[v4] erofs: support to readahead dirent blocks in erofs_readdir()

[PATCH v4] erofs: support to readahead dirent blocks in erofs_readdir()

Posted by Chao Yu 6 months, 3 weeks ago

This patch supports to readahead more blocks in erofs_readdir(), it can
enhance readdir performance in large direcotry.

readdir test in a large directory which contains 12000 sub-files.

		files_per_second
Before:		926385.54
After:		2380435.562

Meanwhile, let's introduces a new sysfs entry to control readahead
bytes to provide more flexible policy for readahead of readdir().
- location: /sys/fs/erofs/<disk>/dir_ra_bytes
- default value: 16384
- disable readahead: set the value to 0

Signed-off-by: Chao Yu <chao@kernel.org>
---
v4:
- clean up codes and comments
 Documentation/ABI/testing/sysfs-fs-erofs |  8 ++++++++
 fs/erofs/dir.c                           | 14 ++++++++++++++
 fs/erofs/internal.h                      |  4 ++++
 fs/erofs/super.c                         |  2 ++
 fs/erofs/sysfs.c                         |  2 ++
 5 files changed, 30 insertions(+)

diff --git a/Documentation/ABI/testing/sysfs-fs-erofs b/Documentation/ABI/testing/sysfs-fs-erofs
index bf3b6299c15e..85fa56ca092c 100644
--- a/Documentation/ABI/testing/sysfs-fs-erofs
+++ b/Documentation/ABI/testing/sysfs-fs-erofs
@@ -35,3 +35,11 @@ Description:	Used to set or show hardware accelerators in effect
 		and multiple accelerators are separated by '\n'.
 		Supported accelerator(s): qat_deflate.
 		Disable all accelerators with an empty string (echo > accel).
+
+What:		/sys/fs/erofs/<disk>/dir_ra_bytes
+Date:		July 2025
+Contact:	"Chao Yu" <chao@kernel.org>
+Description:	Used to set or show readahead bytes during readdir(), by
+		default the value is 16384.
+
+		- 0: disable readahead.
diff --git a/fs/erofs/dir.c b/fs/erofs/dir.c
index 3e4b38bec0aa..99745c272b60 100644
--- a/fs/erofs/dir.c
+++ b/fs/erofs/dir.c
@@ -47,8 +47,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 	struct inode *dir = file_inode(f);
 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	struct super_block *sb = dir->i_sb;
+	struct file_ra_state *ra = &f->f_ra;
 	unsigned long bsz = sb->s_blocksize;
 	unsigned int ofs = erofs_blkoff(sb, ctx->pos);
+	pgoff_t ra_pages = DIV_ROUND_UP_POW2(
+			EROFS_I_SB(dir)->dir_ra_bytes, PAGE_SIZE);
+	pgoff_t nr_pages = DIV_ROUND_UP_POW2(dir->i_size, PAGE_SIZE);
 	int err = 0;
 	bool initial = true;
 
@@ -63,6 +67,16 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 			break;
 		}
 
+		/* readahead blocks to enhance performance in large directory */
+		if (ra_pages) {
+			pgoff_t idx = DIV_ROUND_UP(ctx->pos, PAGE_SIZE);
+			pgoff_t pages = min(nr_pages - idx, ra_pages);
+
+			if (pages > 1 && !ra_has_index(ra, idx))
+				page_cache_sync_readahead(dir->i_mapping, ra,
+							f, idx, pages);
+		}
+
 		de = erofs_bread(&buf, dbstart, true);
 		if (IS_ERR(de)) {
 			erofs_err(sb, "failed to readdir of logical block %llu of nid %llu",
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index 0d19bde8c094..8b1372521790 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -157,6 +157,7 @@ struct erofs_sb_info {
 	/* sysfs support */
 	struct kobject s_kobj;		/* /sys/fs/erofs/<devname> */
 	struct completion s_kobj_unregister;
+	erofs_off_t dir_ra_bytes;
 
 	/* fscache support */
 	struct fscache_volume *volume;
@@ -238,6 +239,9 @@ EROFS_FEATURE_FUNCS(xattr_filter, compat, COMPAT_XATTR_FILTER)
 #define EROFS_I_BL_XATTR_BIT	(BITS_PER_LONG - 1)
 #define EROFS_I_BL_Z_BIT	(BITS_PER_LONG - 2)
 
