Re: [PATCH] xfs: bump INUMBERS cursor correctly in xfs_inumbers_walk

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Tue, Jul 09, 2019 at 06:59:43AM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
> 
> There's a subtle unit conversion error when we increment the INUMBERS
> cursor at the end of xfs_inumbers_walk.  If there's an inode chunk at
> the very end of the AG /and/ the AG size is a perfect power of two, that
> means we can have inodes, that means that the startino of that last

"...is a perfect power of two, the startino of that last chunk..."

--D

> chunk (which is in units of AG inodes) will be 63 less than (1 <<
> agino_log).  If we add XFS_INODES_PER_CHUNK to the startino, we end up
> with a startino that's larger than (1 << agino_log) and when we convert
> that back to fs inode units we'll rip off that upper bit and wind up
> back at the start of the AG.
> 
> Fix this by converting to units of fs inodes before adding
> XFS_INODES_PER_CHUNK so that we'll harmlessly end up pointing to the
> next AG.
> 
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
>  fs/xfs/xfs_itable.c |    5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/xfs/xfs_itable.c b/fs/xfs/xfs_itable.c
> index cda8ae94480c..a8a06bb78ea8 100644
> --- a/fs/xfs/xfs_itable.c
> +++ b/fs/xfs/xfs_itable.c
> @@ -338,15 +338,14 @@ xfs_inumbers_walk(
>  		.xi_version	= XFS_INUMBERS_VERSION_V5,
>  	};
>  	struct xfs_inumbers_chunk *ic = data;
> -	xfs_agino_t		agino;
>  	int			error;
>  
>  	error = ic->formatter(ic->breq, &inogrp);
>  	if (error && error != XFS_IBULK_ABORT)
>  		return error;
>  
> -	agino = irec->ir_startino + XFS_INODES_PER_CHUNK;
> -	ic->breq->startino = XFS_AGINO_TO_INO(mp, agno, agino);
> +	ic->breq->startino = XFS_AGINO_TO_INO(mp, agno, irec->ir_startino) +
> +			XFS_INODES_PER_CHUNK;
>  	return error;
>  }
>