Re: [PATCH v2 2/6] maple_tree: Introduce interfaces __mt_dup() and mtree_dup()

[Peng Zhang <zhangpeng.00@bytedance.com>]

在 2023/9/8 04:13, Liam R. Howlett 写道:
> * Peng Zhang <zhangpeng.00@bytedance.com> [230830 08:57]:
>> Introduce interfaces __mt_dup() and mtree_dup(), which are used to
>> duplicate a maple tree. Compared with traversing the source tree and
>> reinserting entry by entry in the new tree, it has better performance.
>> The difference between __mt_dup() and mtree_dup() is that mtree_dup()
>> handles locks internally.
> 
> __mt_dup() should be called mas_dup() to indicate the advanced interface
> which requires users to handle their own locks.
Ok, I'll change __mt_dup() to mas_dup().
> 
>>
>> Signed-off-by: Peng Zhang <zhangpeng.00@bytedance.com>
>> ---
>>   include/linux/maple_tree.h |   3 +
>>   lib/maple_tree.c           | 265 +++++++++++++++++++++++++++++++++++++
>>   2 files changed, 268 insertions(+)
>>
>> diff --git a/include/linux/maple_tree.h b/include/linux/maple_tree.h
>> index e41c70ac7744..44fe8a57ecbd 100644
>> --- a/include/linux/maple_tree.h
>> +++ b/include/linux/maple_tree.h
>> @@ -327,6 +327,9 @@ int mtree_store(struct maple_tree *mt, unsigned long index,
>>   		void *entry, gfp_t gfp);
>>   void *mtree_erase(struct maple_tree *mt, unsigned long index);
>>   
>> +int mtree_dup(struct maple_tree *mt, struct maple_tree *new, gfp_t gfp);
>> +int __mt_dup(struct maple_tree *mt, struct maple_tree *new, gfp_t gfp);
>> +
>>   void mtree_destroy(struct maple_tree *mt);
>>   void __mt_destroy(struct maple_tree *mt);
>>   
>> diff --git a/lib/maple_tree.c b/lib/maple_tree.c
>> index ef234cf02e3e..8f841682269c 100644
>> --- a/lib/maple_tree.c
>> +++ b/lib/maple_tree.c
>> @@ -6370,6 +6370,271 @@ void *mtree_erase(struct maple_tree *mt, unsigned long index)
>>   }
>>   EXPORT_SYMBOL(mtree_erase);
>>   
>> +/*
>> + * mas_dup_free() - Free a half-constructed tree.
> 
> Maybe "Free an incomplete duplication of a tree" ?
> 
>> + * @mas: Points to the last node of the half-constructed tree.
> 
> Your use of "Points to" seems to indicate someone knows you are talking
> about a "maple state that has a node pointing to".  Can this be made
> more clear?
> @mas: The maple state of a incomplete tree.
> 
> Then add a note that @mas->node points to the last successfully
> allocated node?
> 
> Or something along those lines.
Ok, I'll revise the comment.
> 
>> + *
>> + * This function frees all nodes starting from @mas->node in the reverse order
>> + * of mas_dup_build(). There is no need to hold the source tree lock at this
>> + * time.
>> + */
>> +static void mas_dup_free(struct ma_state *mas)
>> +{
>> +	struct maple_node *node;
>> +	enum maple_type type;
>> +	void __rcu **slots;
>> +	unsigned char count, i;
>> +
>> +	/* Maybe the first node allocation failed. */
>> +	if (!mas->node)
>> +		return;
>> +
>> +	while (!mte_is_root(mas->node)) {
>> +		mas_ascend(mas);
>> +
>> +		if (mas->offset) {
>> +			mas->offset--;
>> +			do {
>> +				mas_descend(mas);
>> +				mas->offset = mas_data_end(mas);
>> +			} while (!mte_is_leaf(mas->node));
> 
> Can you blindly descend and check !mte_is_leaf()?  What happens when the
> tree duplication fails at random internal nodes?  Maybe I missed how
> this cannot happen?
This cannot happen. Note the mas_ascend(mas) at the beginning of the
outermost loop.

