On Mon, Apr 23, 2018 at 8:12 AM, Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > On Mon, Apr 23, 2018 at 4:50 PM, Amir Goldstein <amir73il@xxxxxxxxx> wrote: >> On Mon, Apr 23, 2018 at 6:37 AM, Miklos Szeredi <miklos@xxxxxxxxxx> wrote: >>> On Wed, Apr 18, 2018 at 4:30 PM, Amir Goldstein <amir73il@xxxxxxxxx> wrote: >>>> On Wed, Apr 18, 2018 at 5:23 PM, Miklos Szeredi <miklos@xxxxxxxxxx> wrote: >>>>> On Wed, Apr 18, 2018 at 4:16 PM, Amir Goldstein <amir73il@xxxxxxxxx> wrote: >>>>> >>>>>> FYI, Miklos has a patch set in review to fix the last of the >>>>>> non-standard behavior tracking test that is still failing (overlay/016). >>>>>> After that patch set is merged, all overlay tests will be expected to >>>>>> pass on master - new era ;-) >>>>> >>>>> Oh, we can write new tests that don't pass ;) >>> >>> Another one: >>> >>> fstat() returns incorrect st_ctime after unlink of a lower file/directory. >>> >>> Alternative fixes: >>> >>> 1) take some attributes from whiteout >>> >>> 2) metacopy up to tmpfile I have patches to copy up to an "orphan index": https://github.com/amir73il/linux/commits/ovl-rocopyup combine that with metacopy and you have a ready and inexpensive place holder for st_ctime of unlinked file. >>> >>> But first of all, we need a test case that fails, otherwise there's no >>> motivation to do the fix ;) >>> >> >> I admit that the failing xfstests I wrote have had a "motivating" >> effect, but going forward, I feel that xfstests is not really the place >> for these sort of weird behavior tests. I find it to be more suiting as >> a regression tracking test suite and special test case coverage. >> >> IMO, parent ctime fits better to a POSIX compliance test suite, such >> as pdjfstest. > > I'm not talking about parent ctime, rather victim ctime (updated due > to nlink change). > >> >> And besides, overlayfs now feels it is in a functional state that is worthy >> of PASS. Sure, that is a very subjective observation. >> >>> I'm wondering, would it make sense to manually add "rotation points" >>> to plain xfstest test cases? >> >> Can you give an example of an xfstest and a rotation point? > > Any test where there is a setup phase and a test phase is asking for a > rotation point (i.e. set up on lower (no-overlay) and then modify on > overlay). xfstest cases do not have a setup phase and run phase, so I don't see a way that this could be done elegantly without modifying all test cases. > > I haven't thought about individual test cases, just wondering if > there's already a testcase testing st_ctime on unlink for example, > that is only not failing because there isn't a rotation of overlay > after the setup phase. > >From what I can see there are a few tests checking ctime update, mainly regression tests for btrfs ctime update bugs: $ git grep ctime.update tests/ tests/generic/221:# Check ctime updates when calling futimens without UTIME_OMIT for the tests/generic/236:# Check ctime updated or not if file linked tests/generic/236:# check ctime updated tests/generic/277:# Check if ctime update caused by chattr is written to disk tests/generic/408: echo "ctime updated" generic/236 tests the link use case, but no test for unlink AFAIK. Thanks, Amir. -- To unsubscribe from this list: send the line "unsubscribe linux-unionfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
