[Ocfs2-devel] [PATCH 24/39] ocfs2: Add caching info for refcount tree.

[Tao Ma <tao.ma@oracle.com>]

Joel Becker wrote:
> On Thu, Apr 30, 2009 at 06:58:36AM +0800, Tao Ma wrote:
>> +/* Caller must hold refcount tree lock. */
>>  int ocfs2_decrease_refcount(struct inode *inode, struct buffer_head *di_bh,
>>  			    handle_t *handle, u32 cpos, u32 len,
>>  			    struct ocfs2_alloc_context *meta_ac,
>> @@ -1528,11 +1544,19 @@ int ocfs2_decrease_refcount(struct inode *inode, struct buffer_head *di_bh,
>>  	struct ocfs2_inode_info *oi = OCFS2_I(inode);
>>  	struct buffer_head *ref_root_bh = NULL;
>>  	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
>> +	struct ocfs2_refcount_tree *tree;
> 
> 	If the caller holds the lock, don't they have the tree
> structure?  Can they pass it in?
I have thought of them before. But the most painful thing is in xattr 
part. There are so many places in xattr.c which will call 
ocfs2_decrease_refcount(in-body xattr set, in block xattr set, bucket 
xattr set, xattr final remove etc),  and I finally gave it up because 
there are so many functions that is needed to change. And what's more, 
if the xattr isn't refcounted, we passed so many NULL and used up too 
much stack.

>>  static int ocfs2_replace_cow(struct inode *inode,
>>  			     struct buffer_head *di_bh,
>>  			     struct buffer_head *ref_root_bh,
>> +			     struct ocfs2_caching_info *ref_ci,
>>  			     u32 cow_start, u32 cow_len,
>>  			     struct page **pages,
>>  			     int num_pages)
> 
> 	It would seem that you could pass the tree in here instead of
> just ref_ci.  Then if you need to change what the functions do with the
> refcount_tree fields, it's already there.
yeah, you are right. Will do. Thanks.

Regards,
Tao