xfs: use separate btree cursor cache for each btree type

Now that we have the infrastructure to track the max possible height of
each btree type, we can create a separate slab cache for cursors of each
type of btree. For smaller indices like the free space btrees, this
means that we can pack more cursors into a slab page, improving slab
utilization.

Signed-off-by: Darrick J. Wong <[e-mail]>
Reviewed-by: Dave Chinner <[e-mail]>

fs/xfs/libxfs/xfs_alloc_btree.c

···
 #include "xfs_trans.h"
 #include "xfs_ag.h"
 
+static kmem_zone_t	*xfs_allocbt_cur_cache;
 
 STATIC struct xfs_btree_cur *
 xfs_allocbt_dup_cursor(
···
 
 	ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
 
-	cur = xfs_btree_alloc_cursor(mp, tp, btnum, mp->m_alloc_maxlevels);
+	cur = xfs_btree_alloc_cursor(mp, tp, btnum, mp->m_alloc_maxlevels,
+			xfs_allocbt_cur_cache);
 	cur->bc_ag.abt.active = false;
 
 	if (btnum == XFS_BTNUM_CNT) {
···
 	unsigned long long	len)
 {
 	return xfs_btree_calc_size(mp->m_alloc_mnr, len);
+}
+
+int __init
+xfs_allocbt_init_cur_cache(void)
+{
+	xfs_allocbt_cur_cache = kmem_cache_create("xfs_bnobt_cur",
+			xfs_btree_cur_sizeof(xfs_allocbt_maxlevels_ondisk()),
+			0, 0, NULL);
+
+	if (!xfs_allocbt_cur_cache)
+		return -ENOMEM;
+	return 0;
+}
+
+void
+xfs_allocbt_destroy_cur_cache(void)
+{
+	kmem_cache_destroy(xfs_allocbt_cur_cache);
+	xfs_allocbt_cur_cache = NULL;
 }

fs/xfs/libxfs/xfs_alloc_btree.h

···
 
 unsigned int xfs_allocbt_maxlevels_ondisk(void);
 
+int __init xfs_allocbt_init_cur_cache(void);
+void xfs_allocbt_destroy_cur_cache(void);
+
 #endif	/* __XFS_ALLOC_BTREE_H__ */

fs/xfs/libxfs/xfs_bmap_btree.c

···
 #include "xfs_trace.h"
 #include "xfs_rmap.h"
 
+static kmem_zone_t	*xfs_bmbt_cur_cache;
+
 /*
  * Convert on-disk form of btree root to in-memory form.
  */
···
 	ASSERT(whichfork != XFS_COW_FORK);
 
 	cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
-			mp->m_bm_maxlevels[whichfork]);
+			mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
 	cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
 
···
 	unsigned long long	len)
 {
 	return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
+}
+
+int __init
+xfs_bmbt_init_cur_cache(void)
+{
+	xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
+			xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
+			0, 0, NULL);
+
+	if (!xfs_bmbt_cur_cache)
+		return -ENOMEM;
+	return 0;
+}
+
+void
+xfs_bmbt_destroy_cur_cache(void)
+{
+	kmem_cache_destroy(xfs_bmbt_cur_cache);
+	xfs_bmbt_cur_cache = NULL;
 }

fs/xfs/libxfs/xfs_bmap_btree.h

···
 
 unsigned int xfs_bmbt_maxlevels_ondisk(void);
 
+int __init xfs_bmbt_init_cur_cache(void);
+void xfs_bmbt_destroy_cur_cache(void);
+
 #endif	/* __XFS_BMAP_BTREE_H__ */

fs/xfs/libxfs/xfs_btree.c

···
 #include "xfs_log.h"
 #include "xfs_btree_staging.h"
 #include "xfs_ag.h"
-
-/*
- * Cursor allocation zone.
- */
-kmem_zone_t	*xfs_btree_cur_zone;
+#include "xfs_alloc_btree.h"
+#include "xfs_ialloc_btree.h"
+#include "xfs_bmap_btree.h"
+#include "xfs_rmap_btree.h"
+#include "xfs_refcount_btree.h"
 
 /*
  * Btree magic numbers.
···
 		kmem_free(cur->bc_ops);
 	if (!(cur->bc_flags & XFS_BTREE_LONG_PTRS) && cur->bc_ag.pag)
 		xfs_perag_put(cur->bc_ag.pag);
-	kmem_cache_free(xfs_btree_cur_zone, cur);
+	kmem_cache_free(cur->bc_cache, cur);
 }
 
