Re: [PATCH] ext4: properly check for dirty state in ext4_inode_datasync_dirty()

[Ritesh Harjani <riteshh@linux.ibm.com>]

On 10/28/20 8:59 PM, Theodore Y. Ts'o wrote:
> On Wed, Oct 28, 2020 at 08:57:03AM +0530, Ritesh Harjani wrote:
>>
>> Well, I too noticed this yesterday while I was testing xfstests -g swap.
>> Those tests were returning _notrun, hence that could be the reason why
>> it didn't get notice in XFSTESTing from Ted.
> 
> Yeah, one of the things I discussed with Harshad is we really need a
> test that looks like generic/472, but which is in shared/NNN, and
> which unconditionally tries to use swapon for those file systems where
> swapfiles are expected to work.  This is actually the second
> regression caused by our breaking swapfile support (the other being
> the iomap bmap change), which escaped our testing because we didn't
> notice that generic/472 was skipped.

Yes, agreed this is second in a row.
So with fast-commit, swap tests returned _not_run, since
swapon syscall returned -EINVAL in _require_scratch_swapfile() itself.
This is due to some old commit in fstests to make swap tests work on
btrfs on both kernels (with and w/o support of swapon in btrfs), it
first checks in _require_scratch_swapfile() to see if swapon even works
or not. Hence it skips to run further if _require_scratch_swapfile()
fails.

Secondly with bmap to iomap interface, I guess it should pass
all tests except for case with fallocate files, which I think is
tests/generic/496. But here too it assumes that if 1st time it fails
with falloc then swapon may not be supported for that fs and hence does
_notrun.

I am actually working on this to make these swap tests return some
definitive pass or failure status. Will be sending some patches soon.
I could use your idea to add a test in shared/NNN for testing swap with
fallocate files for ext4 and xfs (for bmap to iomap ext4 regression
category of tests)

Thanks
-ritesh