> 
>> +
>> +			mas_ascend(mas);
>> +		}
>> +
>> +		node = mte_to_node(mas->node);
>> +		type = mte_node_type(mas->node);
>> +		slots = (void **)ma_slots(node, type);
>> +		count = mas_data_end(mas) + 1;
>> +		for (i = 0; i < count; i++)
>> +			((unsigned long *)slots)[i] &= ~MAPLE_NODE_MASK;
>> +
>> +		mt_free_bulk(count, slots);
>> +	}
> 
> 
>> +
>> +	node = mte_to_node(mas->node);
>> +	mt_free_one(node);
>> +}
>> +
>> +/*
>> + * mas_copy_node() - Copy a maple node and allocate child nodes.
> 
> if required. "..and allocate child nodes if required."
> 
>> + * @mas: Points to the source node.
>> + * @new_mas: Points to the new node.
>> + * @parent: The parent node of the new node.
>> + * @gfp: The GFP_FLAGS to use for allocations.
>> + *
>> + * Copy @mas->node to @new_mas->node, set @parent to be the parent of
>> + * @new_mas->node and allocate new child nodes for @new_mas->node.
>> + * If memory allocation fails, @mas is set to -ENOMEM.
>> + */
>> +static inline void mas_copy_node(struct ma_state *mas, struct ma_state *new_mas,
>> +		struct maple_node *parent, gfp_t gfp)
>> +{
>> +	struct maple_node *node = mte_to_node(mas->node);
>> +	struct maple_node *new_node = mte_to_node(new_mas->node);
>> +	enum maple_type type;
>> +	unsigned long val;
>> +	unsigned char request, count, i;
>> +	void __rcu **slots;
>> +	void __rcu **new_slots;
>> +
>> +	/* Copy the node completely. */
>> +	memcpy(new_node, node, sizeof(struct maple_node));
>> +
>> +	/* Update the parent node pointer. */
>> +	if (unlikely(ma_is_root(node)))
>> +		val = MA_ROOT_PARENT;
>> +	else
>> +		val = (unsigned long)node->parent & MAPLE_NODE_MASK;
> 
> If you treat the root as special and outside the loop, then you can
> avoid the check for root for every non-root node.  For root, you just
> need to copy and do this special parent thing before the main loop in
> mas_dup_build().  This will avoid an extra branch for each VMA over 14,
> so that would add up to a lot of instructions.
I'll handle the root node outside.
However, do you think it makes sense to have the parent of the root node
point to the struct maple_tree? I don't see it used anywhere.

> 
>> +
>> +	new_node->parent = ma_parent_ptr(val | (unsigned long)parent);
>> +
>> +	if (mte_is_leaf(mas->node))
>> +		return;
> 
> You are checking here and in mas_dup_build() for the leaf, splitting the
> function into parent assignment and allocate would allow you to check
> once. Copy could be moved to the main loop or with the parent setting,
> depending on how you handle the root suggestion above.
I'll try to reduce some checks.
> 
>> +
>> +	/* Allocate memory for child nodes. */
>> +	type = mte_node_type(mas->node);
>> +	new_slots = ma_slots(new_node, type);
>> +	request = mas_data_end(mas) + 1;
>> +	count = mt_alloc_bulk(gfp, request, new_slots);
>> +	if (unlikely(count < request)) {
>> +		if (count)
>> +			mt_free_bulk(count, new_slots);
> 
> The new_slots will still contain the addresses of the freed nodes.
> Don't you need to clear it here to avoid a double free?  Is there a
> test case for this in your testing?  Again, I may have missed how this
> is not possible..
It's impossible, because in mt_free_bulk(), the first thing to do with
the incoming node is to go up. We free all child nodes at the parent
node.

We guarantee that the node passed to mas_dup_free() is "clean".
mas_dup_free() also follows this so will not free children of this node.

