Re: [PATCH] xfs/444: test log replay after XFS_IOC_SWAPEXT

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

On Fri, Feb 23, 2018 at 04:58:30PM -0600, Eric Sandeen wrote:
> 
> 
> On 2/23/18 3:54 PM, Darrick J. Wong wrote:
> > On Fri, Feb 23, 2018 at 12:33:41PM -0600, Eric Sandeen wrote:
> >> This is a mashup of xfs/042 and some of the log replay tests;
> >> it checks whether the log can be replayed if we crash immediately
> >> after an xfs_fsr / XFS_IOC_SWAPEXT.
> >>
> >> Hint: it can't.  It fails because the temporary donor inode has
> >> been deleted and has invalid mode 0 when we try to replay its
> >> swapext operation.  Kernel patches to fix it will follow soon.
> > 
> > Hmm, does this filesystem have rmap enabled or not?
> 
> Not when I tested it.  I can try with it.
> 
> > Different swapext strategy in play depending on the answer to that
> > question.
> 
> Hm, ok.
> 
> 
> > Maybe I should wait for patches to show up. :P
> 
> Something like this, with a better comment, and also skipping
> XFS_ILOG_[DA]OWNER on replay if mode == 0 to catch it on an
> upgrade if needed.
> 
> diff --git a/fs/xfs/xfs_inode_item.c b/fs/xfs/xfs_inode_item.c
> index 26f2413..72a0223 100644
> --- a/fs/xfs/xfs_inode_item.c
> +++ b/fs/xfs/xfs_inode_item.c
> @@ -436,6 +436,12 @@ xfs_inode_item_format(
>  			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
>  	}
>  
> +	/* If this inode has been deleted do not try to recover swapext */
> +	if (VFS_I(ip)->i_mode == 0) {
> +		ilf->ilf_fields &= ~(XFS_ILOG_DOWNER | XFS_ILOG_AOWNER);
> +		iip->ili_fields &= ~(XFS_ILOG_DOWNER | XFS_ILOG_AOWNER);

Definitely the !rmap case then, since rmap swapext uses defer_ops
to move the extents instead of [AD]OWNER.

--D

> +	}
> +
>  	/* update the format with the exact fields we actually logged */
>  	ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
>  }