This is very neat. I'd love to know more information about it. The paper never discusses the exact notation of the annotations, for instance. And is the source code available yet?
Very neat work. I understand about ext-family and those fs that have a clear boundary of the data/metadata block -- they can be annotated and identified later. Then how about NTFS where metadata and data can mix in one block? Is there any potential challenge or same approach can still apply?
Not without interfacing the with filesystem driver. You cannot and must not modify the filesystem on disk behind the driver's back while the filesystem is mounted. The only exception to this is filesystems built with that in mind, and which implement locking protocols that these applications would then have to implement. (Sun used to have such a filesystem for clusters. Each host in the cluster would mount the same filesystems from the same storage, and did locking on-disk in order to not step on each other.)