The child nodes of this node cannot be freed in mt_free_bulk() because
the node is not completely constructed and data_end cannot be obtained.
data_end cannot be set on this node because the number of successfully
allocated child nodes can be 0.
> 
>> +		mas_set_err(mas, -ENOMEM);
>> +		return;
>> +	}
>> +
>> +	/* Restore node type information in slots. */
>> +	slots = ma_slots(node, type);
>> +	for (i = 0; i < count; i++)
>> +		((unsigned long *)new_slots)[i] |=
>> +			((unsigned long)mt_slot_locked(mas->tree, slots, i) &
>> +			MAPLE_NODE_MASK);
> 
> Can you expand this to multiple lines to make it more clear what is
> going on?
I will try to do that.

> 
>> +}
>> +
>> +/*
>> + * mas_dup_build() - Build a new maple tree from a source tree
>> + * @mas: The maple state of source tree.
>> + * @new_mas: The maple state of new tree.
>> + * @gfp: The GFP_FLAGS to use for allocations.
>> + *
>> + * This function builds a new tree in DFS preorder. If the memory allocation
>> + * fails, the error code -ENOMEM will be set in @mas, and @new_mas points to the
>> + * last node. mas_dup_free() will free the half-constructed tree.
>> + *
>> + * Note that the attributes of the two trees must be exactly the same, and the
>> + * new tree must be empty, otherwise -EINVAL will be returned.
>> + */
>> +static inline void mas_dup_build(struct ma_state *mas, struct ma_state *new_mas,
>> +		gfp_t gfp)
>> +{
>> +	struct maple_node *node, *parent;
> 
> Could parent be struct maple_pnode?
I'll rename it.

> 
>> +	struct maple_enode *root;
>> +	enum maple_type type;
>> +
>> +	if (unlikely(mt_attr(mas->tree) != mt_attr(new_mas->tree)) ||
>> +	    unlikely(!mtree_empty(new_mas->tree))) {
>> +		mas_set_err(mas, -EINVAL);
>> +		return;
>> +	}
>> +
>> +	mas_start(mas);
>> +	if (mas_is_ptr(mas) || mas_is_none(mas)) {
>> +		/*
>> +		 * The attributes of the two trees must be the same before this.
>> +		 * The following assignment makes them the same height.
>> +		 */
>> +		new_mas->tree->ma_flags = mas->tree->ma_flags;
>> +		rcu_assign_pointer(new_mas->tree->ma_root, mas->tree->ma_root);
>> +		return;
>> +	}
>> +
>> +	node = mt_alloc_one(gfp);
>> +	if (!node) {
>> +		new_mas->node = NULL;
> 
> We don't have checks around for node == NULL, MAS_NONE would be a safer
> choice.  It is unlikely that someone would dup the tree and fail then
> call something else, but I avoid setting node to NULL.
I will set it to MAS_NONE in the next version.

> 
>> +		mas_set_err(mas, -ENOMEM);
>> +		return;
>> +	}
>> +
>> +	type = mte_node_type(mas->node);
>> +	root = mt_mk_node(node, type);
>> +	new_mas->node = root;
>> +	new_mas->min = 0;
>> +	new_mas->max = ULONG_MAX;
>> +	parent = ma_mnode_ptr(new_mas->tree);
>> +
>> +	while (1) {
>> +		mas_copy_node(mas, new_mas, parent, gfp);
>> +
>> +		if (unlikely(mas_is_err(mas)))
>> +			return;
>> +
>> +		/* Once we reach a leaf, we need to ascend, or end the loop. */
>> +		if (mte_is_leaf(mas->node)) {
>> +			if (mas->max == ULONG_MAX) {
>> +				new_mas->tree->ma_flags = mas->tree->ma_flags;
>> +				rcu_assign_pointer(new_mas->tree->ma_root,
>> +						   mte_mk_root(root));
>> +				break;
> 
> If you move this to the end of the function, you can replace the same
> block above with a goto.  That will avoid breaking the line up.
I can do this, but it doesn't seem to make a difference.
> 
>> +			}
>> +
>> +			do {
>> +				/*
>> +				 * Must not at the root node, because we've
>> +				 * already end the loop when we reach the last
>> +				 * leaf.
>> +				 */
> 
> I'm not sure what the comment above is trying to say.  Do you mean "This
> won't reach the root node because the loop will break when the last leaf
> is hit"?  I don't think that is accurate.. it will hit the root node but
> not the end of the root node, right?  Anyways, the comment isn't clear
> so please have a look.
Yes, it will hit the root node but not the end of the root node. I'll
fix this comment. Thanks.

