Hello,
I discovered a potential bug in the truncate mechanism.
If the system crashes while an entry is in the truncate list, the operation is repeated during recovery.
During this repeated truncate operation, the inode timestamps are set to the current time and not the time when the operation occurred.
Therefore, the "recovered" state is neither the original nor the intended final state which imo is invalid behavior for a file system and should be addressed.
One fix would be to add the new mtime and ctime of the inode to the truncate list which is then recovered.
My implementation also uses a marker field to distinguish between the old and new version of the truncate list entries.
Hello, I discovered a potential bug in the truncate mechanism. If the system crashes while an entry is in the truncate list, the operation is repeated during recovery. During this repeated truncate operation, the inode timestamps are set to the current time and not the time when the operation occurred. Therefore, the "recovered" state is neither the original nor the intended final state which imo is invalid behavior for a file system and should be addressed.
One fix would be to add the new mtime and ctime of the inode to the truncate list which is then recovered. My implementation also uses a marker field to distinguish between the old and new version of the truncate list entries.