+/* default readahead size of directories */
+#define EROFS_DIR_RA_BYTES	16384
+
 struct erofs_inode {
 	erofs_nid_t nid;
 
diff --git a/fs/erofs/super.c b/fs/erofs/super.c
index e1e9f06e8342..38fc4813a896 100644
--- a/fs/erofs/super.c
+++ b/fs/erofs/super.c
@@ -715,6 +715,8 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
 	if (err)
 		return err;
 
+	sbi->dir_ra_bytes = EROFS_DIR_RA_BYTES;
+
 	erofs_info(sb, "mounted with root inode @ nid %llu.", sbi->root_nid);
 	return 0;
 }
diff --git a/fs/erofs/sysfs.c b/fs/erofs/sysfs.c
index eed8797a193f..9d9f820a5621 100644
--- a/fs/erofs/sysfs.c
+++ b/fs/erofs/sysfs.c
@@ -65,12 +65,14 @@ EROFS_ATTR_FUNC(drop_caches, 0200);
 #ifdef CONFIG_EROFS_FS_ZIP_ACCEL
 EROFS_ATTR_FUNC(accel, 0644);
 #endif
+EROFS_ATTR_RW_UI(dir_ra_bytes, erofs_sb_info);
 
 static struct attribute *erofs_sb_attrs[] = {
 #ifdef CONFIG_EROFS_FS_ZIP
 	ATTR_LIST(sync_decompress),
 	ATTR_LIST(drop_caches),
 #endif
+	ATTR_LIST(dir_ra_bytes),
 	NULL,
 };
 ATTRIBUTE_GROUPS(erofs_sb);
-- 
2.49.0

Re: [PATCH v4] erofs: support to readahead dirent blocks in erofs_readdir()

Posted by Gao Xiang 6 months, 3 weeks ago


On 2025/7/21 10:13, Chao Yu wrote:
> This patch supports to readahead more blocks in erofs_readdir(), it can
> enhance readdir performance in large direcotry.
> 
> readdir test in a large directory which contains 12000 sub-files.
> 
> 		files_per_second
> Before:		926385.54
> After:		2380435.562
> 
> Meanwhile, let's introduces a new sysfs entry to control readahead
> bytes to provide more flexible policy for readahead of readdir().
> - location: /sys/fs/erofs/<disk>/dir_ra_bytes
> - default value: 16384
> - disable readahead: set the value to 0
> 
> Signed-off-by: Chao Yu <chao@kernel.org>
> ---
> v4:
> - clean up codes and comments
>   Documentation/ABI/testing/sysfs-fs-erofs |  8 ++++++++
>   fs/erofs/dir.c                           | 14 ++++++++++++++
>   fs/erofs/internal.h                      |  4 ++++
>   fs/erofs/super.c                         |  2 ++
>   fs/erofs/sysfs.c                         |  2 ++
>   5 files changed, 30 insertions(+)
> 
> diff --git a/Documentation/ABI/testing/sysfs-fs-erofs b/Documentation/ABI/testing/sysfs-fs-erofs
> index bf3b6299c15e..85fa56ca092c 100644
> --- a/Documentation/ABI/testing/sysfs-fs-erofs
> +++ b/Documentation/ABI/testing/sysfs-fs-erofs
> @@ -35,3 +35,11 @@ Description:	Used to set or show hardware accelerators in effect
>   		and multiple accelerators are separated by '\n'.
>   		Supported accelerator(s): qat_deflate.
>   		Disable all accelerators with an empty string (echo > accel).
> +
> +What:		/sys/fs/erofs/<disk>/dir_ra_bytes
> +Date:		July 2025
> +Contact:	"Chao Yu" <chao@kernel.org>
> +Description:	Used to set or show readahead bytes during readdir(), by
> +		default the value is 16384.
> +
> +		- 0: disable readahead.
> diff --git a/fs/erofs/dir.c b/fs/erofs/dir.c
> index 3e4b38bec0aa..99745c272b60 100644
> --- a/fs/erofs/dir.c
> +++ b/fs/erofs/dir.c
> @@ -47,8 +47,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
>   	struct inode *dir = file_inode(f);
>   	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
>   	struct super_block *sb = dir->i_sb;
> +	struct file_ra_state *ra = &f->f_ra;
>   	unsigned long bsz = sb->s_blocksize;
>   	unsigned int ofs = erofs_blkoff(sb, ctx->pos);
> +	pgoff_t ra_pages = DIV_ROUND_UP_POW2(
> +			EROFS_I_SB(dir)->dir_ra_bytes, PAGE_SIZE);
> +	pgoff_t nr_pages = DIV_ROUND_UP_POW2(dir->i_size, PAGE_SIZE);
>   	int err = 0;
>   	bool initial = true;
>   
> @@ -63,6 +67,16 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
>   			break;
>   		}
>   
> +		/* readahead blocks to enhance performance in large directory */
> +		if (ra_pages) {
> +			pgoff_t idx = DIV_ROUND_UP(ctx->pos, PAGE_SIZE);

Can we use DIV_ROUND_UP_POW2 here too? If it's okay,
I will update the patch manually when applied.

