diff mbox

btrfs ulist use rbtree instead

Message ID 1349341866-29320-1-git-send-email-zimilo@code-trick.com (mailing list archive)
State New, archived
Headers show

Commit Message

Rock Lee Oct. 4, 2012, 9:11 a.m. UTC
From: Rock <zimilo@code-trick.com>

---
 fs/btrfs/backref.c |   10 ++--
 fs/btrfs/qgroup.c  |   16 +++---
 fs/btrfs/send.c    |    2 +-
 fs/btrfs/ulist.c   |  154 +++++++++++++++++++++++++++++++++++++---------------
 fs/btrfs/ulist.h   |   45 ++++++++++++---
 5 files changed, 161 insertions(+), 66 deletions(-)

Comments

David Sterba Oct. 4, 2012, 9:26 a.m. UTC | #1
> @@ -207,16 +266,23 @@ EXPORT_SYMBOL(ulist_add);
>   * end is reached. No guarantee is made with respect to the order in which
>   * the elements are returned. They might neither be returned in order of
>   * addition nor in ascending order.
> - * It is allowed to call ulist_add during an enumeration. Newly added items
> - * are guaranteed to show up in the running enumeration.
>   */
>  struct ulist_node *ulist_next(struct ulist *ulist, struct ulist_iterator *uiter)

Quick observation:

If there's code relying on the behaviour stated in the removed part of
the comment, it will break. Have you verified this is not the case?


david
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Arne Jansen Oct. 4, 2012, 9:44 a.m. UTC | #2
On 04.10.2012 11:26, David Sterba wrote:
>> @@ -207,16 +266,23 @@ EXPORT_SYMBOL(ulist_add);
>>   * end is reached. No guarantee is made with respect to the order in which
>>   * the elements are returned. They might neither be returned in order of
>>   * addition nor in ascending order.
>> - * It is allowed to call ulist_add during an enumeration. Newly added items
>> - * are guaranteed to show up in the running enumeration.
>>   */
>>  struct ulist_node *ulist_next(struct ulist *ulist, struct ulist_iterator *uiter)
> 
> Quick observation:
> 
> If there's code relying on the behaviour stated in the removed part of
> the comment, it will break. Have you verified this is not the case?

It's a good thing to use rb-trees when the small inline cache is exhausted,
but of course it should keep the semantics. We heavily rely on the removed
part.
It should be possible to keep the semantics if the elements are also kept
in a linked list. As it inflates the size of struct ulist_node even more,
it might make sense to use a smaller struct for the inline cache to keep
the footprint low.

Also, a commit message might be good that explains the motivation for the
change. Have you done any measurements?

Thanks for working on this.

-Arne

> 
> 
> david
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Rock Lee Oct. 4, 2012, 2:28 p.m. UTC | #3
Detailed explanations for the patch:

     When the small inline cache is exhausted, created a new rbtree,

     and the new rbtree uses  original spaces the inline nodes placed for saving memory.

     By using the rbtree can gain a better performance when nnodes gets larger.

     
Sorry for I doest't did much more measurements, but the average lookup time increases slower then the original linear policy when nnodes goes larger.


For this is my first patch I submitted, I'm too excited to find something I can hack the kernel, however I didn't consider the whole thing.

I will continue to dive into the btrfs implementation, and work harder. 

:-)

- Rock

On 2012-10-4, at ??5:44, Arne Jansen <sensille@gmx.net> wrote:

> On 04.10.2012 11:26, David Sterba wrote:
>>> @@ -207,16 +266,23 @@ EXPORT_SYMBOL(ulist_add);
>>>  * end is reached. No guarantee is made with respect to the order in which
>>>  * the elements are returned. They might neither be returned in order of
>>>  * addition nor in ascending order.
>>> - * It is allowed to call ulist_add during an enumeration. Newly added items
>>> - * are guaranteed to show up in the running enumeration.
>>>  */
>>> struct ulist_node *ulist_next(struct ulist *ulist, struct ulist_iterator *uiter)
>> 
>> Quick observation:
>> 
>> If there's code relying on the behaviour stated in the removed part of
>> the comment, it will break. Have you verified this is not the case?
> 
> It's a good thing to use rb-trees when the small inline cache is exhausted,
> but of course it should keep the semantics. We heavily rely on the removed
> part.
> It should be possible to keep the semantics if the elements are also kept
> in a linked list. As it inflates the size of struct ulist_node even more,
> it might make sense to use a smaller struct for the inline cache to keep
> the footprint low.
> 
> Also, a commit message might be good that explains the motivation for the
> change. Have you done any measurements?
> 
> Thanks for working on this.
> 
> -Arne
> 
>> 
>> 
>> david
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/







