On Fri, Apr 22, 2016 at 02:04:51PM -0400, Jan Harkes wrote:
> On Fri, Apr 22, 2016 at 07:40:00PM +0200, u-myfx_at_aetey.se wrote:
> > May be we are especially lucky, but this was the only visible problem.
> 
> This particular VIOC_GETVOLSTAT pioctl call is used in 4 different
> places in cfs and 2 places in repair, your fix addressed one place. Also

It is likely that we never exercised repair on ARM machines.

(The repair process has unfortunately never been robust even on intel :(
but we did not have the resources to analyze and fix it.
I don't think we ever lost data but it is not unusual to see repair
misbehave.)

> things like the pioctl in/out structures seem to be defined at the spot
> of the implementation which makes it very easy to make a change
> somewhere that is not picked up elsewhere.
> 
> Considering how many places need to change I'm no longer sure I want to
> reshuffle and am starting to think that having common pioctl pack/unpack
> functions in a shared place might be a better approach. Going to have to
> ponder this a bit.

Nice, thanks for this cleanup!

Rune