgr/volumes/fs/operations/template.py
@@ -58,6 +58,7 @@ class SubvolumeOpType(Enum):
     CLONE_CREATE          = 'clone-create'
     CLONE_STATUS          = 'clone-status'
     CLONE_CANCEL          = 'clone-cancel'
+    CLONE_FAILED          = 'clone-failed'
     CLONE_INTERNAL        = 'clone_internal'
     ALLOW_ACCESS          = 'allow-access'
     DENY_ACCESS           = 'deny-access'

diff --git a/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v2.py b/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v2.py
index 164448133e0f85b77349f3c242e4906f9540f57f..e4aa2591bf99cfc44715836ca8a311226fc6dda1 100644 (file)
--- a/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v2.py
+++ b/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v2.py
@@ -288,7 +288,8 @@ class SubvolumeV2(SubvolumeV1):
                 SubvolumeOpType.SNAP_INFO,
                 SubvolumeOpType.SNAP_PROTECT,
                 SubvolumeOpType.SNAP_UNPROTECT,
-                SubvolumeOpType.CLONE_SOURCE
+                SubvolumeOpType.CLONE_SOURCE,
+                SubvolumeOpType.CLONE_FAILED
             }
 
         return {SubvolumeOpType.REMOVE_FORCE,
@@ -296,7 +297,8 @@ class SubvolumeV2(SubvolumeV1):
                 SubvolumeOpType.CLONE_STATUS,
                 SubvolumeOpType.CLONE_CANCEL,
                 SubvolumeOpType.CLONE_INTERNAL,
-                SubvolumeOpType.CLONE_SOURCE}
+                SubvolumeOpType.CLONE_SOURCE,
+                SubvolumeOpType.CLONE_FAILED}
 
     def open(self, op_type):
         if not isinstance(op_type, SubvolumeOpType):
@@ -343,6 +345,11 @@ class SubvolumeV2(SubvolumeV1):
                           f"path '{subvol_path}' for subvolume "
                           f"{self.subvolname}'")
                 return
+            elif op_type == SubvolumeOpType.CLONE_FAILED:
+                log.debug('since clone failed, letting that register in .meta '
+                          'file and ignoring missing subvolume path '
+                          f'{subvol_path} for subvolume {self.subvolname}')
+                return
             log.debug("missing subvolume path '{0}' for subvolume '{1}'".format(subvol_path, self.subvolname))
             raise VolumeException(-errno.ENOENT, "mount path missing for subvolume '{0}'".format(self.subvolname))
         except cephfs.Error as e: