The following series of patches add support creating and checking
filesystems with the FLEX_BG feature. This feature currently groups
meta-data from a series of groups at the beginning of a flex group in
order to improve performance during heavy meta-data operations.
Some light testing on meta-data filesystem and fsck times already show
some improvements.
This is still very experimental code I meant mostly as a prototype, but
comments are welcome.
-JRS
From: Jose R. Santos <[email protected]>
Reserve the INCOMPAT feature number for FLEX_BG.
Signed-off-by: Jose R. Santos <[email protected]>
--
lib/ext2fs/ext2_fs.h | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)
diff --git a/lib/ext2fs/ext2_fs.h b/lib/ext2fs/ext2_fs.h
index a316665..2394857 100644
--- a/lib/ext2fs/ext2_fs.h
+++ b/lib/ext2fs/ext2_fs.h
@@ -640,6 +640,7 @@ struct ext2_super_block {
#define EXT3_FEATURE_INCOMPAT_EXTENTS 0x0040
#define EXT4_FEATURE_INCOMPAT_64BIT 0x0080
#define EXT4_FEATURE_INCOMPAT_MMP 0x0100
+#define EXT4_FEATURE_INCOMPAT_FLEX_BG 0x0200
#define EXT2_FEATURE_COMPAT_SUPP 0
From: Jose R. Santos <[email protected]>
Allow FLEX_BG to be use as a feature option at mke2fs time.
Signed-off-by: Jose R. Santos <[email protected]>
--
lib/e2p/feature.c | 2 ++
lib/ext2fs/ext2fs.h | 6 ++++--
misc/mke2fs.c | 3 ++-
3 files changed, 8 insertions(+), 3 deletions(-)
diff --git a/lib/e2p/feature.c b/lib/e2p/feature.c
index fe7e65a..4bf5630 100644
--- a/lib/e2p/feature.c
+++ b/lib/e2p/feature.c
@@ -67,6 +67,8 @@ static struct feature feature_list[] = {
"extent" },
{ E2P_FEATURE_INCOMPAT, EXT4_FEATURE_INCOMPAT_64BIT,
"64bit" },
+ { E2P_FEATURE_INCOMPAT, EXT4_FEATURE_INCOMPAT_FLEX_BG,
+ "flex_bg"},
{ 0, 0, 0 },
};
diff --git a/lib/ext2fs/ext2fs.h b/lib/ext2fs/ext2fs.h
index 20c63c0..d1cda2f 100644
--- a/lib/ext2fs/ext2fs.h
+++ b/lib/ext2fs/ext2fs.h
@@ -441,12 +441,14 @@ typedef struct ext2_icount *ext2_icount_t;
EXT2_FEATURE_INCOMPAT_COMPRESSION|\
EXT3_FEATURE_INCOMPAT_JOURNAL_DEV|\
EXT2_FEATURE_INCOMPAT_META_BG|\
- EXT3_FEATURE_INCOMPAT_RECOVER)
+ EXT3_FEATURE_INCOMPAT_RECOVER|\
+ EXT4_FEATURE_INCOMPAT_FLEX_BG)
#else
#define EXT2_LIB_FEATURE_INCOMPAT_SUPP (EXT2_FEATURE_INCOMPAT_FILETYPE|\
EXT3_FEATURE_INCOMPAT_JOURNAL_DEV|\
EXT2_FEATURE_INCOMPAT_META_BG|\
- EXT3_FEATURE_INCOMPAT_RECOVER)
+ EXT3_FEATURE_INCOMPAT_RECOVER|\
+ EXT4_FEATURE_INCOMPAT_FLEX_BG)
#endif
#define EXT2_LIB_FEATURE_RO_COMPAT_SUPP (EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER|\
EXT2_FEATURE_RO_COMPAT_LARGE_FILE)
diff --git a/misc/mke2fs.c b/misc/mke2fs.c
index 0c6d4f3..0af92e2 100644
--- a/misc/mke2fs.c
+++ b/misc/mke2fs.c
@@ -873,7 +873,8 @@ static __u32 ok_features[3] = {
EXT2_FEATURE_COMPAT_LAZY_BG, /* Compat */
EXT2_FEATURE_INCOMPAT_FILETYPE| /* Incompat */
EXT3_FEATURE_INCOMPAT_JOURNAL_DEV|
- EXT2_FEATURE_INCOMPAT_META_BG,
+ EXT2_FEATURE_INCOMPAT_META_BG|
+ EXT4_FEATURE_INCOMPAT_FLEX_BG,
EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER /* R/O compat */
};
From: Jose R. Santos <[email protected]>
Relax group descriptor checking.
