Re: [RFC] jbd2: reduce the number of writes when commiting a transacation

[Ted Ts'o <tytso@mit.edu>]

On Tue, Apr 24, 2012 at 11:57:09PM +0200, Jan Kara wrote:
>   Also currently the async commit code has essentially unfixable bugs in
> handling of cache flushes as I wrote in
> http://www.spinics.net/lists/linux-ext4/msg30452.html. Because data blocks
> are not part of journal checksum, it can happen with async commit code that
> data is not safely on disk although transaction is completely committed. So
> async commit code isn't really safe to use unless you are fine with
> exposure of uninitialized data...

With the old journal checksum, the data blocks *are* part of the
journal checksum.  That's not the reason I haven't enabled it as a
default (even though it would close to double fs_mark benchmarks).
The main issue is that e2fsck doesn't deal intelligently if some
commit *other* than the last one has a bad intelligent.

With the new journal checksum patches, each individual data block has
its own checksum, so we don't need to discard the entire commit;
instead we can just drop the individual block(s) that have a bad
checksum, and then force a full fsck run afterwards.

						- Ted