--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c
index ff6475f..a5bebc8 100644
--- a/fs/btrfs/backref.c
+++ b/fs/btrfs/backref.c
@@ -360,7 +360,7 @@  static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 		}
 
 		/* we put the first parent into the ref at hand */
-		ULIST_ITER_INIT(&uiter);
+		ULIST_ITER_INIT(parents, &uiter);
 		node = ulist_next(parents, &uiter);
 		ref->parent = node ? node->val : 0;
 		ref->inode_list =
@@ -955,7 +955,7 @@  static void free_leaf_list(struct ulist *blocks)
 	struct extent_inode_elem *eie_next;
 	struct ulist_iterator uiter;
 
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(blocks, &uiter);
 	while ((node = ulist_next(blocks, &uiter))) {
 		if (!node->aux)
 			continue;
@@ -1038,7 +1038,7 @@  int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 	}
 
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(tmp, &uiter);
 	while (1) {
 		ret = find_parent_nodes(trans, fs_info, bytenr,
 					time_seq, tmp, *roots, NULL);
@@ -1395,13 +1395,13 @@  int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
 	if (ret)
 		goto out;
 
-	ULIST_ITER_INIT(&ref_uiter);
+	ULIST_ITER_INIT(refs, &ref_uiter);
 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
 		ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
 					   tree_mod_seq_elem.seq, &roots);
 		if (ret)
 			break;
-		ULIST_ITER_INIT(&root_uiter);
+		ULIST_ITER_INIT(roots, &root_uiter);
 		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
 			pr_debug("root %llu references leaf %llu, data list "
 				 "%#lx\n", root_node->val, ref_node->val,
diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
index b650155..a0aad87 100644
--- a/fs/btrfs/qgroup.c
+++ b/fs/btrfs/qgroup.c
@@ -1133,7 +1133,7 @@  int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 	seq = fs_info->qgroup_seq;
 	fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
 
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(roots, &uiter);
 	while ((unode = ulist_next(roots, &uiter))) {
 		struct ulist_node *tmp_unode;
 		struct ulist_iterator tmp_uiter;
@@ -1146,7 +1146,7 @@  int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 		ulist_reinit(tmp);
 						/* XXX id not needed */
 		ulist_add(tmp, qg->qgroupid, (unsigned long)qg, GFP_ATOMIC);
-		ULIST_ITER_INIT(&tmp_uiter);
+		ULIST_ITER_INIT(tmp, &tmp_uiter);
 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
 			struct btrfs_qgroup_list *glist;
 
@@ -1169,7 +1169,7 @@  int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 	 */
 	ulist_reinit(tmp);
 	ulist_add(tmp, qgroup->qgroupid, (unsigned long)qgroup, GFP_ATOMIC);
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(tmp, &uiter);
 	while ((unode = ulist_next(tmp, &uiter))) {
 		struct btrfs_qgroup *qg;
 		struct btrfs_qgroup_list *glist;
@@ -1197,7 +1197,7 @@  int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 	/*
 	 * step 3: walk again from old refs
 	 */
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(roots, &uiter);
 	while ((unode = ulist_next(roots, &uiter))) {
 		struct btrfs_qgroup *qg;
 		struct ulist_node *tmp_unode;
@@ -1209,7 +1209,7 @@  int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 
 		ulist_reinit(tmp);
 		ulist_add(tmp, qg->qgroupid, (unsigned long)qg, GFP_ATOMIC);
-		ULIST_ITER_INIT(&tmp_uiter);
+		ULIST_ITER_INIT(tmp, &tmp_uiter);
 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
 			struct btrfs_qgroup_list *glist;
 
