LU-1943 zfs-osd: different fixes for zfs

LU-1943 zfs-osd: different fixes for zfs

- echo_create_md_object() to mark la_mode is valid

- ofd_statfs() to set avail/free blocks to 2 when the counters
  are converted to 4K units. otherwise with zfs 2 blocks of size
  128K can be > 0.01% leading to failures in sanity/27*

- zfs-osd to initialize LMA

- zfs-osd's osd_attr_get() to calculate size of directory from
  blocks and do not use cached size (as we don't control it as
  with regular files)

- zfs-osd to track enabled/disabled status for SA and follow it

- take from orion additional locking in osd_object_sa_dirty_add()
  and osd_object_sa_dirty_rele()

- lu_cache_shrink() do not purge objects if __GFP_FS is not set
  this should prevent deadlock when used with zfs backend

- ost_checksum_bulk() to allocate new page to simulate corruption,
  otherwise it corrupts zfs's internal buffers which do not rely
  on PageUptoDate flag

- t-f to modprobe zfs module if osd-zfs is in use

Signed-off-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Change-Id: I027483162735cd0fb4be9fa8bd12c2f2dc43b44a
Reviewed-on: http://review.whamcloud.com/4141
Tested-by: Hudson
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Mike Pershin <tappro@whamcloud.com>