Ceph » rgw

The issue was found in 12.2.2.

The step to reproduce the issue is that

1. Create a bucket (Say A-bucket) and upload Object X through multipart upload in SSE-C.
2. Create another bucket (Say B-Bucket) but B-Bucket and A-Bucket have different data pools
3. Use s3cmd copy X in A-bucket to X_B in B-bucket
4. Download X_B from B-bucket and notice that the data is corrupted

Per check through the code, the rationale is that

• The encrypted data in X_B was in Multipart manner since it is copied from X
• The mannifest in X_B has been changed to Atomic manner by the copy operation
• The GET operation follow the X_B's manifest to to decrypt the data in atomic way while the data is in fact encrypted in multipart way

It seems that current implementation does not support this kind of copy operation.
Before we figure out a comprehensive implementation to make it fully functional (maybe the compression manner instead of relying on the volatile manifest is preferred?), I think we should reject the cross-pool copy operation for multipart SSE-C objects explicitly instead of failing silently. Any suggestion?