Revision 1:
  cvs dm 1.0

Revision 2:
  Only every other metadata area was being read when loading a snapshot!
  [Kevin Corry]

Revision 3:
  Don't initialise static variables to zero/NULL.

Revision 4:
  Change resume/suspend to do_resume/do_suspend to avoid name clash.

Revision 5:
  Hello all,
  
  The current version of the VFS locking patch adds a new semaphore to
  fs/super.c.  This is used to make sure a filesystem does not get mounted
  on a logical volume while a snapshot is being taken.  It also results in
  all mounts on the system being serialized, and isn't in line with the
  VFS locking scheme in general.
  
  I've been meaning to fix it forever, here's an updated version that adds
  a super with s->s_dev set to the source volume if nothing is currently
  mounted on the source volume.  This allows me to use the s_umount
  semaphore in the super block to keep things safe, which is cleaner
  overall.  
  
  The other benefit over the existing patch is this one has zero footprint
  outside the lockfs calls.  You're only running new code if you take a
  snapshot.
  
  I've done some testing here, but wanted to let LVM people review it
  before going further.  Patch is below against 2.4.21-rc6.  
  
  This provides zero new functionality over the existing VFS locking
  patch, and is experimental.  Do not apply this on production servers,
  and do not apply unless you want to help test.
  
  -chris
  
  ===== drivers/md/lvm.c 1.19 vs edited =====

Revision 6:
  missing parts of the previous vfs patch (merge).

Revision 7:
  Lift vfs locking to dm_suspend/resume.

Revision 8:
  Correct error message when start a dm-daemon.

Revision 9:
  When multiple load ioctls are issued the reference count on older
  'new_tables' wasn't being dropped.  [Christophe Saout]

Revision 10:
  Stop labelling dm as 'experimental'.

Revision 11:
  Move retrieve_status up so dev_wait() can use it.

Revision 12:
  dev_wait was meant to return table status not dev status.  [Alasdair Kergon]

Revision 13:
  Fix error message when linear targets gets handed more than 2 arguments.
  [Alasdair Kergon]

Revision 14:
  Support an arbitrary number of target parameters.  [Alasdair Kergon]

Revision 15:
  Fix VFS patch

Revision 16:
  The ioctl interface always knows how many targets are going to be in
  the table, so remove the dynamic array sizing code in dm-table.c.
  This fixes a problem with large tables where the dm_target pointer
  passed to the target ctr was becoming invalid.

Revision 17:
  Correct calculation of the dirty logs bitset size.

Revision 18:
  Correct the sector calculation in map_buffer().

Revision 19:
  If a kcopyd client hadn't allocated enough pages and then submitted a
  large io that was being split into sub jobs we could stall waiting for
  pages.  There is now a kcopyd_client->max_split field that is an
  appropriate number of sub_jobs to split the io into based on the
  number of allocated pages.

Revision 20:
  Fix bug in dm-io.c block size calculation.  [Alasdair Kergon]

Revision 21:
  bh->b_rdev wasn't being set properly.  Bug from earlier patch.

Revision 22:
  The number of targets in a table needs to be rounded up to
  KEYS_PER_NODE.  [Kevin Corry]