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 <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>

Author: Darrick J. Wong

fs/xfs/libxfs/xfs_alloc_btree.c

@@ -20,6 +20,7 @@
 #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(
@@ -477,7 +478,8 @@ xfs_allocbt_init_common(
 
 	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) {
@@ -617,3 +619,22 @@ xfs_allocbt_calc_size(
 {
 	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

@@ -62,4 +62,7 @@ void xfs_allocbt_commit_staged_btree(struct xfs_btree_cur *cur,
 
 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

@@ -22,6 +22,8 @@
 #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.
  */
@@ -553,7 +555,7 @@ xfs_bmbt_init_cursor(
 	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);
 
@@ -675,3 +677,22 @@ xfs_bmbt_calc_size(
 {
 	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

@@ -112,4 +112,7 @@ extern unsigned long long xfs_bmbt_calc_size(struct xfs_mount *mp,
 
 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

@@ -22,11 +22,11 @@
 #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.
@@ -379,7 +379,7 @@ xfs_btree_del_cursor(
 		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);
 }
 
 /*
@@ -4962,3 +4962,42 @@ xfs_btree_has_more_records(
 	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

@@ -13,8 +13,6 @@ struct xfs_trans;
 struct xfs_ifork;
 struct xfs_perag;
 
-extern kmem_zone_t	*xfs_btree_cur_zone;
-
 /*
  * Generic key, ptr and record wrapper structures.
  *
@@ -92,12 +90,6 @@ uint32_t xfs_btree_magic(int crc, xfs_btnum_t btnum);
 #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 */
 	size_t	key_len;
@@ -238,6 +230,7 @@ struct xfs_btree_cur
 	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 */
@@ -592,19 +585,22 @@ xfs_btree_alloc_cursor(
 	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

@@ -22,6 +22,8 @@
 #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,
@@ -433,7 +435,7 @@ xfs_inobt_init_common(
 	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;
@@ -812,3 +814,22 @@ xfs_iallocbt_calc_size(
 {
 	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

@@ -77,4 +77,7 @@ void xfs_inobt_commit_staged_btree(struct xfs_btree_cur *cur,
 
 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

@@ -21,6 +21,8 @@
 #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)
@@ -323,7 +325,7 @@ xfs_refcountbt_init_common(
 	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;
@@ -514,3 +516,22 @@ xfs_refcountbt_calc_reserves(
 
 	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

@@ -67,4 +67,7 @@ void xfs_refcountbt_commit_staged_btree(struct xfs_btree_cur *cur,
 
 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__ */