In order for tools such as dump2efs, e2fsck and debugfs to open a ext4
filesystem with FLEX_BG feature enable, some descriptor checking needs
to be relaxed. This patch changes the group desciptor checking so
that bitmaps and inode tables can be located anywhere in the
partitions block range.
Eventually, a more thorough check would restrict bitmaps and inode
tables to be located at the beginning of a flex block group range.
Since the super block does not currently know about the number of
groups per flex group, this will do for now.
Signed-off-by: Jose R. Santos <[email protected]>
--
e2fsck/super.c | 10 ++++++++--
lib/ext2fs/check_desc.c | 10 ++++++++--
2 files changed, 16 insertions(+), 4 deletions(-)
diff --git a/e2fsck/super.c b/e2fsck/super.c
index 00a131c..8e58e5c 100644
--- a/e2fsck/super.c
+++ b/e2fsck/super.c
@@ -578,8 +578,14 @@ void check_super_block(e2fsck_t ctx)
for (i = 0, gd=fs->group_desc; i < fs->group_desc_count; i++, gd++) {
pctx.group = i;
- first_block = ext2fs_group_first_block(fs, i);
- last_block = ext2fs_group_last_block(fs, i);
+ if (EXT2_HAS_INCOMPAT_FEATURE (fs->super,
+ EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
+ first_block = fs->super->s_first_data_block;
+ last_block = fs->super->s_blocks_count;
+ } else {
+ first_block = ext2fs_group_first_block(fs, i);
+ last_block = ext2fs_group_last_block(fs, i);
+ }
if ((gd->bg_block_bitmap < first_block) ||
(gd->bg_block_bitmap > last_block)) {
diff --git a/lib/ext2fs/check_desc.c b/lib/ext2fs/check_desc.c
index 146f9e5..bb65c06 100644
--- a/lib/ext2fs/check_desc.c
+++ b/lib/ext2fs/check_desc.c
@@ -38,8 +38,14 @@ errcode_t ext2fs_check_desc(ext2_filsys fs)
EXT2_CHECK_MAGIC(fs, EXT2_ET_MAGIC_EXT2FS_FILSYS);
for (i = 0; i < fs->group_desc_count; i++) {
- first_block = ext2fs_group_first_block(fs, i);
- last_block = ext2fs_group_last_block(fs, i);
+ if (EXT2_HAS_INCOMPAT_FEATURE (fs->super, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
+ first_block = fs->super->s_first_data_block;
+ last_block = fs->super->s_blocks_count;
+ }
+ else {
+ first_block = ext2fs_group_first_block(fs, i);
+ last_block = ext2fs_group_last_block(fs, i);
+ }
/*
* Check to make sure block bitmap for group is
From: Jose R. Santos <[email protected]>
New bitmap and inode table allocation for FLEX_BG
Change the way we allocate bitmaps and inode tables if the FLEX_BG
feature is used at mke2fs time. The block and inode bitmaps are
allocated as a one contiguous set for each flex block group. Due to
the size of the inode tables, the inode table for each block group is
allocate individually but packed close together at the beginning of a
flex group. For now, this allow for the inode table to be packed
close to the inode bitmaps in cases where we try to allocate a large
group of inode tables right after the bitmaps and fail.
