nexenta/cinder

cinder's Issues

Improperly url-encoded filesystem names in ns5/nfs.py

The _get_share_datasets() method returns an URL-encoded filesystem name (the second part of the return tuple), but its result is frequently treated as non-encoded in many other methods.

Please note that the buggy behavior only manifests itself when you use "URL-encodable" characters in filesystem names, e.g. in our case the pool name is "data" and filesystem names are "dev/cinder" or "production/cinder". The slash in the filesystem name is the problematic char.

The first type of improper use is, for example, in _do_create_volume(). Here, the URL-encoded FS name is used in the data dictionary that is sent as JSON payload. Since there is no need to URL-encode a payload, the server reads the data as-is and fails with an error of type "Filesystem not found: data/dev%2Fcinder".

The other type of improper use is in revert_to_snapshot(). After we define the fs variable (e.g. "dev%2Fcinder"), it gets copied to fs_path. Afterwards, the url is constructed with this variable, albeit transformed with urllib.parse.quote_plus(fs_path). This results in double-encoding the forward slash and again, an API error.

I think it would be beneficial to always treat the filesystem name as a plain string and then abstract embedding that into JSON data or into the URL in a more generic way.

Cinder volume revert to snapshot: NexentaException: Waited for 120s, but pool pool1 service is still not running