The first patch just changes 'building' into a bool as that's how I
found it used when going to add a new bool for patch 2
Patch 2 addresses the following bz:
https://bugzilla.redhat.com/show_bug.cgi?id=1091866
Essentially, wiping the lv caused it to disappear after a vol-refresh.
This was because the lv that was created was a sparse (or thin or
snapshot via --virtualsize of -V) logical volume. The multiple names
are the ways I found the feature described. When the sparse volume
was written to by the wipe algorithms it actually filled the volume
beyond it's capacity rendering it INACTIVE which causes pool-refresh
to ignore it. As it turns out a sparse lv created of size (for example)
4 MiB has metadata contained within the size created. While we could
adjust the size of the metadata, control over it's location becomes
the issue. It seems there is also some guard data at each end of the
sparse lv as when the wipe was done lost I/O messages were sent to
the system log.
After much searching it just seems that writing a sparse lv with some
sort of wipe algorithm will not work. Although if anyone has suggestions
I'm willing to try. I have found writing '0x00''s, the scrub utility,
and a "dd if=/dev/zero of=/dev/lv_pool/lv_test bs=4096 count=1024"
doesn't work.
It's also of interest to note that the 'scrub' utility doesn't seem
to recognize the logical volume format properly as none of the nonzero
algorithms are run - each fails to open the file - for example:
$ /usr/sbin/lvcreate --name lv_test -L 4096K lv_pool
$ /usr/bin/scrub -f -p bsi /dev/lv_pool/lv_test
scrub: using BSI patterns
scrub: warning: /dev/lv_pool/lv_test is zero length
$
John Ferlan (2):
storage: Convert 'building' into a bool
storage: Disallow vol_wipe for sparse logical volumes
src/conf/storage_conf.h | 2 +-
src/storage/storage_backend_logical.c | 39 ++++++++++++++++++++++++++++++++++-
src/storage/storage_driver.c | 8 +++----
src/util/virstoragefile.h | 1 +
4 files changed, 44 insertions(+), 6 deletions(-)
--
1.9.3