[PATCH 4/5] xfs: free inodes on log recovery error

[Brian Foster]

On Wed, Jul 02, 2014 at 09:32:10AM -0500, Mark Tinguely wrote:
> Recovery may free inodes that end up on the inode
> reclaim RCU. If recovery fails, we leak these inodes.
> The filesystem should be in forced shutdown at this
> point, so a call to xfs_reclaim_inode is a fast path
> to freeing the inodes and RCU entries.
> 
> Signed-off-by: Mark Tinguely <tinguely at sgi.com>
> ---
>  fs/xfs/xfs_log.c   |    2 ++
>  fs/xfs/xfs_mount.c |    1 +
>  2 files changed, 3 insertions(+)
> 
> Index: b/fs/xfs/xfs_log.c
> ===================================================================
> --- a/fs/xfs/xfs_log.c
> +++ b/fs/xfs/xfs_log.c
> @@ -31,6 +31,7 @@
>  #include "xfs_log_priv.h"
>  #include "xfs_log_recover.h"
>  #include "xfs_inode.h"
> +#include "xfs_icache.h"
>  #include "xfs_trace.h"
>  #include "xfs_fsops.h"
>  #include "xfs_cksum.h"
> @@ -720,6 +721,7 @@ xfs_log_mount(
>  	return 0;
>  
>  out_destroy_ail:
> +	xfs_reclaim_inodes(mp, SYNC_WAIT);

So an inode in the perag cache means an xfs_iget(). I see that in
xlog_recover_process_one_iunlink(), which is via
xfs_log_mount_finish(). Assuming I'm following that correctly, why the
reclaim here (and in response to failure here by the caller) as opposed
to closer to a failure of xfs_log_mount_finish()?

Brian

>  	xfs_trans_ail_destroy(mp);
>  out_free_log:
>  	xlog_dealloc_log(mp->m_log);
> Index: b/fs/xfs/xfs_mount.c
> ===================================================================
> --- a/fs/xfs/xfs_mount.c
> +++ b/fs/xfs/xfs_mount.c
> @@ -980,6 +980,7 @@ xfs_mountfs(
>   out_log_dealloc:
>  	xfs_log_unmount(mp);
>   out_fail_wait:
> +	xfs_reclaim_inodes(mp, SYNC_WAIT);
>  	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
>  		xfs_wait_buftarg(mp->m_logdev_targp);
>  	xfs_wait_buftarg(mp->m_ddev_targp);
> 
> 
> _______________________________________________
> xfs mailing list
> xfs at oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs