Re: [PATCH v4 13/13] ext4: Remove ordered data support from ext4_writepage()

[Jan Kara <jack@suse.cz>]

On Thu 08-12-22 09:23:56, Theodore Ts'o wrote:
> On Wed, Dec 07, 2022 at 12:27:16PM +0100, Jan Kara wrote:
> > ext4_writepage() should not be called for ordered data anymore. Remove
> > support for it from the function.
> > 
> > Reviewed-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
> > Reviewed-by: Christoph Hellwig <hch@lst.de>
> > Signed-off-by: Jan Kara <jack@suse.cz>
> 
> This commit (determined via bisection) is causing a large number of
> test failures for ext4/data_journal case:
> 
> ext4/data_journal: 521 tests, 33 failures, 98 skipped, 4248 seconds
>   Failures: ext4/004 generic/012 generic/016 generic/021 generic/022 
>     generic/029 generic/030 generic/031 generic/032 generic/058 
>     generic/060 generic/061 generic/063 generic/071 generic/075 
>     generic/098 generic/112 generic/127 generic/231 generic/393 
>     generic/397 generic/404 generic/439 generic/455 generic/477 
>     generic/491 generic/567 generic/572 generic/574 generic/577 
>     generic/634 generic/639 generic/679
> 
> Many/most of these failures appear to be data plane failures.  For
> example ext4/004 is a "dump | restore" followed by a "diff -r
> $DUMP_DIR $RESTORE_DIR".  The generic/012, generic/021, generic/022
> failures are md5sum checksum failures after testing the punch hole
> operation.  The generic/029 failure are hexdump mismatches after
> calling a combination of truncate, pwrites, and mmap'ed writes, etc.
> 
> Since this is the last patch in the series, and we've already dropped
> the writepage hook (which is one of the things Christoph was going
> for), so one approach might be drop this patch from the series at
> least for this upcoming merge window.
> 
> Jan, what do you think?

Yes, please drop it and I'm sorry for the breakage. I did only very limited
testing with data=journal mode (which passed) and apparently something
unexpected breaks. I suspect some path doing writeout of committed (but not
yet checkpointed) data pages breaks. I'll have a look.

								Honza

-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR