Re: [PATCH 05/11] quota: make per-sb hash array

[Jan Kara <jack@suse.cz>]

On Tue 05-10-10 22:20:21, Dmitry Monakhov wrote:
> @@ -1925,6 +1927,47 @@ int dquot_file_open(struct inode *inode, struct file *file)
>  }
>  EXPORT_SYMBOL(dquot_file_open);
>  
> +int dquot_hash_alloc(struct super_block *sb, int order)
> +{
> +	unsigned long nr_hash, i;
> +	struct hlist_head *hash_array;
> +	struct dquot_hash *dq_hash = &sb_dqopt(sb)->dq_hash;
> +
> +	hash_array = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
> +	if (!hash_array)
> +		return -ENOMEM;
> +
> +	/* Find power-of-two hlist_heads which can fit into allocation */
> +	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
> +	dq_hash->bits = 0;
> +	do {
> +		dq_hash->bits++;
> +	} while (nr_hash >> dq_hash->bits);
> +	dq_hash->bits--;
  We have ilog2() function for this these days...

> +void dquot_hash_destroy(struct super_block *sb)
> +{
> +
> +	struct dquot_hash * dq_hash = &sb_dqopt(sb)->dq_hash;
> +	unsigned long i, nr_hash = 1UL << dq_hash->bits;
> +	unsigned long addr = (unsigned long )dq_hash->head;
					   ^ superfluous space
> +
> +	for (i = 0; i < nr_hash; i++)
> +		BUG_ON(!hlist_empty(dq_hash->head + i));
  Just WARN_ON + return would be better. It's probably better to leak a
page of kernel memory than crash the whole kernel...

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR