On Thu, Aug 29, 2019 at 6:20 PM Mike Marshall hubcap@omnibond.com wrote:
Hi Deepa...
I installed this patch series on top of Linux 5.3-rc6 and ran xfstests on orangefs and got a regression... generic/258 failed with: "Timestamp wrapped"...
# cat results/generic/258.out.bad QA output created by 258 Creating file with timestamp of Jan 1, 1960 Testing for negative seconds since epoch Timestamp wrapped: 0 Timestamp wrapped (see /home/hubcap/xfstests-dev/results//generic/258.full for details)
Note that patch [16/20] https://lkml.org/lkml/2019/8/18/193 assumes that orangefs does not support negative timestamps. And, the reason was pointed out in the commit text:
---------------------- Assume the limits as unsigned according to the below commit 98e8eef557a9 ("changed PVFS_time from int64_t to uint64_t") in https://github.com/waltligon/orangefs
Author: Neill Miller neillm@mcs.anl.gov Date: Thu Sep 2 15:00:38 2004 +0000 --------------------
So the timestamp being wrapped to 0 in this case is correct behavior according to my patchset. The generic/258 assumes that the timestamps can be negative. If this is not true then it should not be run for this fs.
But, if you think the timestamp should support negative timestamps for orangefs, I'd be happy to change it.
Just to double check, I compiled Linux 5.3-rc6 without the timestamp series, and 258 passed... then added the patch series back and the failure returned...
Also.... when I was using "git am" to apply the patch series, the ceph patch failed, so I "git skipped" it...
My series applies cleanly against linux-next as of yesterday.
-Deepa