On Sun, Jan 03, 2021 at 11:25:40AM +0200, Andrei POPESCU wrote:
That would mean all data is written to the disk twice and would make a journaling file system twice as slow compared to a non-journaling file system; the journal is typically on the same storage.
That's almost never how it's actually implemented these days, exactly because of the hideous performance penalty. Barring a non-default option like ext3's data=journal what generally happens is that the data is written to the disk, then the *metadata* is written to the journal, and then the metadata is updated in its final location. The total amount of data written to the journal is relatively low. The primary benefit of this sort of scheme is avoiding a fsck on unclean shutdown; you may end up with old data, new data, or partially written data. Applications which need more data integrity guarantees generally implement some sort of journalling and/or use atomic filesystem operations. (E.g., write a temporary file, flush/sync, rename--that guarantees either the old file or the new file will be present, but not a partially written file.)