On Wed, Aug 18, 2021 at 01:36:32PM +0800, Qu Wenruo wrote:
quoted
Looks good to me.
I hope there isn't any xfstests/tests/btrfs test case that is hardcoded
to the older block group alloc (like swap file test with relocation?).
But then the test case will need a fix, not btrfs-progs. So
Well, fsck/025 will fail, and it's not the test case needs a fix...
We just suddenly can't mkfs on a 128MiB device due to the enlarged
minimal stripe length of data chunks.
This still exposes a behavior difference between kernel and btrfs-progs...
So I'll reply here as all people involved are also CCed - I'm dropping
this patch for now.
The test 025 fails and my idea was to skip it for now but that still
leaves the problems with creating small filesystems, that's something
that has been working and it should keep doing so.
Proper fix: scale the chunk size for large filesystems and also small
filesystems. We can probably set the minimum size to somewhere like 64M
(for non-mixed setups), and scale the chunk sizes according to that.
d.