Re: [linux-lvm] Pvmove question

[Joe Thornber <thornber@btconnect.com>]

On Mon, Aug 27, 2001 at 08:46:33AM -0500, Kevin M Corry wrote:
> > On Fri, Aug 24, 2001 at 12:31:18AM +0200, Jean-Eric Cuendet wrote:
> > >
> > > Hi,
> > > I have a pvmove question:
> > > In the man page, it's said to be careful with pvmove if you use it
> while
> > > LVM is in use.
> > > I don't really understand what is dangerous and what is safe...
> >
> > pvmove on a mounted (or otherwise in use) LV might cause curruption of
> > your data.
> 
> In what cases could pvmove cause corruption on a mounted LV? Looking
> at the man page and scanning through pvmove.c, it seems to indicate
> that only a system crash during pvmove would corrupt the data. Is
> this the only case, or are there others?

Search back through the list for a response from Andrea Arcangeli.  We
need to track buffer_heads that have been trhough the make_request
function so we can really can flush the LV when suspending io for the
dureation of the move.  If you're not using raw io then the only
problem is a possible deadlock when under high memory pressure.

> If there are other possibilities for corruption, wouldn't it be a
> good idea to not allow pvmove's on PV's that contain mounted LVs?

IMO yes.

- Joe