Re: raidframe with R5/RS not reconstructing?

[Greg Oster <oster%cs.usask.ca@localhost>]

On Thu, 16 Feb 2012 15:09:40 -0600
Greg Oster <oster%cs.usask.ca@localhost> wrote:

> On Sun, 12 Feb 2012 06:50:52 -0600 (CST)
> "John D. Baker" <jdbaker%mylinuxisp.com@localhost> wrote:
> 
> > An experimental oversight caused one unit of my raid to get marked
> > as failed, even though nothing's wrong with the hardware.  It's now
> > running in degraded mode and all the data appear to be intact.
> > 
> > No problem, I thought.  'raidctl -R /dev/wd8a raid0' should put it
> > back to its old self.
> > 
> > But it doesn't.  It immediately reports 12% completion and sits
> > there (almost 9 hours now) with no apparent progress and no ETA
> > computed (00:00).
> > 
> > I suppose this is the acid test for the little-used RAID 5 w/Rotated
> > Sparing option.  The last time I needed it (in NetBSD-4 days on
> > sparc), it worked just fine.
> >
> 
> For those playing along at home:  
> 
> I am able to replicate the issue, and have an understanding of what
> the problem is (basically, with rotated sparing every nth stripe of an
> n-disk RAID5_RS set doesn't generate any writes during reconstruction,
> and the logic in the code is expecting there to be writes for every
> stripe). 
> 
> Just a simple matter of finding the correct fix... 
> 
> (the reason this worked back in NetBSD-4 days is that the
> reconstruction map code has been reworked since then, and this bug
> was introduced...)

The following two changes fix this issue:

cvs rdiff -u -r1.33 -r1.34 src/sys/dev/raidframe/rf_reconmap.c
cvs rdiff -u -r1.117 -r1.118 src/sys/dev/raidframe/rf_reconstruct.c

A request for a pullup to netbsd-6 will happen in the next little
while...

Later...

Greg Oster