On Wed, Jul 13, 2022 at 10:33:05AM -0700, Darrick J. Wong wrote: > On Thu, Jul 14, 2022 at 01:04:26AM +0800, Zorro Lang wrote: > > On Tue, Jul 12, 2022 at 05:57:07PM -0700, Darrick J. Wong wrote: > > > From: Darrick J. Wong <djwong@xxxxxxxxxx> > > > > > > Skip the generic fpunch tests on a file when the file's allocation unit > > > size is not congruent with the proposed testing operations. > > > > > > This can be the case when we're testing reflink and fallocate on the XFS > > > realtime device. For those configurations, the file allocation unit is > > > a realtime extent, which can be any integer multiple of the block size. > > > If the request length isn't an exact multiple of the allocation unit > > > size, reflink and fallocate will fail due to alignment issues, so > > > there's no point in running these tests. > > > > > > Assuming this edgecase configuration of an edgecase feature is > > > vanishingly rare, let's just _notrun the tests instead of rewriting a > > > ton of tests to do their integrity checking by hand. > > > > > > Signed-off-by: Darrick J. Wong <djwong@xxxxxxxxxx> > > > --- > > > common/punch | 1 + > > > 1 file changed, 1 insertion(+) > > > > > > > > > diff --git a/common/punch b/common/punch > > > index 4d16b898..7560edf8 100644 > > > --- a/common/punch > > > +++ b/common/punch > > > @@ -250,6 +250,7 @@ _test_generic_punch() > > > _8k="$((multiple * 8))k" > > > _12k="$((multiple * 12))k" > > > _20k="$((multiple * 20))k" > > > + _require_congruent_file_oplen $TEST_DIR $((multiple * 4096)) > > > > Should the $TEST_DIR be $testfile, or $(dirname $testfile) ? > > Ah, right, that ought to be $(dirname $testfile), thanks for catching > that. I guess I didn't catch that because all the current callers pass > in $TEST_DIR/<somefile>, which is functionally the same, but a landmine > nonetheless. Yeah, I checked all cases which call _test_generic_punch(), they all use $TEST_DIR. In this patchset (e.g. patch 2/8) you sometimes use $testdir for _require_congruent_file_oplen, sometimes use $TEST_DIR or $SCRATCH_MNT directly (even there's $testdir too), although they're not wrong :) This patchset really touch many cases, they looks good mostly, but to avoid bringing in regressions, I have to give them more tests before merging. And welcome more reviewing from others :) Thanks, Zorro > > --D > > > > > > > # initial test state must be defined, otherwise the first test can fail > > > # due ot stale file state left from previous tests. > > > > > >
