Re: [PATCH 018/119] xfs: enable the xfs_defer mechanism to process extents to free

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

On Tue, Jun 28, 2016 at 08:32:39AM -0400, Brian Foster wrote:
> On Mon, Jun 27, 2016 at 03:00:03PM -0700, Darrick J. Wong wrote:
> > On Mon, Jun 27, 2016 at 02:41:45PM -0700, Darrick J. Wong wrote:
> > > On Mon, Jun 27, 2016 at 09:15:08AM -0400, Brian Foster wrote:
> > > > On Thu, Jun 16, 2016 at 06:19:47PM -0700, Darrick J. Wong wrote:
> > > > > Connect the xfs_defer mechanism with the pieces that we'll need to
> > > > > handle deferred extent freeing.  We'll wire up the existing code to
> > > > > our new deferred mechanism later.
> > > > > 
> > > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > ---
> > > > 
> > > > Could we merge this with the xfs_trans_*efi/efd* bits? We'd need to
> > > > preserve some calls for recovery, but it looks like other parts are only
> > > > used by the deferred ops infrastructure at this point.
> > > 
> > > Yes, we could replace xfs_bmap_free_create_{intent,done} with
> > > xfs_trans_get_ef[id] and lose the silly functions.  I'll go take
> > > care of all of them.
> > 
> > Hah, gcc complains about the mismatch in pointer types for the second
> > argument.
> > 
> > fs/xfs/xfs_defer_item.c:504:17: error: initialization from incompatible pointer type [-Werror=incompatible-pointer-types]
> >   .create_done = xfs_trans_get_bud,
> >                  ^
> > fs/xfs/xfs_defer_item.c:504:17: note: (near initialization for ‘xfs_bmap_update_defer_type.create_done’)
> > 
> > I guess one could put in an ugly cast to coerce the types at a cost of
> > uglifying the code.  <shrug> Opinions?
> > 
> 
> Not sure what you mean here... I should be more clear. What I was
> thinking is to nuke xfs_defer_item.c, define the 'const struct
> xfs_defer_op_type xfs_extent_free_defer_type' right in
> xfs_trans_extfree.c (perhaps rename the file) and wire up the functions
> appropriately. Just change the function signatures if you need to. For
> something like xfs_trans_get_efd(), maybe refactor the guts into a
> static helper that both the defer callback and xfs_trans_get_efd() can
> use, since we need the latter for log recovery. Will something like that
> work?
> 
> To put it simply, it looks like we have at least a few places where
> defer item has a callback that calls yet another interface, but is the
> only caller of the latter (e.g.,
> xfs_bmap_free_create_intent()->xfs_trans_get_efi()). 

*Oh*.  Yes, certainly all the stuff in xfs_defer_item.c can be broken up by
type and moved into xfs_trans_*.c.

The libxfs version will retain libxfs/defer_item.c since the deferred op types
there simply call the appropriate libxfs functions without any logging.

