image_conversion_dir of the cinder-volume service can be filled up while performing upload-to-image of the large volume, because there is no check_available_space() being executed during this operation.
By default policy volume_extension:volume_actions:upload_image is being set to the admin_or_owner be default, so everyone can perform this operation. Also option image_upload_use_cinder_backend is being set to false.
How to reproduce:
1) Create a large volume (for example 4Tb)
2) Execute cinder upload-to-image for this volume
3) See, that image_conversion_dir is being filled up until all free space on the image_conversion_dir is gone, even if there was no enough free space on the partition with image_conversion_dir.
https://github.com/openstack/cinder/blob/stable/rocky/cinder/volume/driver.py#L866
https://github.com/openstack/cinder/blob/stable/rocky/cinder/volume/drivers/rbd.py#L1451
On the Glance side, there is a possibility to limit the maximum size of the uploaded image with image_size_cap option defaulted to 1Tb. But, for example, for rbd driver rbd export operation will be executed before glanceclient invocation and request to the Glance API, which can refuse to create a large image because of the overall limit or image_size_cap option.
This issue affects all Cinder releases at least since Pike.
Looks like this is a limitation acknowledge by Cinder https:/