Optimize microzaps

Microzap on-disk format does not include a hash tree, expecting one to
be built in RAM during mzap_open().  The built tree is linked to DMU
user buffer, freed when original DMU buffer is dropped from cache. I've
found that workloads accessing many large directories and having active
eviction from DMU cache spend significant amount of time building and
then destroying the trees.  I've also found that for each 64 byte mzap
element additional 64 byte tree element is allocated, that is a waste
of memory and CPU caches.

Improve memory efficiency of the hash tree by switching from AVL-tree
to B-tree.  It allows to save 24 bytes per element just on pointers.
Save 32 bits on mze_hash by storing only upper 32 bits since lower 32
bits are always zero for microzaps.  Save 16 bits on mze_chunkid, since
microzap can never have so many elements.  Respectively with the 16 bits
there can be no more than 16 bits of collision differentiators.  As
result, struct mzap_ent now drops from 48 (rounded to 64) to 8 bytes.

Tune B-trees for small data.  Reduce BTREE_CORE_ELEMS from 128 to 126
to allow struct zfs_btree_core in case of 8 byte elements to pack into
2KB instead of 4KB.  Aside of the microzaps it should also help 32bit
range trees.  Allow custom B-tree leaf size to reduce memmove() time.

Split zap_name_alloc() into zap_name_alloc() and zap_name_init_str().
It allows to not waste time allocating/freeing memory when processing
multiple names in a loop during mzap_open().

Together on a pool with 10K directories of 1800 files each and DMU
cache limited to 128MB this reduces time of `find . -name zzz` by 41%
from 7.63s to 4.47s, and saves additional ~30% of CPU time on the DMU
cache reclamation.

Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Matthew Ahrens <[email protected]>
Reviewed-by: Ryan Moeller <[email protected]>
Signed-off-by:	Alexander Motin <[email protected]>
Sponsored by:	iXsystems, Inc.
Closes openzfs#14039

include/sys/btree.h

@@ -65,7 +65,7 @@ extern "C" {
  * them, and increased memory overhead. Increasing these values results in
  * higher variance in operation time, and reduces memory overhead.
  */
-#define	BTREE_CORE_ELEMS	128
+#define	BTREE_CORE_ELEMS	126
 #define	BTREE_LEAF_SIZE		4096
 
 extern kmem_cache_t *zfs_btree_leaf_cache;
@@ -95,9 +95,6 @@ typedef struct zfs_btree_leaf {
 	uint8_t		btl_elems[];
 } zfs_btree_leaf_t;
 
-#define	BTREE_LEAF_ESIZE	(BTREE_LEAF_SIZE - \
-    offsetof(zfs_btree_leaf_t, btl_elems))
-
 typedef struct zfs_btree_index {
 	zfs_btree_hdr_t	*bti_node;
 	uint32_t	bti_offset;
@@ -109,14 +106,15 @@ typedef struct zfs_btree_index {
 } zfs_btree_index_t;
 
 typedef struct btree {
-	zfs_btree_hdr_t		*bt_root;
-	int64_t			bt_height;
+	int (*bt_compar) (const void *, const void *);
 	size_t			bt_elem_size;
+	size_t			bt_leaf_size;
 	uint32_t		bt_leaf_cap;
+	int32_t			bt_height;
 	uint64_t		bt_num_elems;
 	uint64_t		bt_num_nodes;
+	zfs_btree_hdr_t		*bt_root;
 	zfs_btree_leaf_t	*bt_bulk; // non-null if bulk loading
-	int (*bt_compar) (const void *, const void *);
 } zfs_btree_t;
 
 /*
@@ -132,9 +130,12 @@ void zfs_btree_fini(void);
  * compar - function to compare two nodes, it must return exactly: -1, 0, or +1
  *          -1 for <, 0 for ==, and +1 for >
  * size   - the value of sizeof(struct my_type)
+ * lsize  - custom leaf size
  */
 void zfs_btree_create(zfs_btree_t *, int (*) (const void *, const void *),
     size_t);
+void zfs_btree_create_custom(zfs_btree_t *, int (*)(const void *, const void *),
+    size_t, size_t);
 
