zfswin-2.2.0rc6 testing

[lundman]

1:
Creating:
BOOM
BOOM/lower
BOOM/lowerer

Had inconsistent zfs mount output.

2:
rclone sync . E:/lower/
does not work at all. Could it be rclone does not handle reparse-points? Test with vhd.

3:
snapshot mount showed directory, but no contents, bug?

4:
zpool export fails with "busy". Because snapshot mount went wrong? bug?

[EchterAgo]

I also have found another issue with the export/unmount.

So I import an existing dataset that has my backups on it, then I let rclone sync. It just checks the metadata of ~95k files and does not write anything new because things are already up to date. Then I try to export the pool and it will just hang forever.

What is interesting here is that all the VREG nodes get reclaimed very fast but then on the filesanddirs pass there are only directory nodes (VDIR) left and they can not be reclaimed because of v_usecount > 0. I let this run for quite a while and just nothing changes. I also checked if there are still any programs that have handles to the path open, but there are none.

These notes all have v_type == VDIR, v_usecount > 0, v_iocount == 0, SectionObjectPointers == NULL, v_data == NULL, except for the VNODE_MARKROOT nodes.

Is there a way to diagnose what is keeping their v_usecount > 0?

[EchterAgo]

Nevermind, I think what I'm seeing here is vflush just taking a long time to process 95k vnodes. I was able to optimize this a lot by reducing the locking calls in the loop, I'll try to make a PR for that.

[EchterAgo]

I managed to get the reclaim time for the 95k vnodes down from minutes to a second.

[EchterAgo]

See #328