[PATCH 00/16] gfs2: unlinked inodes

[Andreas Gruenbacher <agruenba@redhat.com>]

We have recently run into cases where unlinked inodes end up
accumulating.  This should only happen as a result of cluster node
failures, and those unlinked inodes should be cleaned up during
allocations.  Sometimes, for some reason, neiter happened.

In turns out that this is mainly due to a bug in commit f0e56edc2ec7
("gfs2: Split the two kinds of glock "delete" work"); see patch "gfs2:
Fix unlinked inode cleanup".

While going through the relevant parts of the code, I noticed a number
of additional minor problems which this patch set addresses as well.

Any thoughts?

Thanks,
Andreas

Andreas Gruenbacher (16):
  gfs2: Rename GLF_VERIFY_EVICT to GLF_VERIFY_DELETE
  gfs2: Initialize gl_no_formal_ino earlier
  gfs2: Allow immediate GLF_VERIFY_DELETE work
  gfs2: Fix unlinked inode cleanup
  gfs2: Faster gfs2_upgrade_iopen_glock wakeups
  gfs2: Rename GIF_{DEFERRED -> DEFER}_DELETE
  gfs2: Rename dinode_demise to evict_behavior
  gfs2: Return enum evict_behavior from gfs2_upgrade_iopen_glock
  gfs2: Minor delete_work_func cleanup
  gfs2: Clean up delete work processing
  gfs2: Call gfs2_queue_verify_delete from gfs2_evict_inode
  gfs2: Update to the evict / remote delete documentation
  gfs2: Use mod_delayed_work in gfs2_queue_try_to_evict
  gfs2: Randomize GLF_VERIFY_DELETE work delay
  gfs2: Use get_random_u32 in gfs2_orlov_skip
  gfs2: Make gfs2_inode_refresh static

 fs/gfs2/glock.c  | 100 ++++++++++++++++-------------------------------
 fs/gfs2/glock.h  |   7 ++++
 fs/gfs2/glops.c  |  11 +++++-
 fs/gfs2/incore.h |   4 +-
 fs/gfs2/inode.c  |   1 +
 fs/gfs2/inode.h  |   2 -
 fs/gfs2/rgrp.c   |   6 +--
 fs/gfs2/super.c  |  76 +++++++++++++++++++++--------------
 8 files changed, 101 insertions(+), 106 deletions(-)

-- 
2.46.0