Thanks!  for report.

Sequense was:
1) create FS (RAID5), mount, write 40 MB file, sync
2) detach drive
3) start replace detached (but still not closed) drive by another
4) touch mount point, call 'btrfs fi sync' (I use udev script to initiate write immediately after drive removal)
4) <device is closed here> -> oops.

  Right. It didn't handle this case. Now I have fixed this locally.
  But not sent yet, as I am also waiting to base this patch set on
  top of my recent patches.
    [PATCH 0/2] [RFC] btrfs: create degraded-RAID1 chunks

- Anand