@@ -1470,7 +1470,7 @@  int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
 	 */
 	ulist = ulist_alloc(GFP_ATOMIC);
 	ulist_add(ulist, qgroup->qgroupid, (unsigned long)qgroup, GFP_ATOMIC);
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(ulist, &uiter);
 	while ((unode = ulist_next(ulist, &uiter))) {
 		struct btrfs_qgroup *qg;
 		struct btrfs_qgroup_list *glist;
@@ -1498,7 +1498,7 @@  int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
 	/*
 	 * no limits exceeded, now record the reservation into all qgroups
 	 */
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(ulist, &uiter);
 	while ((unode = ulist_next(ulist, &uiter))) {
 		struct btrfs_qgroup *qg;
 
@@ -1542,7 +1542,7 @@  void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
 
 	ulist = ulist_alloc(GFP_ATOMIC);
 	ulist_add(ulist, qgroup->qgroupid, (unsigned long)qgroup, GFP_ATOMIC);
-	ULIST_ITER_INIT(&uiter);
+	ULIST_ITER_INIT(ulist, &uiter);
 	while ((unode = ulist_next(ulist, &uiter))) {
 		struct btrfs_qgroup *qg;
 		struct btrfs_qgroup_list *glist;
diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c
index fb5ffe9..d4a2254 100644
--- a/fs/btrfs/send.c
+++ b/fs/btrfs/send.c
@@ -2898,7 +2898,7 @@  verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
 	 * deletion and if it's finally possible to perform the rmdir now.
 	 * We also update the inode stats of the parent dirs here.
 	 */
-	ULIST_ITER_INIT(&uit);
+	ULIST_ITER_INIT(check_dirs, &uit);
 	while ((un = ulist_next(check_dirs, &uit))) {
 		if (un->val > sctx->cur_ino)
 			continue;
diff --git a/fs/btrfs/ulist.c b/fs/btrfs/ulist.c
index ab942f4..2040905 100644
--- a/fs/btrfs/ulist.c
+++ b/fs/btrfs/ulist.c
@@ -2,6 +2,8 @@ 
  * Copyright (C) 2011 STRATO AG
  * written by Arne Jansen <sensille@gmx.net>
  * Distributed under the GNU GPL license version 2.
+ *
+ * Copyright 2012 Rock Lee <zimilo@code-trick.com>
  */
 
 #include <linux/slab.h>
@@ -23,10 +25,10 @@ 
  *
  * ulist = ulist_alloc();
  * ulist_add(ulist, root);
- * ULIST_ITER_INIT(&uiter);
+ * ULIST_ITER_INIT(ulist, &uiter);
  *
  * while ((elem = ulist_next(ulist, &uiter)) {
- * 	for (all child nodes n in elem)
+ *	for (all child nodes n in elem)
  *		ulist_add(ulist, n);
  *	do something useful with the node;
  * }
@@ -51,8 +53,10 @@ 
 void ulist_init(struct ulist *ulist)
 {
 	ulist->nnodes = 0;
-	ulist->nodes = ulist->int_nodes;
 	ulist->nodes_alloced = ULIST_SIZE;
+	ulist->pools = NULL;
+	ulist->root = RB_ROOT;
+
 }
 EXPORT_SYMBOL(ulist_init);
 
@@ -65,13 +69,18 @@  EXPORT_SYMBOL(ulist_init);
  */
 void ulist_fini(struct ulist *ulist)
 {
-	/*
-	 * The first ULIST_SIZE elements are stored inline in struct ulist.
-	 * Only if more elements are alocated they need to be freed.
-	 */
-	if (ulist->nodes_alloced > ULIST_SIZE)
-		kfree(ulist->nodes);
-	ulist->nodes_alloced = 0;	/* in case ulist_fini is called twice */
+	if (ulist->pools) {
+		struct list_head *p, *n;
+		struct ulist_node_pool *pool;
+		list_for_each_safe(p, n, &(ulist->pools->list)) {
+			pool = list_entry(p, struct ulist_node_pool, list);
+			kfree(pool);
+		}
+		ulist->pools = NULL;
+	}
+	ulist->nnodes = 0;
+	ulist->nodes_alloced = ULIST_SIZE;
+	ulist->root = RB_ROOT;
 }
 EXPORT_SYMBOL(ulist_fini);
 
