On Thu, Aug 12, 2021 at 08:03:26PM +0300, Kari Argillander wrote:

On Sun, Aug 01, 2021 at 11:23:16PM -0400, Theodore Ts'o wrote:

quoted

On Thu, Jul 29, 2021 at 09:24:59AM -0700, Darrick J. Wong wrote:

quoted

I have the same (still unanswered) questions as last time:

1. What happens when you run ntfs3 through fstests with '-g all'?  I get
that the pass rate isn't going to be as high with ntfs3 as it is with
ext4/xfs/btrfs, but fstests can be adapted (see the recent attempts to
get exfat under test).

Indeed, it's not that hard at all.  I've included a patch to
xfstests-bld[1] so that you can just run "kvm-xfstests -c ntfs3 -g
auto".

Konstantin, I would *strongly* encourage you to try running fstests,
about 60 seconds into a run, we discover that generic/013 will trigger
locking problems that could lead to deadlocks.

It seems at least at my testing that if acl option is used then
generic/013 will pass. I have tested this with old linux-next commit
5a4cee98ea757e1a2a1354b497afdf8fafc30a20 I have still some of my own
code in it but I will test this tomorrow so I can be sure.

It also seems that acl support is broken. I also suspect ntfs-3g mkfs in
some failure cases. So maybe ntfs-3g mkfs will give different result than
Paragons mkfs. It would be nice to test with Paragons mkfs software or
that Paragon will test with ntfs-3g.

I have made more testing and it was actually my code which cause 013 not
fail. It is still pretty strange. I have made code for new mount api (fs
context) and 013 still get deadlock but still test will pass. This only
happends if acl is on. Though this is intresting why this happends. I
will not use more time for this now. I will try to focus fs context
mount api for now.