2007-07-05 18:09:59

by Badari Pulavarty

[permalink] [raw]
Subject: [PATCH] ext2 statfs speed up

This is a patch that speeds up statfs. It is very simple -
the "overhead" calculation, which takes a huge amount of time
for large filesystems, never changes unless the size of the
filesystem itself changes. That means we can store it in memory
and only recalculate if the filesystem has been resized (almost
never).

It also fixes a minor problem that we never update the on-disk
superblock free blocks/inodes counts until the filesystem is
unmounted. While not fatal, we may as well update that on disk
when we have the information, and it makes things like debugfs
and dumpe2fs report a bit more accurate info.

Signed-off-by: Badari Pulavarty <[email protected]>
Signed-off-by: Andreas Dilger <[email protected]>

fs/ext2/super.c | 20 ++++++++++++++------
include/linux/ext2_fs_sb.h | 2 ++
2 files changed, 16 insertions(+), 6 deletions(-)

Index: linux-2.6.22-rc7/include/linux/ext2_fs_sb.h
===================================================================
--- linux-2.6.22-rc7.orig/include/linux/ext2_fs_sb.h 2007-07-01 12:54:24.000000000 -0700
+++ linux-2.6.22-rc7/include/linux/ext2_fs_sb.h 2007-07-04 22:02:01.000000000 -0700
@@ -33,6 +33,8 @@ struct ext2_sb_info {
unsigned long s_gdb_count; /* Number of group descriptor blocks */
unsigned long s_desc_per_block; /* Number of group descriptors per block */
unsigned long s_groups_count; /* Number of groups in the fs */
+ unsigned long s_overhead_last; /* Last calculated overhead */
+ unsigned long s_blocks_last; /* Last seen block count */
struct buffer_head * s_sbh; /* Buffer containing the super block */
struct ext2_super_block * s_es; /* Pointer to the super block in the buffer */
struct buffer_head ** s_group_desc;
Index: linux-2.6.22-rc7/fs/ext2/super.c
===================================================================
--- linux-2.6.22-rc7.orig/fs/ext2/super.c 2007-07-01 12:54:24.000000000 -0700
+++ linux-2.6.22-rc7/fs/ext2/super.c 2007-07-05 12:29:03.000000000 -0700
@@ -1099,15 +1099,18 @@ static int ext2_statfs (struct dentry *
struct super_block *sb = dentry->d_sb;
struct ext2_sb_info *sbi = EXT2_SB(sb);
struct ext2_super_block *es = sbi->s_es;
- unsigned long overhead;
- int i;
u64 fsid;

if (test_opt (sb, MINIX_DF))
- overhead = 0;
- else {
+ sbi->s_overhead_last = 0;
+ else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
+ unsigned long i, overhead = 0;
+ smp_rmb();
+
/*
- * Compute the overhead (FS structures)
+ * Compute the overhead (FS structures). This is constant
+ * for a given filesystem unless the number of block groups
+ * changes so we cache the previous value until it does.
*/

/*
@@ -1131,17 +1134,22 @@ static int ext2_statfs (struct dentry *
*/
overhead += (sbi->s_groups_count *
(2 + sbi->s_itb_per_group));
+ sbi->s_overhead_last = overhead;
+ smp_wmb();
+ sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
}

buf->f_type = EXT2_SUPER_MAGIC;
buf->f_bsize = sb->s_blocksize;
- buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
+ buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
buf->f_bfree = ext2_count_free_blocks(sb);
+ es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
buf->f_bavail = 0;
buf->f_files = le32_to_cpu(es->s_inodes_count);
buf->f_ffree = ext2_count_free_inodes(sb);
+ es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
buf->f_namelen = EXT2_NAME_LEN;
fsid = le64_to_cpup((void *)es->s_uuid) ^
le64_to_cpup((void *)es->s_uuid + sizeof(u64));


2007-07-05 21:06:53

by Andreas Dilger

[permalink] [raw]
Subject: Re: [PATCH] ext2 statfs speed up

On Jul 05, 2007 11:11 -0700, Badari Pulavarty wrote:
> @@ -1131,17 +1134,22 @@ static int ext2_statfs (struct dentry *
> buf->f_bfree = ext2_count_free_blocks(sb);
> + es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
> buf->f_ffree = ext2_count_free_inodes(sb);
> + es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);

