RE: [PATCH] exfat: speed up iterate/lookup by fixing start point of traversing fat chain

[Sungjong Seo]

> When directory iterate and lookup is called, there is a buggy rewinding of
> start point for traversing fat chain to the directory entry's first
> cluster. This caused repeated fat chain traversing from the first entry of
> the directory that would show worse performance if huge amounts of files
> exist under single directory.
> Fix not to rewind, make continue from currently referenced cluster and dir
> entry.
> 
> Tested with 50,000 files under single directory / 256GB sdcard, with
> command "time ls -l > /dev/null",
> Before :     0m08.69s real     0m00.27s user     0m05.91s system
> After  :     0m07.01s real     0m00.25s user     0m04.34s system
> 
> Signed-off-by: Hyeongseok Kim <hyeongseok@xxxxxxxxx>
> ---
>  fs/exfat/dir.c | 42 +++++++++++++++++++++++++++++++++---------
>  1 file changed, 33 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/exfat/dir.c b/fs/exfat/dir.c index
> e1d5536de948..59d12eaa0649 100644
> --- a/fs/exfat/dir.c
> +++ b/fs/exfat/dir.c
> @@ -147,7 +147,7 @@ static int exfat_readdir(struct inode *inode, loff_t
> *cpos, struct exfat_dir_ent
>  					0);
> 
>  			*uni_name.name = 0x0;
> -			exfat_get_uniname_from_ext_entry(sb, &dir, dentry,
> +			exfat_get_uniname_from_ext_entry(sb, &clu, i,
>  				uni_name.name);

Looks good. Old code looks like a bug as you said.

>  			exfat_utf16_to_nls(sb, &uni_name,
>  				dir_entry->namebuf.lfn,
> @@ -911,10 +911,15 @@ enum {
>  };
> 
>  /*
> - * return values:
> - *   >= 0	: return dir entiry position with the name in dir
> - *   -ENOENT	: entry with the name does not exist
> - *   -EIO	: I/O error
> + * @ei:         inode info of directory
> + * @p_dir:      input as directory structure in which we search name
> + *              if found, output as a cluster dir where the name exists
> + *              if not found, not changed from input
> + * @num_entries entry size of p_uniname
> + * @return:
> + *   >= 0:      dir entry position from output p_dir.dir
> + *   -ENOENT:   entry with the name does not exist
> + *   -EIO:      I/O error
>   */
>  int exfat_find_dir_entry(struct super_block *sb, struct exfat_inode_info
> *ei,
>  		struct exfat_chain *p_dir, struct exfat_uni_name *p_uniname,
> @@ -925,14 +930,16 @@ int exfat_find_dir_entry(struct super_block *sb,
> struct exfat_inode_info *ei,
[snip]
  			hint_stat->clu = p_dir->dir;
>  			hint_stat->eidx = 0;
> -			return (dentry - num_ext);
> +
> +			exfat_chain_dup(p_dir, &tmp_clu);
> +			return dentry_in_cluster;
>  		}
>  	}
> 
>  	hint_stat->clu = clu.dir;
>  	hint_stat->eidx = dentry + 1;
> -	return dentry - num_ext;
> +
> +	exfat_chain_dup(p_dir, &tmp_clu);
> +	return dentry_in_cluster;
>  }

Changing the functionality of exfat find_dir_entry() will affect
exfat_find() and exfat_lookup(), breaking the concept of ei->dir.dir
which should have the starting cluster of its parent directory.

Well, is there any missing patch related to exfat_find()?
It would be nice to modify the caller of this function, exfat_find(),
so that this change in functionality doesn't affect other functions.

Thanks.

> 
>  int exfat_count_ext_entries(struct super_block *sb, struct exfat_chain
> *p_dir,
> --
> 2.27.0.83.g0313f36