Re: [PATCHv2 5/9] generic/031: Fix the test case for 64k blocksize config

On Tue, Aug 03, 2021 at 10:30:33AM +0530, Ritesh Harjani wrote:

On 21/08/02 12:00AM, Eryu Guan wrote:

quoted

On Wed, Jul 21, 2021 at 10:57:58AM +0530, Ritesh Harjani wrote:

quoted

This test fails with blocksize 64k since the test assumes 4k blocksize
in fcollapse param. This patch fixes that and also tests for 64k
blocksize.

Signed-off-by: Ritesh Harjani <redacted>
---
 tests/generic/031     | 14 +++++++++-----
 tests/generic/031.out | 16 ++++++++--------
 2 files changed, 17 insertions(+), 13 deletions(-)

diff --git a/tests/generic/031 b/tests/generic/031
index 313ce9ff..11961c54 100755
--- a/tests/generic/031
+++ b/tests/generic/031

@@ -26,11 +26,16 @@ testfile=$SCRATCH_MNT/testfile
 _scratch_mkfs > /dev/null 2>&1
 _scratch_mount

+# fcollapse need offset and len to be multiple of blocksize for filesystems
+# So let's make the offsets and len required for fcollapse multiples of 64K
+# so that it works for all configurations (including on dax on 64K page size
+# systems)
+fact=$((65536/4096))
 $XFS_IO_PROG -f \
-	-c "pwrite 185332 55756" \
-	-c "fcollapse 28672 40960" \
-	-c "pwrite 133228 63394" \
-	-c "fcollapse 0 4096" \
+	-c "pwrite $((185332*fact + 12)) $((55756*fact + 12))" \

Where does this 12 come from?

A random number so that the offset and length are not bocksize aligned.
If you see the final .out file, you will see the offset of the writes
remains the same with and before this patch.

quoted

And I'm wondering if this still reproduces the original bug.

I am not sure how to trigger this. I know that this test was intended for
bs < ps cases. If someone can help me / point me to the kernel fix for this,
I can try to reproduce the original bug too.

I found this link for this test patch series. Couldn't find the kernel fixes
link though.
https://www.spinics.net/lists/fstests/msg00340.html

I think it's a regression test for this patchset.

https://www.spinics.net/lists/xfs/msg29807.html

quoted

And looks like that the original test setups came from a specific
fsstress or fsx run, and aimed to the specific bug, perhaps we could
require the test with <= 4k block size, and _notrun in 64k case.

It would be good to know whether this code could trigger the original bug or
not. Then we need not make _notrun for 64k case.

Agreed, if we could make sure that updated test still triggers the
original bug, there's no reason _notrun for 64k case.

Thanks,
Eryu