Otherwise it looks good to me,
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>

Thanks,
Gao Xiang

Re: [PATCH v4] erofs: support to readahead dirent blocks in erofs_readdir()

Posted by Chao Yu 6 months, 3 weeks ago

On 7/21/25 10:24, Gao Xiang wrote:
> 
> 
> On 2025/7/21 10:13, Chao Yu wrote:
>> This patch supports to readahead more blocks in erofs_readdir(), it can
>> enhance readdir performance in large direcotry.
>>
>> readdir test in a large directory which contains 12000 sub-files.
>>
>>         files_per_second
>> Before:        926385.54
>> After:        2380435.562
>>
>> Meanwhile, let's introduces a new sysfs entry to control readahead
>> bytes to provide more flexible policy for readahead of readdir().
>> - location: /sys/fs/erofs/<disk>/dir_ra_bytes
>> - default value: 16384
>> - disable readahead: set the value to 0
>>
>> Signed-off-by: Chao Yu <chao@kernel.org>
>> ---
>> v4:
>> - clean up codes and comments
>>   Documentation/ABI/testing/sysfs-fs-erofs |  8 ++++++++
>>   fs/erofs/dir.c                           | 14 ++++++++++++++
>>   fs/erofs/internal.h                      |  4 ++++
>>   fs/erofs/super.c                         |  2 ++
>>   fs/erofs/sysfs.c                         |  2 ++
>>   5 files changed, 30 insertions(+)
>>
>> diff --git a/Documentation/ABI/testing/sysfs-fs-erofs b/Documentation/ABI/testing/sysfs-fs-erofs
>> index bf3b6299c15e..85fa56ca092c 100644
>> --- a/Documentation/ABI/testing/sysfs-fs-erofs
>> +++ b/Documentation/ABI/testing/sysfs-fs-erofs
>> @@ -35,3 +35,11 @@ Description:    Used to set or show hardware accelerators in effect
>>           and multiple accelerators are separated by '\n'.
>>           Supported accelerator(s): qat_deflate.
>>           Disable all accelerators with an empty string (echo > accel).
>> +
>> +What:        /sys/fs/erofs/<disk>/dir_ra_bytes
>> +Date:        July 2025
>> +Contact:    "Chao Yu" <chao@kernel.org>
>> +Description:    Used to set or show readahead bytes during readdir(), by
>> +        default the value is 16384.
>> +
>> +        - 0: disable readahead.
>> diff --git a/fs/erofs/dir.c b/fs/erofs/dir.c
>> index 3e4b38bec0aa..99745c272b60 100644
>> --- a/fs/erofs/dir.c
>> +++ b/fs/erofs/dir.c
>> @@ -47,8 +47,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
>>       struct inode *dir = file_inode(f);
>>       struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
>>       struct super_block *sb = dir->i_sb;
>> +    struct file_ra_state *ra = &f->f_ra;
>>       unsigned long bsz = sb->s_blocksize;
>>       unsigned int ofs = erofs_blkoff(sb, ctx->pos);
>> +    pgoff_t ra_pages = DIV_ROUND_UP_POW2(
>> +            EROFS_I_SB(dir)->dir_ra_bytes, PAGE_SIZE);
>> +    pgoff_t nr_pages = DIV_ROUND_UP_POW2(dir->i_size, PAGE_SIZE);
>>       int err = 0;
>>       bool initial = true;
>>   @@ -63,6 +67,16 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
>>               break;
>>           }
>>   +        /* readahead blocks to enhance performance in large directory */
>> +        if (ra_pages) {
>> +            pgoff_t idx = DIV_ROUND_UP(ctx->pos, PAGE_SIZE);
> 
> Can we use DIV_ROUND_UP_POW2 here too? If it's okay,
> I will update the patch manually when applied.

Xiang,

Yeah, looks fine to me.

Please go ahead to update it while applying, thanks for your help. :)

Thanks,

> 
> Otherwise it looks good to me,
> Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
> 
> Thanks,
> Gao Xiang