rbd: CephCSI cannot determine correct clone depth in certain case

[Rakshith-R]

Describe the bug

If parent PVC of snapshot/ restore/clone pvc is deleted,
then

ceph-csi/internal/rbd/rbd_util.go

Line 700 in 35eb347

func (ri *rbdImage) getCloneDepth(ctx context.Context) (uint, error) {

will not work as intended.
Therefore, the clone depth returned is not correct since this function requires all parent images in chain to be present in cluster(deleted parent images will be trash).

One side affect is the issue described here rook/rook#12312.
Creating a chain of clone/snapshot+restore and deleting parent snapshot and PVC renders the final child PVC
to be unmountable.

Environment details

• Image/version of Ceph CSI driver : All supported cephcsi versions.
• Helm chart version : -
• Kernel version : -
• Mounter used for mounting PVC (for cephFS its fuse or kernel. for rbd its
  krbd or rbd-nbd) : krbd
• Kubernetes cluster version : all
• Ceph cluster version : all

Steps to reproduce

Steps to reproduce the behavior:

1. Create a chain of clones/ snapshot+restores
2. Delete parent snapshot/clone immediately after child pvc/snapshot creation
3. Try to mount child PVC to a pod.

Actual results

rbd map fails with the below error:

E0531 18:09:09.891563 15892 utils.go:210] ID: 198 Req-ID: 0001-0009-rook-ceph-0000000000000001-8b978541-8495-4c20-bcab-0a42fa927b5a GRPC error: rpc error: code = Internal desc = rbd: map failed with error an error (exit status 22) occurred while running rbd args: [--id csi-rbd-node -m 10.110.0.127:6789,10.109.223.145:6789,10.108.43.136:6789 --keyfile=***stripped*** map replicapool/csi-vol-8b978541-8495-4c20-bcab-0a42fa927b5a --device-type krbd --options noudev], rbd error output: rbd: sysfs write failed rbd: map failed: (22) Invalid argument

Expected behavior

No error.

Workaround

Flatten the child PVC image manually.

Possible Solution

From CephCSI point of view, we have no other to determine clone depth.
We need a API change possibly in rbd image info providing us with clone depth from ceph.

cc @ceph/ceph-csi-contributors

[Rakshith-R]

Steps to reproduce:

1. Create [Restore]PVC
2. Create Snapshot
3. Delete parent PVC and restore PVC
4. Delete snapshot
5. Repeat step 2 (untill there is 17 images in trash rbd trash ls <pool_name>| wc -l)
6. Mount child PVC to a pod

Nodeplugin logs:

I0724 12:32:42.749654    7461 utils.go:195] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 GRPC call: /csi.v1.Node/NodeStageVolume
I0724 12:32:42.749810    7461 utils.go:206] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 GRPC request: {"secrets":"***stripped***","staging_target_path":"/var/lib/kubelet/plugins/kubernetes.io/csi/rook-ceph.rbd.csi.ceph.com/86f184afd78d2e10464567a1eaa6d77fd2b52867f9c4f76350333f39fc3dc557/globalmount","volume_capability":{"AccessType":{"Mount":{"fs_type":"ext4"}},"access_mode":{"mode":7}},"volume_context":{"clusterID":"rook-ceph","imageFeatures":"layering","imageFormat":"2","imageName":"csi-vol-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6","journalPool":"replicapool","pool":"replicapool","storage.kubernetes.io/csiProvisionerIdentity":"1690201077472-8271-rook-ceph.rbd.csi.ceph.com"},"volume_id":"0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6"}
I0724 12:32:42.750671    7461 omap.go:88] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 got omap values: (pool="replicapool", namespace="", name="csi.volume.c36ef26f-d656-4f3b-9b12-d54b9d9c98a6"): map[csi.imageid:12f02815b4b2 csi.imagename:csi-vol-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 csi.volname:pvc-475deef4-7f81-4750-b75f-e1e1b07a538f csi.volume.owner:rook-ceph]
I0724 12:32:43.170381    7461 rbd_util.go:352] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 checking for ImageFeatures: [layering operations]
I0724 12:32:43.198124    7461 cephcmds.go:105] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 command succeeded: rbd [device list --format=json --device-type krbd]
I0724 12:32:43.353638    7461 rbd_attach.go:419] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 rbd: map mon 192.168.39.232:6789
I0724 12:32:43.669135    7461 cephcmds.go:98] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 an error (exit status 22) occurred while running rbd args: [--id csi-rbd-node -m 192.168.39.232:6789 --keyfile=***stripped*** map replicapool/csi-vol-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 --device-type krbd --options noudev]
W0724 12:32:43.669173    7461 rbd_attach.go:468] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 rbd: map error an error (exit status 22) occurred while running rbd args: [--id csi-rbd-node -m 192.168.39.232:6789 --keyfile=***stripped*** map replicapool/csi-vol-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 --device-type krbd --options noudev], rbd output: rbd: sysfs write failed
rbd: map failed: (22) Invalid argument
E0724 12:32:43.669309    7461 utils.go:210] ID: 90 Req-ID: 0001-0009-rook-ceph-0000000000000001-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 GRPC error: rpc error: code = Internal desc = rbd: map failed with error an error (exit status 22) occurred while running rbd args: [--id csi-rbd-node -m 192.168.39.232:6789 --keyfile=***stripped*** map replicapool/csi-vol-c36ef26f-d656-4f3b-9b12-d54b9d9c98a6 --device-type krbd --options noudev], rbd error output: rbd: sysfs write failed
rbd: map failed: (22) Invalid argument