@@ -152,48 +161,98 @@  int ulist_add(struct ulist *ulist, u64 val, unsigned long aux,
 int ulist_add_merge(struct ulist *ulist, u64 val, unsigned long aux,
 		    unsigned long *old_aux, gfp_t gfp_mask)
 {
-	int i;
-
-	for (i = 0; i < ulist->nnodes; ++i) {
-		if (ulist->nodes[i].val == val) {
+	struct ulist_node *node;
+	struct ulist_node_pool *pool = NULL;
+	if (ulist->nnodes <= ULIST_SIZE) {
+		int i;
+		for (i = 0; i < ulist->nnodes; ++i) {
+			if (ulist->int_nodes[i].val == val) {
+				if (old_aux)
+					*old_aux = ulist->int_nodes[i].aux;
+				return 0;
+			}
+		}
+	} else {
+		node = __ulist_rbtree_search(ulist, val);
+		if (node) {
 			if (old_aux)
-				*old_aux = ulist->nodes[i].aux;
+				*old_aux = node->aux;
 			return 0;
 		}
 	}
 
-	if (ulist->nnodes >= ulist->nodes_alloced) {
-		u64 new_alloced = ulist->nodes_alloced + 128;
-		struct ulist_node *new_nodes;
-		void *old = NULL;
-
-		/*
-		 * if nodes_alloced == ULIST_SIZE no memory has been allocated
-		 * yet, so pass NULL to krealloc
-		 */
-		if (ulist->nodes_alloced > ULIST_SIZE)
-			old = ulist->nodes;
-
-		new_nodes = krealloc(old, sizeof(*new_nodes) * new_alloced,
-				     gfp_mask);
-		if (!new_nodes)
+	if (ulist->nodes_alloced == ulist->nnodes) {
+		pool = kmalloc(sizeof(*pool), gfp_mask);
+		if (!pool)
 			return -ENOMEM;
+		pool->free_idx = 0;
 
-		if (!old)
-			memcpy(new_nodes, ulist->int_nodes,
-			       sizeof(ulist->int_nodes));
+		if (unlikely(ulist->nodes_alloced == ULIST_SIZE)) {
+			int i;
+			for (i = 0; i < ULIST_SIZE; ++i)
+				__ulist_rbtree_add_node(ulist,
+							&ulist->int_nodes[i]);
+			ulist->pools = pool;
+		} else {
+			list_add_tail(&pool->list, &ulist->pools->list);
+		}
+		ulist->nodes_alloced += ULIST_NODE_POOL_SIZE;
+	}
 
-		ulist->nodes = new_nodes;
-		ulist->nodes_alloced = new_alloced;
+	if (ulist->nnodes >= ULIST_SIZE) {
+		pool = list_entry(ulist->pools->list.prev,
+				struct ulist_node_pool,
+				list);
+		node = &pool->nodes[pool->free_idx++];
+		node->val = val;
+		__ulist_rbtree_add_node(ulist, node);
+	} else {
+		node = &ulist->int_nodes[ulist->nnodes];
+		node->val = val;
 	}
-	ulist->nodes[ulist->nnodes].val = val;
-	ulist->nodes[ulist->nnodes].aux = aux;
+
+	node->aux = aux;
 	++ulist->nnodes;
-
 	return 1;
 }
 EXPORT_SYMBOL(ulist_add);
 