> 
>> +				mas_ascend(mas);
>> +				mas_ascend(new_mas);
>> +			} while (mas->offset == mas_data_end(mas));
>> +
>> +			mas->offset++;
>> +			new_mas->offset++;
>> +		}
>> +
>> +		mas_descend(mas);
>> +		parent = mte_to_node(new_mas->node);
>> +		mas_descend(new_mas);
>> +		mas->offset = 0;
>> +		new_mas->offset = 0;
>> +	}
>> +}
>> +
>> +/**
>> + * __mt_dup(): Duplicate a maple tree
>> + * @mt: The source maple tree
>> + * @new: The new maple tree
>> + * @gfp: The GFP_FLAGS to use for allocations
>> + *
>> + * This function duplicates a maple tree using a faster method than traversing
>> + * the source tree and inserting entries into the new tree one by one.
> 
> Can you make this comment more about what your code does instead of the
> "one by one" description?
> 
>> + * The user needs to ensure that the attributes of the source tree and the new
>> + * tree are the same, and the new tree needs to be an empty tree, otherwise
>> + * -EINVAL will be returned.
>> + * Note that the user needs to manually lock the source tree and the new tree.
>> + *
>> + * Return: 0 on success, -ENOMEM if memory could not be allocated, -EINVAL If
>> + * the attributes of the two trees are different or the new tree is not an empty
>> + * tree.
>> + */
>> +int __mt_dup(struct maple_tree *mt, struct maple_tree *new, gfp_t gfp)
>> +{
>> +	int ret = 0;
>> +	MA_STATE(mas, mt, 0, 0);
>> +	MA_STATE(new_mas, new, 0, 0);
>> +
>> +	mas_dup_build(&mas, &new_mas, gfp);
>> +
>> +	if (unlikely(mas_is_err(&mas))) {
>> +		ret = xa_err(mas.node);
>> +		if (ret == -ENOMEM)
>> +			mas_dup_free(&new_mas);
>> +	}
>> +
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL(__mt_dup);
>> +
>> +/**
>> + * mtree_dup(): Duplicate a maple tree
>> + * @mt: The source maple tree
>> + * @new: The new maple tree
>> + * @gfp: The GFP_FLAGS to use for allocations
>> + *
>> + * This function duplicates a maple tree using a faster method than traversing
>> + * the source tree and inserting entries into the new tree one by one.
> 
> Again, it's more interesting to state it uses the DFS preorder copy.
> 
> It is also worth mentioning the superior allocation behaviour since that
> is a desirable trait for many.  In fact, you should add the allocation
> behaviour in your cover letter.
Okay, I will describe more in the next version.

> 
>> + * The user needs to ensure that the attributes of the source tree and the new
>> + * tree are the same, and the new tree needs to be an empty tree, otherwise
>> + * -EINVAL will be returned.
>> + *
>> + * Return: 0 on success, -ENOMEM if memory could not be allocated, -EINVAL If
>> + * the attributes of the two trees are different or the new tree is not an empty
>> + * tree.
>> + */
>> +int mtree_dup(struct maple_tree *mt, struct maple_tree *new, gfp_t gfp)
>> +{
>> +	int ret = 0;
>> +	MA_STATE(mas, mt, 0, 0);
>> +	MA_STATE(new_mas, new, 0, 0);
>> +
>> +	mas_lock(&new_mas);
>> +	mas_lock(&mas);
>> +
>> +	mas_dup_build(&mas, &new_mas, gfp);
>> +	mas_unlock(&mas);
>> +
>> +	if (unlikely(mas_is_err(&mas))) {
>> +		ret = xa_err(mas.node);
>> +		if (ret == -ENOMEM)
>> +			mas_dup_free(&new_mas);
>> +	}
>> +
>> +	mas_unlock(&new_mas);
>> +
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL(mtree_dup);
>> +
>>   /**
>>    * __mt_destroy() - Walk and free all nodes of a locked maple tree.
>>    * @mt: The maple tree
>> -- 
>> 2.20.1
>>