--D

> 
> Brian
> 
> > --D
> > 
> > > 
> > > --D
> > > 
> > > > 
> > > > Brian
> > > > 
> > > > >  fs/xfs/libxfs/xfs_defer.h |    1 
> > > > >  fs/xfs/xfs_defer_item.c   |  108 +++++++++++++++++++++++++++++++++++++++++++++
> > > > >  2 files changed, 109 insertions(+)
> > > > > 
> > > > > 
> > > > > diff --git a/fs/xfs/libxfs/xfs_defer.h b/fs/xfs/libxfs/xfs_defer.h
> > > > > index 85c7a3a..743fc32 100644
> > > > > --- a/fs/xfs/libxfs/xfs_defer.h
> > > > > +++ b/fs/xfs/libxfs/xfs_defer.h
> > > > > @@ -51,6 +51,7 @@ struct xfs_defer_pending {
> > > > >   * find all the space it needs.
> > > > >   */
> > > > >  enum xfs_defer_ops_type {
> > > > > +	XFS_DEFER_OPS_TYPE_FREE,
> > > > >  	XFS_DEFER_OPS_TYPE_MAX,
> > > > >  };
> > > > >  
> > > > > diff --git a/fs/xfs/xfs_defer_item.c b/fs/xfs/xfs_defer_item.c
> > > > > index 4c2ba28..127a54e 100644
> > > > > --- a/fs/xfs/xfs_defer_item.c
> > > > > +++ b/fs/xfs/xfs_defer_item.c
> > > > > @@ -29,9 +29,117 @@
> > > > >  #include "xfs_defer.h"
> > > > >  #include "xfs_trans.h"
> > > > >  #include "xfs_trace.h"
> > > > > +#include "xfs_bmap.h"
> > > > > +#include "xfs_extfree_item.h"
> > > > > +
> > > > > +/* Extent Freeing */
> > > > > +
> > > > > +/* Sort bmap items by AG. */
> > > > > +static int
> > > > > +xfs_bmap_free_diff_items(
> > > > > +	void				*priv,
> > > > > +	struct list_head		*a,
> > > > > +	struct list_head		*b)
> > > > > +{
> > > > > +	struct xfs_mount		*mp = priv;
> > > > > +	struct xfs_bmap_free_item	*ra;
> > > > > +	struct xfs_bmap_free_item	*rb;
> > > > > +
> > > > > +	ra = container_of(a, struct xfs_bmap_free_item, xbfi_list);
> > > > > +	rb = container_of(b, struct xfs_bmap_free_item, xbfi_list);
> > > > > +	return  XFS_FSB_TO_AGNO(mp, ra->xbfi_startblock) -
> > > > > +		XFS_FSB_TO_AGNO(mp, rb->xbfi_startblock);
> > > > > +}
> > > > > +
> > > > > +/* Get an EFI. */
> > > > > +STATIC void *
> > > > > +xfs_bmap_free_create_intent(
> > > > > +	struct xfs_trans		*tp,
> > > > > +	unsigned int			count)
> > > > > +{
> > > > > +	return xfs_trans_get_efi(tp, count);
> > > > > +}
> > > > > +
> > > > > +/* Log a free extent to the intent item. */
> > > > > +STATIC void
> > > > > +xfs_bmap_free_log_item(
> > > > > +	struct xfs_trans		*tp,
> > > > > +	void				*intent,
> > > > > +	struct list_head		*item)
> > > > > +{
> > > > > +	struct xfs_bmap_free_item	*free;
> > > > > +
> > > > > +	free = container_of(item, struct xfs_bmap_free_item, xbfi_list);
> > > > > +	xfs_trans_log_efi_extent(tp, intent, free->xbfi_startblock,
> > > > > +			free->xbfi_blockcount);
> > > > > +}
> > > > > +
> > > > > +/* Get an EFD so we can process all the free extents. */
> > > > > +STATIC void *
> > > > > +xfs_bmap_free_create_done(
> > > > > +	struct xfs_trans		*tp,
> > > > > +	void				*intent,
> > > > > +	unsigned int			count)
> > > > > +{
> > > > > +	return xfs_trans_get_efd(tp, intent, count);
> > > > > +}
> > > > > +
> > > > > +/* Process a free extent. */
> > > > > +STATIC int
> > > > > +xfs_bmap_free_finish_item(
> > > > > +	struct xfs_trans		*tp,
> > > > > +	struct xfs_defer_ops		*dop,
> > > > > +	struct list_head		*item,
> > > > > +	void				*done_item,
> > > > > +	void				**state)
> > > > > +{
> > > > > +	struct xfs_bmap_free_item	*free;
> > > > > +	int				error;
> > > > > +
> > > > > +	free = container_of(item, struct xfs_bmap_free_item, xbfi_list);
> > > > > +	error = xfs_trans_free_extent(tp, done_item,
> > > > > +			free->xbfi_startblock,
> > > > > +			free->xbfi_blockcount);
> > > > > +	kmem_free(free);
> > > > > +	return error;
> > > > > +}
> > > > > +
> > > > > +/* Abort all pending EFIs. */
> > > > > +STATIC void
> > > > > +xfs_bmap_free_abort_intent(
> > > > > +	void				*intent)
> > > > > +{
> > > > > +	xfs_efi_release(intent);
> > > > > +}
> > > > > +
> > > > > +/* Cancel a free extent. */
> > > > > +STATIC void
> > > > > +xfs_bmap_free_cancel_item(
> > > > > +	struct list_head		*item)
> > > > > +{
> > > > > +	struct xfs_bmap_free_item	*free;
> > > > > +
> > > > > +	free = container_of(item, struct xfs_bmap_free_item, xbfi_list);
> > > > > +	kmem_free(free);
> > > > > +}
> > > > > +
> > > > > +const struct xfs_defer_op_type xfs_extent_free_defer_type = {
> > > > > +	.type		= XFS_DEFER_OPS_TYPE_FREE,
> > > > > +	.max_items	= XFS_EFI_MAX_FAST_EXTENTS,
> > > > > +	.diff_items	= xfs_bmap_free_diff_items,
> > > > > +	.create_intent	= xfs_bmap_free_create_intent,
> > > > > +	.abort_intent	= xfs_bmap_free_abort_intent,
> > > > > +	.log_item	= xfs_bmap_free_log_item,
> > > > > +	.create_done	= xfs_bmap_free_create_done,
> > > > > +	.finish_item	= xfs_bmap_free_finish_item,
> > > > > +	.cancel_item	= xfs_bmap_free_cancel_item,
> > > > > +};
> > > > > +
> > > > > +/* Deferred Item Initialization */
> > > > >  
> > > > >  /* Initialize the deferred operation types. */
> > > > >  void
> > > > >  xfs_defer_init_types(void)
> > > > >  {
> > > > > +	xfs_defer_init_op_type(&xfs_extent_free_defer_type);
> > > > >  }
> > > > > 
> > > > > _______________________________________________
> > > > > xfs mailing list
> > > > > xfs@oss.sgi.com
> > > > > http://oss.sgi.com/mailman/listinfo/xfs
> > > 
> > > _______________________________________________
> > > xfs mailing list
> > > xfs@oss.sgi.com
> > > http://oss.sgi.com/mailman/listinfo/xfs
> > 
> > _______________________________________________
> > xfs mailing list
> > xfs@oss.sgi.com
> > http://oss.sgi.com/mailman/listinfo/xfs