Hmm, this is still sub-optimal. For ext3 and ext4 it just uses
percpu_counter_sum() instead of the slow ext*_count_free_blocks(), which
walks all of the groups. Not that this is a reason to hold this patch,
because at least we are removing 1/2 of the overhead for ext2.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

2007-07-06 02:07:19

by Badari Pulavarty

[permalink] [raw]
Subject: Re: [PATCH] ext2 statfs speed up

On Thu, 2007-07-05 at 15:06 -0600, Andreas Dilger wrote:
> On Jul 05, 2007 11:11 -0700, Badari Pulavarty wrote:
> > @@ -1131,17 +1134,22 @@ static int ext2_statfs (struct dentry *
> > buf->f_bfree = ext2_count_free_blocks(sb);
> > + es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
> > buf->f_ffree = ext2_count_free_inodes(sb);
> > + es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
>
> Hmm, this is still sub-optimal. For ext3 and ext4 it just uses
> percpu_counter_sum() instead of the slow ext*_count_free_blocks(), which
> walks all of the groups. Not that this is a reason to hold this patch,
> because at least we are removing 1/2 of the overhead for ext2.

Andrea,

I am wondering why we are not currently using percpu_counter_sum()
for ext2 ? I see that ext2 already has all the stuff it needs.
Can't I just do following ?

Thanks,
Badari

fs/ext2/super.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

Index: linux-2.6.22-rc7/fs/ext2/super.c
===================================================================
--- linux-2.6.22-rc7.orig/fs/ext2/super.c 2007-07-05 12:35:15.000000000 -0700
+++ linux-2.6.22-rc7/fs/ext2/super.c 2007-07-05 20:37:32.000000000 -0700
@@ -1142,13 +1142,13 @@ static int ext2_statfs (struct dentry *
buf->f_type = EXT2_SUPER_MAGIC;
buf->f_bsize = sb->s_blocksize;
buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
- buf->f_bfree = ext2_count_free_blocks(sb);
+ buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
buf->f_bavail = 0;
buf->f_files = le32_to_cpu(es->s_inodes_count);
- buf->f_ffree = ext2_count_free_inodes(sb);
+ buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
buf->f_namelen = EXT2_NAME_LEN;
fsid = le64_to_cpup((void *)es->s_uuid) ^

2007-07-06 04:52:27

by Andreas Dilger

[permalink] [raw]
Subject: Re: [PATCH] ext2 statfs speed up

On Jul 05, 2007 19:08 -0700, Badari Pulavarty wrote:
> On Thu, 2007-07-05 at 15:06 -0600, Andreas Dilger wrote:
> > On Jul 05, 2007 11:11 -0700, Badari Pulavarty wrote:
> > > @@ -1131,17 +1134,22 @@ static int ext2_statfs (struct dentry *
> > > buf->f_bfree = ext2_count_free_blocks(sb);
> > > + es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
> > > buf->f_ffree = ext2_count_free_inodes(sb);
> > > + es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
> >
> > Hmm, this is still sub-optimal. For ext3 and ext4 it just uses
> > percpu_counter_sum() instead of the slow ext*_count_free_blocks(), which
> > walks all of the groups. Not that this is a reason to hold this patch,
> > because at least we are removing 1/2 of the overhead for ext2.
>
> I am wondering why we are not currently using percpu_counter_sum()
> for ext2 ? I see that ext2 already has all the stuff it needs.
> Can't I just do following ?

Yes, that is exactly what I was asking above.

> fs/ext2/super.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> Index: linux-2.6.22-rc7/fs/ext2/super.c
> ===================================================================
> --- linux-2.6.22-rc7.orig/fs/ext2/super.c 2007-07-05 12:35:15.000000000 -0700
> +++ linux-2.6.22-rc7/fs/ext2/super.c 2007-07-05 20:37:32.000000000 -0700
> @@ -1142,13 +1142,13 @@ static int ext2_statfs (struct dentry *
> buf->f_type = EXT2_SUPER_MAGIC;
> buf->f_bsize = sb->s_blocksize;
> buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
> - buf->f_bfree = ext2_count_free_blocks(sb);
> + buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
> es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
> buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
> if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
> buf->f_bavail = 0;
> buf->f_files = le32_to_cpu(es->s_inodes_count);
> - buf->f_ffree = ext2_count_free_inodes(sb);
> + buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
> es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
> buf->f_namelen = EXT2_NAME_LEN;
> fsid = le64_to_cpup((void *)es->s_uuid) ^

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.