on 2019/05/08 0:37, Darrick J. Wong wrote:
[cc fstests]
On Tue, May 07, 2019 at 04:51:35PM +0800, xuyang wrote:
Hi darrick
since commit d0e484a("check: wipe scratch devices between tests")is
merged into xfstests, when I running xfs/297 on kernel 5.1.0-rc5+ with
xfsprogs-4.18.0-3.el8.x86_64, it causes a failure that log size is too
small to reach the minimum size, as below:
#wipefs -a /dev/sda11 (20G)
/dev/sda11: 4 bytes were erased at offset 0x00000000 (xfs): 58 46 53 42
#mkfs.xfs -f -d agcount=16,su=256k,sw=12 -l su=256k,size=5120b /dev/sda11
log size 5120 blocks too small, minimum size is 5184 blocks
That's not related to wipefs at all.
The problem here is that your vendor's xfsprogs package turns on reflink
by default. The reflink feature increases the minimum log size
requirements, which this test doesn't account for, and hence it misses
by 64 blocks. Evidently nobody at your vendor's QA department noticed?
I only noticed because I started carrying an "enable reflink by default"
patch last Thursday and it caused a bunch of regressions on tests that
call mkfs.xfs without looping in MKFS_OPTIONS. I will be sending out
patches to fix all that shortly and will cc you on them.
I got it .
upstream xfsprogs doesn't have this problem.
Upstream xfsprogs doesn't enable reflink by default.
I am confused about why the min_logblocks becomes larger after wipefs.
Is it a calculating minimum log size bug? Perhaps, I can adjust the
logsize to 5184b. Can you give some advise?
Wait for the corrections and help me test them, please? :)
--D
Hi Darrick
I have tested your patchset for 20 times(run seven affected cases), it's ok on my machine.
Kind regards,
Yang Xu
