Re: [PATCH] mdadm reshape hangs on external grow chunk

[Mariusz Tkaczyk <mariusz.tkaczyk@linux.intel.com>]

On Thu, 29 Sep 2022 11:35:21 +0200
Kinga Tanska <kinga.tanska@linux.intel.com> wrote:

> On Fri, 23 Sep 2022 10:26:35 -0400
> Nigel Croxon <ncroxon@redhat.com> wrote:
> 
> > After creating a raid array on top of a imsm container. Try to
> > grow the chunk size and the reshape will hang with zero progress.
> > The reason is the computation of sync_max_to_set value:
> > if (before_data_disks <= data_disks)
> >         sync_max_to_set = sra->reshape_progress / data_disks;
> >     else
> >         sync_max_to_set = (sra->component_size * data_disks
> >                        - sra->reshape_progress) / data_disks;
> > 
> > Can produce a zero result. Which is then used to set the maximum
> > sync value, causing zero progress to the reshape.  The change is to
> > test if the sync_max_to_set value is zero. And if so, set the sysfs
> > sync_max to "max".
> > 
> > Steps to Reproduce:
> > 1. Create a container and RAID0 array
> > mdadm -CR /dev/md/imsm -e imsm -n2 /dev/nvme0n1 /dev/nvme1n1
> > mdadm -CR  /dev/md/vol -l0 --chunk=16 -n2 /dev/nvme0n1 /dev/nvme1n1
> > 2. Wait for resync
> > 3. Try to grow the chunk size
> > mdadm --grow /dev/md/vol --chunk=256
> > 
> > Signed-off-by: Nigel Croxon <ncroxon@redhat.com>
> > ---
> >  Grow.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/Grow.c b/Grow.c
> > index 0f07a894..6c5021bc 100644
> > --- a/Grow.c
> > +++ b/Grow.c
> > @@ -943,7 +943,7 @@ int start_reshape(struct mdinfo *sra, int
> > already_running, if (!already_running)
> >  		sysfs_set_num(sra, NULL, "sync_min",
> > sync_max_to_set); 
> > -        if (st->ss->external)
> > +        if (sync_max_to_set)
> >  		err = err ?: sysfs_set_num(sra, NULL, "sync_max",
> > sync_max_to_set); else
> >  		err = err ?: sysfs_set_str(sra, NULL, "sync_max",
> > "max");
> 
> Hi Nigel,
> 
> I was trying to retest with your patch but still have the defect. I
> analyzed it and found another reason, which causes this defect. In
> validate_geometry_imsm function freesize and super is being checked and
> return 1 if any of those is NULL. In my opinion 0 shall be returned
> here, because it is an error and reshape should be stopped here. I will
> prepare proper patch and send to review immediately.
> 
Hi Nigel,
I agree with Kinga.
https://patchwork.kernel.org/project/linux-raid/patch/20221028025117.27048-1-kinga.tanska@intel.com/
Could you please retest the proposed patch on your side and provide feedback?

Thanks,
Mariusz