[PATCH v3] exfat: speed up iterate/lookup by fixing start point of traversing cluster chain

[PATCH v3] exfat: speed up iterate/lookup by fixing start point of traversing cluster chain

[Hyeongseok Kim]

When directory iterate and lookup is called, there's a buggy rewinding
of start point for traversing cluster chain to the parent directory
entry's first cluster. This caused repeated cluster chain traversing
from the first entry of the parent directory that would show worse
performance if huge amounts of files exist under the parent 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@gmail.com>
---
 fs/exfat/dir.c      | 19 +++++++++++++------
 fs/exfat/exfat_fs.h |  2 +-
 fs/exfat/namei.c    |  9 ++++++++-
 3 files changed, 22 insertions(+), 8 deletions(-)

diff --git a/fs/exfat/dir.c b/fs/exfat/dir.c
index 7efb1c6d4808..c4523648472a 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);
 			exfat_utf16_to_nls(sb, &uni_name,
 				dir_entry->namebuf.lfn,
@@ -911,14 +911,19 @@ 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 parent directory
+ * @p_dir:      directory structure of parent directory
+ * @num_entries:entry size of p_uniname
+ * @hint_opt:   If p_uniname is found, filled with optimized dir/entry
+ *              for traversing cluster chain.
+ * @return:
+ *   >= 0:      file directory entry position where the name exists
+ *   -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,
-		int num_entries, unsigned int type)
+		int num_entries, unsigned int type, struct exfat_hint *hint_opt)
 {
 	int i, rewind = 0, dentry = 0, end_eidx = 0, num_ext = 0, len;
 	int order, step, name_len = 0;
@@ -995,6 +1000,8 @@ int exfat_find_dir_entry(struct super_block *sb, struct exfat_inode_info *ei,
 
 			if (entry_type == TYPE_FILE || entry_type == TYPE_DIR) {
 				step = DIRENT_STEP_FILE;
+				hint_opt->clu = clu.dir;
+				hint_opt->eidx = i;
 				if (type == TYPE_ALL || type == entry_type) {
 					num_ext = ep->dentry.file.num_ext;
 					step = DIRENT_STEP_STRM;
diff --git a/fs/exfat/exfat_fs.h b/fs/exfat/exfat_fs.h
index e77fe2f45cf2..1d6da61157c9 100644
--- a/fs/exfat/exfat_fs.h
+++ b/fs/exfat/exfat_fs.h
@@ -457,7 +457,7 @@ void exfat_update_dir_chksum_with_entry_set(struct exfat_entry_set_cache *es);
 int exfat_calc_num_entries(struct exfat_uni_name *p_uniname);
 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,
-		int num_entries, unsigned int type);
+		int num_entries, unsigned int type, struct exfat_hint *hint_opt);
 int exfat_alloc_new_dir(struct inode *inode, struct exfat_chain *clu);
 int exfat_find_location(struct super_block *sb, struct exfat_chain *p_dir,
 		int entry, sector_t *sector, int *offset);
diff --git a/fs/exfat/namei.c b/fs/exfat/namei.c
index 1f7b3dc66fcd..24b41103d1cc 100644
--- a/fs/exfat/namei.c
+++ b/fs/exfat/namei.c
@@ -596,6 +596,8 @@ static int exfat_find(struct inode *dir, struct qstr *qname,
 	struct exfat_inode_info *ei = EXFAT_I(dir);
 	struct exfat_dentry *ep, *ep2;
 	struct exfat_entry_set_cache *es;
+	/* for optimized dir & entry to prevent long traverse of cluster chain */
+	struct exfat_hint hint_opt;
 
 	if (qname->len == 0)
 		return -ENOENT;
@@ -619,7 +621,7 @@ static int exfat_find(struct inode *dir, struct qstr *qname,
 
 	/* search the file name for directories */
 	dentry = exfat_find_dir_entry(sb, ei, &cdir, &uni_name,
-			num_entries, TYPE_ALL);
+			num_entries, TYPE_ALL, &hint_opt);
 
 	if (dentry < 0)
 		return dentry; /* -error value */
@@ -628,6 +630,11 @@ static int exfat_find(struct inode *dir, struct qstr *qname,
 	info->entry = dentry;
 	info->num_subdirs = 0;
 
+	/* adjust cdir to the optimized value */
+	cdir.dir = hint_opt.clu;
+	if (cdir.flags & ALLOC_NO_FAT_CHAIN)
+		cdir.size -= dentry / sbi->dentries_per_clu;
+	dentry = hint_opt.eidx;
 	es = exfat_get_dentry_set(sb, &cdir, dentry, ES_2_ENTRIES);
 	if (!es)
 		return -EIO;
-- 
2.27.0.83.g0313f36

RE: [PATCH v3] exfat: speed up iterate/lookup by fixing start point of traversing cluster chain

[Sungjong Seo]

> When directory iterate and lookup is called, there's a buggy rewinding of
> start point for traversing cluster chain to the parent directory entry's
> first cluster. This caused repeated cluster chain traversing from the
> first entry of the parent directory that would show worse performance if
> huge amounts of files exist under the parent 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@gmail.com>

Looks good.
Thanks for your contribution.

Reviewed-by: Sungjong Seo <sj1557.seo@samsung.com>

> ---
>  fs/exfat/dir.c      | 19 +++++++++++++------
>  fs/exfat/exfat_fs.h |  2 +-
>  fs/exfat/namei.c    |  9 ++++++++-
>  3 files changed, 22 insertions(+), 8 deletions(-)

RE: [PATCH v3] exfat: speed up iterate/lookup by fixing start point of traversing cluster chain

[Namjae Jeon]

> > When directory iterate and lookup is called, there's a buggy rewinding
> > of start point for traversing cluster chain to the parent directory
> > entry's first cluster. This caused repeated cluster chain traversing
> > from the first entry of the parent directory that would show worse
> > performance if huge amounts of files exist under the parent 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@gmail.com>
> 
> Looks good.
> Thanks for your contribution.
> 
> Reviewed-by: Sungjong Seo <sj1557.seo@samsung.com>
Applied. Thanks!