Dmesg logs :
dmesg.txt

images in trash:

bash-4.4$ rbd trash ls replicapool | wc -l
18

cc @ceph/ceph-csi-contributors

@idryomov
Can you please provide your inputs on this issue?

[idryomov]

@idryomov
Can you please provide your inputs on this issue?

The kernel client is failing to map an image because it has more than 16 images in the parent chain, most or all of which are in the trash:

[ 3327.965645] rbd: id 12f02b4046bf: unable to get image name
[ 3327.966827] rbd: id 12f041960b82: unable to get image name
[ 3327.968113] rbd: id 12f05858ad42: unable to get image name
[ 3327.969280] rbd: id 12f0c68079be: unable to get image name
[ 3327.970612] rbd: id 12f0cbf47a6: unable to get image name
[ 3327.972276] rbd: id 12f0193d54ed: unable to get image name
[ 3327.973473] rbd: id 115e71c81440: unable to get image name
[ 3327.974756] rbd: id 115e5c165e0c: unable to get image name
[ 3327.976137] rbd: id 115ef49affa9: unable to get image name
[ 3327.977415] rbd: id 115efd7ba12b: unable to get image name
[ 3327.978439] rbd: id 115e63c98b1a: unable to get image name
[ 3327.979750] rbd: id 115e1f145b8: unable to get image name
[ 3327.981078] rbd: id 115e919877e0: unable to get image name
[ 3327.982162] rbd: id 115e2403b185: unable to get image name
[ 3327.983619] rbd: id 115e5a5abd49: unable to get image name
[ 3327.984825] rbd: id 115e3d321ae6: unable to get image name
[ 3327.985458] rbd: parent chain is too long (17)

For further inputs, please translate the steps to reproduce to rbd commands. While doing that, you would likely see how/where the parent images build-up occurs.

[Rakshith-R]

@idryomov
Can you please provide your inputs on this issue?

The kernel client is failing to map an image because it has more than 16 images in the parent chain, most or all of which are in the trash:

Yup, the images being in trash is expected

For further inputs, please translate the steps to reproduce to rbd commands. While doing that, you would likely see how/where the parent images build-up occurs.

Equivalent rbd commands can be found here https://github.com/ceph/ceph-csi/blob/devel/docs/design/proposals/rbd-snap-clone.md.
The build is happening because CephCSI cannot determine the clone chain depth appropriately and therefore is not [adding task to/] flattening the image.

In short,

• CephCSI moves deleted images to trash and adds a task to remove it.
• CephCSI determines clone chain depth by traversing the image chain
  

  ceph-csi/internal/rbd/rbd_util.go

  Line 700 in 35eb347

  func (ri *rbdImage) getCloneDepth(ctx context.Context) (uint, error) {

  • This method fails if a parent PVC/Snapshot is deleted and image is in trash
  • Each kubernetes PVC and snapshot are meant to be independent of parent or child pvc/snapshots. Therefore, we cannot impose any restriction here.
  • We need a way to determine the chain depth in another way, maybe a field in rbd image info ?

[idryomov]

I see. The way getCloneDepth() just returns current depth when it encounters an empty image name looks wrong, as is bailing on ErrImageNotFound error.

When getting parent details via rbd_get_parent() API, the returned rbd_linked_image_spec_t struct has trash and image_id fields. If trash true, it means the parent image is in the trash. In that case, the image ID from image_id can be used to open the parent image with rbd_open_by_id() API, thus moving on to the next iteration.

[Rakshith-R]

I see. The way getCloneDepth() just returns current depth when it encounters an empty image name looks wrong, as is bailing on ErrImageNotFound error.

When getting parent details via rbd_get_parent() API, the returned rbd_linked_image_spec_t struct has trash and image_id fields. If trash true, it means the parent image is in the trash. In that case, the image ID from image_id can be used to open the parent image with rbd_open_by_id() API, thus moving on to the next iteration.

Thanks !,

Currently go-ceph is not reading and exporting those details,
https://github.com/ceph/go-ceph/blob/ce4031e218edce2afdc79a714b7123a74b1e3c78/rbd/snapshot_nautilus.go#L106-L109

We'll need to add it there before using it at cephcsi