+
+struct ulist_node *__ulist_rbtree_search(struct ulist *ulist, u64 val)
+{
+	struct rb_node *node = ulist->root.rb_node;
+	struct ulist_node *v;
+	while (node) {
+		v = rb_entry(node, struct ulist_node, node);
+		if (v->val < val)
+			node = node->rb_left;
+		else if (v->val > val)
+			node = node->rb_right;
+		else
+			return v;
+	}
+	return NULL;
+}
+
+
+int __ulist_rbtree_add_node(struct ulist *ulist, struct ulist_node *node)
+{
+	struct rb_node **new = &(ulist->root.rb_node), *parent = NULL;
+	struct ulist_node *v;
+	while (*new) {
+		v = rb_entry(*new, struct ulist_node, node);
+		if (v->val < node->val)
+			new = &((*new)->rb_left);
+		else if (v->val > node->val)
+			new = &((*new)->rb_right);
+		else
+			return -EEXIST;
+	}
+	rb_link_node(&node->node, parent, new);
+	rb_insert_color(&node->node, &ulist->root);
+	return 0;
+}
+
 /**
  * ulist_next - iterate ulist
  * @ulist:	ulist to iterate
@@ -207,16 +266,23 @@  EXPORT_SYMBOL(ulist_add);
  * end is reached. No guarantee is made with respect to the order in which
  * the elements are returned. They might neither be returned in order of
  * addition nor in ascending order.
- * It is allowed to call ulist_add during an enumeration. Newly added items
- * are guaranteed to show up in the running enumeration.
  */
 struct ulist_node *ulist_next(struct ulist *ulist, struct ulist_iterator *uiter)
 {
 	if (ulist->nnodes == 0)
 		return NULL;
-	if (uiter->i < 0 || uiter->i >= ulist->nnodes)
-		return NULL;
-
-	return &ulist->nodes[uiter->i++];
+	if (ulist->nnodes <= ULIST_SIZE) {
+		if (uiter->d.i >= ulist->nnodes || uiter->d.i < 0)
+			return NULL;
+		return &ulist->int_nodes[uiter->d.i++];
+	} else {
+		struct ulist_node *node;
+		if (uiter->d.node == NULL)
+			return NULL;
+		node = rb_entry(uiter->d.node, struct ulist_node, node);
+		uiter->d.node = rb_next(uiter->d.node);
+		return node;
+	}
+	return NULL;
 }
 EXPORT_SYMBOL(ulist_next);
diff --git a/fs/btrfs/ulist.h b/fs/btrfs/ulist.h
index 21bdc8e..bae5812 100644
--- a/fs/btrfs/ulist.h
+++ b/fs/btrfs/ulist.h
@@ -3,11 +3,15 @@ 
  * written by Arne Jansen <sensille@gmx.net>
  * Distributed under the GNU GPL license version 2.
  *
+ * Copyright 2012 Rock Lee <zimilo@code-trick.com>
  */
 
 #ifndef __ULIST__
 #define __ULIST__
 
+#include <linux/list.h>
+#include <linux/rbtree.h>
+
 /*
  * ulist is a generic data structure to hold a collection of unique u64
  * values. The only operations it supports is adding to the list and
@@ -24,35 +28,53 @@ 
  */
 #define ULIST_SIZE 16
 
+/*
+ * number of elements statically allocated inside the pool
+ */
+#define ULIST_NODE_POOL_SIZE 128
+
 struct ulist_iterator {
-	int i;
+	union {
+		int i;
+		struct rb_node *node;
+	} d;
 };
 
 /*
  * element of the list
  */
 struct ulist_node {
+	struct rb_node node;    /* node for rbtree maintain */
 	u64 val;		/* value to store */
 	unsigned long aux;	/* auxiliary value saved along with the val */
 };
 
+struct ulist_node_pool {
+	struct list_head list;
+	unsigned int free_idx;
+	struct ulist_node nodes[ULIST_NODE_POOL_SIZE];
+};
+
 struct ulist {
 	/*
-	 * number of elements stored in list
+	 * number of elements stored in the list
 	 */
 	unsigned long nnodes;
 
 	/*
-	 * number of nodes we already have room for
+	 * number of nodes we can store in the list
 	 */
 	unsigned long nodes_alloced;
 
 	/*
-	 * pointer to the array storing the elements. The first ULIST_SIZE
-	 * elements are stored inline. In this case the it points to int_nodes.
-	 * After exceeding ULIST_SIZE, dynamic memory is allocated.
+	 * node pools for storing ulist_nodes
 	 */
-	struct ulist_node *nodes;
+	struct ulist_node_pool *pools;
+
+	/*
+	 * when exceeds the ULIST_SIZE, the root field is useful
+	 */
+	struct rb_root root;
 
 	/*
 	 * inline storage space for the first ULIST_SIZE entries
@@ -72,6 +94,13 @@  int ulist_add_merge(struct ulist *ulist, u64 val, unsigned long aux,
 struct ulist_node *ulist_next(struct ulist *ulist,
 			      struct ulist_iterator *uiter);
 
-#define ULIST_ITER_INIT(uiter) ((uiter)->i = 0)
+struct ulist_node *__ulist_rbtree_search(struct ulist *ulist, u64 val);
+int __ulist_rbtree_add_node(struct ulist *ulist, struct ulist_node *node);
+
 
+#define ULIST_ITER_INIT(ulist, uiter) \
+	if ((ulist)->nnodes <= ULIST_SIZE) \
+		(uiter)->d.i = 0;	  \
+	else				  \
+		(uiter)->d.node = rb_first(&((ulist)->root))
 #endif