 /*
  * Find a node with a matching value in the tree. Returns the matching node

include/sys/zap_impl.h

@@ -66,10 +66,9 @@ typedef struct mzap_phys {
 } mzap_phys_t;
 
 typedef struct mzap_ent {
-	avl_node_t mze_node;
-	int mze_chunkid;
-	uint64_t mze_hash;
-	uint32_t mze_cd; /* copy from mze_phys->mze_cd */
+	uint32_t mze_hash;
+	uint16_t mze_cd; /* copy from mze_phys->mze_cd */
+	uint16_t mze_chunkid;
 } mzap_ent_t;
 
 #define	MZE_PHYS(zap, mze) \
@@ -164,7 +163,7 @@ typedef struct zap {
 			int16_t zap_num_entries;
 			int16_t zap_num_chunks;
 			int16_t zap_alloc_next;
-			avl_tree_t zap_avl;
+			zfs_btree_t zap_tree;
 		} zap_micro;
 	} zap_u;
 } zap_t;
@@ -203,7 +202,7 @@ int zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
     zap_t **zapp);
 void zap_unlockdir(zap_t *zap, const void *tag);
 void zap_evict_sync(void *dbu);
-zap_name_t *zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt);
+zap_name_t *zap_name_alloc_str(zap_t *zap, const char *key, matchtype_t mt);
 void zap_name_free(zap_name_t *zn);
 int zap_hashbits(zap_t *zap);
 uint32_t zap_maxcd(zap_t *zap);

module/zfs/btree.c

@@ -102,7 +102,7 @@ zfs_btree_poison_node(zfs_btree_t *tree, zfs_btree_hdr_t *hdr)
 		(void) memset(leaf->btl_elems, 0x0f, hdr->bth_first * size);
 		(void) memset(leaf->btl_elems +
 		    (hdr->bth_first + hdr->bth_count) * size, 0x0f,
-		    BTREE_LEAF_ESIZE -
+		    tree->bt_leaf_size - offsetof(zfs_btree_leaf_t, btl_elems) -
 		    (hdr->bth_first + hdr->bth_count) * size);
 	}
 #endif
@@ -173,16 +173,44 @@ zfs_btree_fini(void)
 	kmem_cache_destroy(zfs_btree_leaf_cache);
 }
 
+static void *
+zfs_btree_leaf_alloc(zfs_btree_t *tree)
+{
+	if (tree->bt_leaf_size == BTREE_LEAF_SIZE)
+		return (kmem_cache_alloc(zfs_btree_leaf_cache, KM_SLEEP));
+	else
+		return (kmem_alloc(tree->bt_leaf_size, KM_SLEEP));
+}
+
+static void
+zfs_btree_leaf_free(zfs_btree_t *tree, void *ptr)
+{
+	if (tree->bt_leaf_size == BTREE_LEAF_SIZE)
+		return (kmem_cache_free(zfs_btree_leaf_cache, ptr));
+	else
+		return (kmem_free(ptr, tree->bt_leaf_size));
+}
+
 void
 zfs_btree_create(zfs_btree_t *tree, int (*compar) (const void *, const void *),
     size_t size)
 {
-	ASSERT3U(size, <=, BTREE_LEAF_ESIZE / 2);
+	zfs_btree_create_custom(tree, compar, size, BTREE_LEAF_SIZE);
+}
+
+void
+zfs_btree_create_custom(zfs_btree_t *tree,
+    int (*compar) (const void *, const void *),
+    size_t size, size_t lsize)
+{
+	size_t esize = lsize - offsetof(zfs_btree_leaf_t, btl_elems);
 
+	ASSERT3U(size, <=, esize / 2);
 	memset(tree, 0, sizeof (*tree));
 	tree->bt_compar = compar;
 	tree->bt_elem_size = size;
-	tree->bt_leaf_cap = P2ALIGN(BTREE_LEAF_ESIZE / size, 2);
+	tree->bt_leaf_size = lsize;
+	tree->bt_leaf_cap = P2ALIGN(esize / size, 2);
 	tree->bt_height = -1;
 	tree->bt_bulk = NULL;
 }
@@ -290,7 +318,7 @@ zfs_btree_find(zfs_btree_t *tree, const void *value, zfs_btree_index_t *where)
 
 	zfs_btree_core_t *node = NULL;
 	uint32_t child = 0;
-	uint64_t depth = 0;
+	uint32_t depth = 0;
 
