Message-ID: <4F0AC4D3.4060701@panasas.com>
Date: Mon, 9 Jan 2012 12:43:31 +0200
From: Boaz Harrosh <bharrosh@panasas.com>
MIME-Version: 1.0
To: Benny Halevy <bhalevy@tonian.com>
CC: NFS list <linux-nfs@vger.kernel.org>
Subject: Re: [PATCH] SQUASHME: pnfsd-exofs: Fix the RAID5 export
References: <4EC13EEA.3030404@tonian.com> <4EDBACDD.3010609@tonian.com> <4F0A9F00.4020308@tonian.com> <4F0AC3C1.5040303@panasas.com>
In-Reply-To: <4F0AC3C1.5040303@panasas.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-nfs-owner@vger.kernel.org

Benny hi, thanks.

You will need all the other ORE patches. But they should be
in Linus tree before the 3.3-rc1 so you might just get them
from there.

And this below! With out this fix, the objlayout driver hits
a BUG_ON. Do you think I should check all raid parameters
at client so a BAD server will not crash a client?


Thanks
Boaz
On 01/09/2012 12:38 PM, Boaz Harrosh wrote:
> 
> BAD RAID Math in layout_get.
> 
> TODO: ore_calc_stripe_info, which is already called, as all
> the needed information calculated by _align_io. Only that
> first_stripe_start is only set in case of RAID, but lo_get
> always needs it. Move the set of first_stripe_start inside
> ore_calc_stripe_info, to the generic part.
> 
> Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
> ---
>  fs/exofs/export.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
> 
> diff --git a/fs/exofs/export.c b/fs/exofs/export.c
> index 8a0e510..621bd11 100644
> --- a/fs/exofs/export.c
> +++ b/fs/exofs/export.c
> @@ -96,7 +96,8 @@ void ore_layout_2_pnfs_layout(struct pnfs_osd_layout *pl,
>  
>  static void _align_io(struct ore_layout *layout, u64 *offset, u64 *length)
>  {
> -	u64 stripe_size = layout->group_width * layout->stripe_unit;
> +	u64 stripe_size = (layout->group_width - layout->parity) *
> +							layout->stripe_unit;
>  	u64 group_size = stripe_size * layout->group_depth;
>  
>  	*offset = div64_u64(*offset, group_size) * group_size;