Re: User space RAID-6 access

[NeilBrown <neilb@suse.de>]

On Tue, 1 Feb 2011 20:21:54 +0100 Piergiorgio Sartor
<piergiorgio.sartor@nexgo.de> wrote:

> On Tue, Feb 01, 2011 at 07:52:59AM +1100, NeilBrown wrote:
> > On Mon, 31 Jan 2011 21:20:55 +0100 Piergiorgio Sartor
> > <piergiorgio.sartor@nexgo.de> wrote:
> > 
> > > Hi all,
> > > 
> > > some times ago, I think was Neil, it was mentioned
> > > about the possibility to access consistently a RAID-6
> > > array from user space, in order to be able to perform
> > > some checks, like the notorius "which HDD has wrong data".
> > > 
> > > Is there any reference or documentation or source code
> > > which can be taken as example for such a case?
> > > 
> > 
> > Look in the mdadm source code, particularly at restripe.c
> > 
> > Also
> >    make test_stripe
> > 
> > make a program the the test suite uses for verify data correctness.
> > 
> > That should give you enough hints to get you started.
> 
> 
> Hi Neil, thanks for the pointer.
> 
> I had a look at the code and there is something I did not get.
> 
> It seems to me the HDDs composing the array are "simply"
> opened with "open".
> 
> In case the array is in use, how is avoided a race condition
> between the "test_stripe" program and the md device?
> 
> I mean, the first could start to read from the HDDs and the
> other could write something in the same place, leading to an
> inconsistent parity, from the "test_stripe" point of view,
> since it missed some update.
> 
> Or there is a locking inside md (in "test_stripe" I could
> not see and it would also be dangerous)?
> 
> Thanks again,
> 
> bye,
> 

I didn't realise that you wanted to look at the members of the array while
the array was active!!  That is a bit harder.  But not impossible.

If you write a couple of numbers to 'suspend_lo' and 'suspend_hi' in sysfs,
then writes to blocks between those two array addresses will be blocked.
So you could suspend a region, look at the blocks, then un-suspend.

NeilBrown