@Stephan Bader: from my research, the *two* bugs are related to the cleaning of the dirty inodes. Umounting a ext4 filesystem with dirty inodes will generate a lot of IO since the journal is mixed up. The same dirty inodes problem exist when there is a lot of IO since jdb2 tries to keep the journal up to date while there are errors. Umounting is just the same process but done in a single operation for the whole filesystem.
@Stephan Bader: from my research, the *two* bugs are related to the cleaning of the dirty inodes. Umounting a ext4 filesystem with dirty inodes will generate a lot of IO since the journal is mixed up. The same dirty inodes problem exist when there is a lot of IO since jdb2 tries to keep the journal up to date while there are errors. Umounting is just the same process but done in a single operation for the whole filesystem.