[lvm-devel] [PATCH 0/3] Add infrastructure for cleaner pv handling, v2

[Dave Wysochanski]

This is the second version of the patchset that adds infrastructure
to move us more towards the goal of removing the standalone PV object.

>From now on, a get/set of a PV property from the pvname requires
a lookup of the vgname, then a vg handle from the vgname, a
pv from the vg handle, and finally using the pv handle to get/set
the property.

The last patch reworks pvchange to use this new functionality.
It passes the testsuite but does have at least one bug.  A
setup with a VG containing a PV with metadatacopies == 0 and
pvchange -u --all end up failing because the PV is on both the
orphan VG and the real VG.  I tried a few ways of working around
this, including processing the non-orphan VGs first, but I believe
the root problem lies in the vg locking and lvmcache.
The lvmcache gets invalidated after each vg lock change, so even
if a PV with metadata copies was known to be in a real VG, once
that lock is dropped the cache is invalidated and the PV shows up
in the vg->pvs list for the orphan VG.