 	/*
 	 * Iterate down the tree, finding which child the value should be in
@@ -811,8 +839,7 @@ zfs_btree_insert_into_leaf(zfs_btree_t *tree, zfs_btree_leaf_t *leaf,
 		move_count++;
 	}
 	tree->bt_num_nodes++;
-	zfs_btree_leaf_t *new_leaf = kmem_cache_alloc(zfs_btree_leaf_cache,
-	    KM_SLEEP);
+	zfs_btree_leaf_t *new_leaf = zfs_btree_leaf_alloc(tree);
 	zfs_btree_hdr_t *new_hdr = &new_leaf->btl_hdr;
 	new_hdr->bth_parent = leaf->btl_hdr.bth_parent;
 	new_hdr->bth_first = (tree->bt_bulk ? 0 : capacity / 4) +
@@ -1078,8 +1105,7 @@ zfs_btree_add_idx(zfs_btree_t *tree, const void *value,
 		ASSERT0(where->bti_offset);
 
 		tree->bt_num_nodes++;
-		zfs_btree_leaf_t *leaf = kmem_cache_alloc(zfs_btree_leaf_cache,
-		    KM_SLEEP);
+		zfs_btree_leaf_t *leaf = zfs_btree_leaf_alloc(tree);
 		tree->bt_root = &leaf->btl_hdr;
 		tree->bt_height++;
 
@@ -1378,7 +1404,7 @@ zfs_btree_node_destroy(zfs_btree_t *tree, zfs_btree_hdr_t *node)
 {
 	tree->bt_num_nodes--;
 	if (!zfs_btree_is_core(node)) {
-		kmem_cache_free(zfs_btree_leaf_cache, node);
+		zfs_btree_leaf_free(tree, node);
 	} else {
 		kmem_free(node, sizeof (zfs_btree_core_t) +
 		    BTREE_CORE_ELEMS * tree->bt_elem_size);
@@ -1991,7 +2017,7 @@ zfs_btree_verify_counts(zfs_btree_t *tree)
  */
 static uint64_t
 zfs_btree_verify_height_helper(zfs_btree_t *tree, zfs_btree_hdr_t *hdr,
-    int64_t height)
+    int32_t height)
 {
 	if (!zfs_btree_is_core(hdr)) {
 		VERIFY0(height);
@@ -2117,8 +2143,10 @@ zfs_btree_verify_poison_helper(zfs_btree_t *tree, zfs_btree_hdr_t *hdr)
 		zfs_btree_leaf_t *leaf = (zfs_btree_leaf_t *)hdr;
 		for (size_t i = 0; i < hdr->bth_first * size; i++)
 			VERIFY3U(leaf->btl_elems[i], ==, 0x0f);
+		size_t esize = tree->bt_leaf_size -
+		    offsetof(zfs_btree_leaf_t, btl_elems);
 		for (size_t i = (hdr->bth_first + hdr->bth_count) * size;
-		    i < BTREE_LEAF_ESIZE; i++)
+		    i < esize; i++)
 			VERIFY3U(leaf->btl_elems[i], ==, 0x0f);
 	} else {
 		zfs_btree_core_t *node = (zfs_btree_core_t *)hdr;

module/zfs/zap_leaf.c

@@ -646,7 +646,7 @@ zap_entry_create(zap_leaf_t *l, zap_name_t *zn, uint32_t cd,
  * form of the name.  But all callers have one of these on hand anyway,
  * so might as well take advantage.  A cleaner but slower interface
  * would accept neither argument, and compute the normalized name as
- * needed (using zap_name_alloc(zap_entry_read_name(zeh))).
+ * needed (using zap_name_alloc_str(zap_entry_read_name(zeh))).
  */
 boolean_t
 zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
@@ -667,7 +667,7 @@ zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
 			continue;
 
 		if (zn == NULL) {
-			zn = zap_name_alloc(zap, name, MT_NORMALIZE);
+			zn = zap_name_alloc_str(zap, name, MT_NORMALIZE);
 			allocdzn = B_TRUE;
 		}
 		if (zap_leaf_array_match(zeh->zeh_leaf, zn,