 /*
···
 		return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
 	else
 		return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);
+}
+
+/* Set up all the btree cursor caches. */
+int __init
+xfs_btree_init_cur_caches(void)
+{
+	int		error;
+
+	error = xfs_allocbt_init_cur_cache();
+	if (error)
+		return error;
+	error = xfs_inobt_init_cur_cache();
+	if (error)
+		goto err;
+	error = xfs_bmbt_init_cur_cache();
+	if (error)
+		goto err;
+	error = xfs_rmapbt_init_cur_cache();
+	if (error)
+		goto err;
+	error = xfs_refcountbt_init_cur_cache();
+	if (error)
+		goto err;
+
+	return 0;
+err:
+	xfs_btree_destroy_cur_caches();
+	return error;
+}
+
+/* Destroy all the btree cursor caches, if they've been allocated. */
+void
+xfs_btree_destroy_cur_caches(void)
+{
+	xfs_allocbt_destroy_cur_cache();
+	xfs_inobt_destroy_cur_cache();
+	xfs_bmbt_destroy_cur_cache();
+	xfs_rmapbt_destroy_cur_cache();
+	xfs_refcountbt_destroy_cur_cache();
 }

fs/xfs/libxfs/xfs_btree.h

···
 struct xfs_ifork;
 struct xfs_perag;
 
-extern kmem_zone_t	*xfs_btree_cur_zone;
-
 /*
  * Generic key, ptr and record wrapper structures.
  *
···
 	XFS_STATS_INC_OFF((cur)->bc_mp, (cur)->bc_statoff + __XBTS_ ## stat)
 #define XFS_BTREE_STATS_ADD(cur, stat, val)	\
 	XFS_STATS_ADD_OFF((cur)->bc_mp, (cur)->bc_statoff + __XBTS_ ## stat, val)
-
-/*
- * The btree cursor zone hands out cursors that can handle up to this many
- * levels.  This is the known maximum for all btree types.
- */
-#define XFS_BTREE_CUR_CACHE_MAXLEVELS	(9)
 
 struct xfs_btree_ops {
 	/* size of the key and record structures */
···
 	struct xfs_trans	*bc_tp;	/* transaction we're in, if any */
 	struct xfs_mount	*bc_mp;	/* file system mount struct */
 	const struct xfs_btree_ops *bc_ops;
+	kmem_zone_t		*bc_cache; /* cursor cache */
 	unsigned int		bc_flags; /* btree features - below */
 	xfs_btnum_t		bc_btnum; /* identifies which btree type */
 	union xfs_btree_irec	bc_rec;	/* current insert/search record value */
···
 	struct xfs_mount	*mp,
 	struct xfs_trans	*tp,
 	xfs_btnum_t		btnum,
-	uint8_t			maxlevels)
+	uint8_t			maxlevels,
+	kmem_zone_t		*cache)
 {
 	struct xfs_btree_cur	*cur;
 
-	ASSERT(maxlevels <= XFS_BTREE_CUR_CACHE_MAXLEVELS);
-
-	cur = kmem_cache_zalloc(xfs_btree_cur_zone, GFP_NOFS | __GFP_NOFAIL);
+	cur = kmem_cache_zalloc(cache, GFP_NOFS | __GFP_NOFAIL);
 	cur->bc_tp = tp;
 	cur->bc_mp = mp;
 	cur->bc_btnum = btnum;
 	cur->bc_maxlevels = maxlevels;
+	cur->bc_cache = cache;
 
 	return cur;
 }
+
+int __init xfs_btree_init_cur_caches(void);
+void xfs_btree_destroy_cur_caches(void);
 
 #endif	/* __XFS_BTREE_H__ */

fs/xfs/libxfs/xfs_ialloc_btree.c

···
 #include "xfs_rmap.h"
 #include "xfs_ag.h"
 
+static kmem_zone_t	*xfs_inobt_cur_cache;
+
 STATIC int
 xfs_inobt_get_minrecs(
 	struct xfs_btree_cur	*cur,
···
 	struct xfs_btree_cur	*cur;
 
 	cur = xfs_btree_alloc_cursor(mp, tp, btnum,
-			M_IGEO(mp)->inobt_maxlevels);
+			M_IGEO(mp)->inobt_maxlevels, xfs_inobt_cur_cache);
 	if (btnum == XFS_BTNUM_INO) {
 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_ibt_2);
 		cur->bc_ops = &xfs_inobt_ops;
···
 	unsigned long long	len)
 {
 	return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);
+}
+
+int __init
+xfs_inobt_init_cur_cache(void)
+{
+	xfs_inobt_cur_cache = kmem_cache_create("xfs_inobt_cur",
+			xfs_btree_cur_sizeof(xfs_inobt_maxlevels_ondisk()),
+			0, 0, NULL);
+
+	if (!xfs_inobt_cur_cache)
+		return -ENOMEM;
+	return 0;
+}
+
+void
+xfs_inobt_destroy_cur_cache(void)
+{
+	kmem_cache_destroy(xfs_inobt_cur_cache);
+	xfs_inobt_cur_cache = NULL;
 }

fs/xfs/libxfs/xfs_ialloc_btree.h

···
 
 unsigned int xfs_iallocbt_maxlevels_ondisk(void);
 
+int __init xfs_inobt_init_cur_cache(void);
+void xfs_inobt_destroy_cur_cache(void);
+
 #endif	/* __XFS_IALLOC_BTREE_H__ */

fs/xfs/libxfs/xfs_refcount_btree.c

···
 #include "xfs_rmap.h"
 #include "xfs_ag.h"
 
+static kmem_zone_t	*xfs_refcountbt_cur_cache;
+
 static struct xfs_btree_cur *
 xfs_refcountbt_dup_cursor(
 	struct xfs_btree_cur	*cur)
···
 	ASSERT(pag->pag_agno < mp->m_sb.sb_agcount);
 
 	cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_REFC,
-			mp->m_refc_maxlevels);
+			mp->m_refc_maxlevels, xfs_refcountbt_cur_cache);
 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2);
 
 	cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
···
 	*used += tree_len;
 
 	return error;
+}
+
+int __init
+xfs_refcountbt_init_cur_cache(void)
+{
+	xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur",
+			xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()),
+			0, 0, NULL);
+
+	if (!xfs_refcountbt_cur_cache)
+		return -ENOMEM;
+	return 0;
+}
+
+void
+xfs_refcountbt_destroy_cur_cache(void)
+{
+	kmem_cache_destroy(xfs_refcountbt_cur_cache);
+	xfs_refcountbt_cur_cache = NULL;
 }

fs/xfs/libxfs/xfs_refcount_btree.h

···
 
 unsigned int xfs_refcountbt_maxlevels_ondisk(void);
 
+int __init xfs_refcountbt_init_cur_cache(void);
+void xfs_refcountbt_destroy_cur_cache(void);
+
 #endif	/* __XFS_REFCOUNT_BTREE_H__ */

fs/xfs/libxfs/xfs_rmap_btree.c

···
 #include "xfs_ag.h"
 #include "xfs_ag_resv.h"
 
+static kmem_zone_t	*xfs_rmapbt_cur_cache;
+
 /*
  * Reverse map btree.
  *
···
 
 	/* Overlapping btree; 2 keys per pointer. */
 	cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_RMAP,
-			mp->m_rmap_maxlevels);
+			mp->m_rmap_maxlevels, xfs_rmapbt_cur_cache);
 	cur->bc_flags = XFS_BTREE_CRC_BLOCKS | XFS_BTREE_OVERLAPPING;
 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_rmap_2);
 	cur->bc_ops = &xfs_rmapbt_ops;
···
 	*used += tree_len;
 
 	return error;
+}
+
+int __init
+xfs_rmapbt_init_cur_cache(void)
+{
+	xfs_rmapbt_cur_cache = kmem_cache_create("xfs_rmapbt_cur",
+			xfs_btree_cur_sizeof(xfs_rmapbt_maxlevels_ondisk()),
+			0, 0, NULL);
+
+	if (!xfs_rmapbt_cur_cache)
+		return -ENOMEM;
+	return 0;
+}
+
+void
+xfs_rmapbt_destroy_cur_cache(void)
+{
+	kmem_cache_destroy(xfs_rmapbt_cur_cache);
+	xfs_rmapbt_cur_cache = NULL;
 }

fs/xfs/libxfs/xfs_rmap_btree.h

···
 
 unsigned int xfs_rmapbt_maxlevels_ondisk(void);
 
+int __init xfs_rmapbt_init_cur_cache(void);
+void xfs_rmapbt_destroy_cur_cache(void);
+
 #endif /* __XFS_RMAP_BTREE_H__ */

fs/xfs/xfs_super.c

···
 #include "xfs_reflink.h"
 #include "xfs_pwork.h"
 #include "xfs_ag.h"
+#include "xfs_btree.h"
 
 #include <linux/magic.h>
 #include <linux/fs_context.h>
···
 STATIC int __init
 xfs_init_zones(void)
 {
+	int		error;
+
 	xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
 						sizeof(struct xlog_ticket),
 						0, 0, NULL);
···
 	if (!xfs_bmap_free_item_zone)
 		goto out_destroy_log_ticket_zone;
 
-	xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
-			xfs_btree_cur_sizeof(XFS_BTREE_CUR_CACHE_MAXLEVELS),
-			0, 0, NULL);
-	if (!xfs_btree_cur_zone)
+	error = xfs_btree_init_cur_caches();
+	if (error)
 		goto out_destroy_bmap_free_item_zone;
 
 	xfs_da_state_zone = kmem_cache_create("xfs_da_state",
···
  out_destroy_da_state_zone:
 	kmem_cache_destroy(xfs_da_state_zone);
  out_destroy_btree_cur_zone:
-	kmem_cache_destroy(xfs_btree_cur_zone);
+	xfs_btree_destroy_cur_caches();
  out_destroy_bmap_free_item_zone:
 	kmem_cache_destroy(xfs_bmap_free_item_zone);
  out_destroy_log_ticket_zone:
···
 	kmem_cache_destroy(xfs_trans_zone);
 	kmem_cache_destroy(xfs_ifork_zone);
 	kmem_cache_destroy(xfs_da_state_zone);
-	kmem_cache_destroy(xfs_btree_cur_zone);
+	xfs_btree_destroy_cur_caches();
 	kmem_cache_destroy(xfs_bmap_free_item_zone);
 	kmem_cache_destroy(xfs_log_ticket_zone);
 }