Signed-off-by: Jose R. Santos <[email protected]>
--
lib/ext2fs/alloc_tables.c | 138 ++++++++++++++++++++++++++++++++++++++++++++-
1 files changed, 134 insertions(+), 4 deletions(-)
diff --git a/lib/ext2fs/alloc_tables.c b/lib/ext2fs/alloc_tables.c
index 4ad2ba9..75252fa 100644
--- a/lib/ext2fs/alloc_tables.c
+++ b/lib/ext2fs/alloc_tables.c
@@ -27,6 +27,130 @@
#include "ext2_fs.h"
#include "ext2fs.h"
+#define ALLOC_BLOCK_BITMAPS 1
+#define ALLOC_INODE_BITMAPS 2
+#define ALLOC_INODE_TABLES 3
+
+#define GROUPS_PER_FLEXBG 64
+#define FIRST_METADATA_GROUP 3
+
+errcode_t ext2fs_allocate_contiguous(ext2_filsys fs, dgrp_t group,
+ int type, blk_t start_blk, blk_t last_blk,
+ int count, ext2fs_block_bitmap bmap)
+{
+ errcode_t retval;
+ blk_t new_blk, blk;
+ int i, j;
+
+ if (!bmap)
+ bmap = fs->block_map;
+
+ switch (type) {
+ case ALLOC_BLOCK_BITMAPS:
+ retval = ext2fs_get_free_blocks(fs, start_blk, last_blk,
+ 1 * count, bmap, &new_blk);
+ if (retval)
+ return retval;
+ for (i=0, blk=new_blk; i < count; i++, blk++) {
+ ext2fs_mark_block_bitmap(bmap, blk);
+ fs->group_desc[group+i].bg_block_bitmap = blk;
+ }
+ break;
+
+ case ALLOC_INODE_BITMAPS:
+ retval = ext2fs_get_free_blocks(fs, start_blk, last_blk,
+ 1 * count, bmap, &new_blk);
+ if (retval)
+ return retval;
+ for (i=0, blk=new_blk; i < count; i++, blk++) {
+ ext2fs_mark_block_bitmap(bmap, blk);
+ fs->group_desc[group+i].bg_inode_bitmap = blk;
+ }
+ break;
+
+ case ALLOC_INODE_TABLES:
+ for (i=0, blk=new_blk; i < count; i++, blk++) {
+ retval = ext2fs_get_free_blocks(fs, start_blk, last_blk,
+ fs->inode_blocks_per_group,
+ bmap, &new_blk);
+ if (retval)
+ return retval;
+ for (j=0, blk = new_blk;
+ j < fs->inode_blocks_per_group; j++, blk++)
+ ext2fs_mark_block_bitmap(bmap, blk);
+ fs->group_desc[group+i].bg_inode_table = new_blk;
+ }
+ break;
+
+ }
+ return 0;
+}
+
+
+
+errcode_t ext2fs_allocate_flex_groups(ext2_filsys fs)
+{
+ errcode_t retval;
+ blk_t start, last, j, blocks;
+ dgrp_t i, k;
+ int gpm;
+
+ gpm = GROUPS_PER_FLEXBG;
+ blocks = 0;
+
+ for (i = 0; i < fs->group_desc_count; i=i+gpm) {
+ if (i == 0 )
+ start = ext2fs_group_first_block(fs,
+ FIRST_METADATA_GROUP);
+ else
+ start = ext2fs_group_first_block(fs, i);
+
+ if (i+gpm-1 > fs->group_desc_count) {
+ last = ext2fs_group_last_block(fs, fs->group_desc_count);
+ gpm = fs->group_desc_count - i;
+ }
+ else
+ last = ext2fs_group_last_block(fs, i+gpm-1);
+
+ retval = ext2fs_allocate_contiguous(fs, i, ALLOC_BLOCK_BITMAPS,
+ start, last, gpm,
+ fs->block_map);
+ if (retval)
+ return retval;
+ retval = ext2fs_allocate_contiguous(fs, i, ALLOC_INODE_BITMAPS,
+ start, last, gpm,
+ fs->block_map);
+ if (retval)
+ return retval;
+ retval = ext2fs_allocate_contiguous(fs, i, ALLOC_INODE_TABLES,
+ start, last, gpm,
+ fs->block_map);
+ if (retval)
+ return retval;
+
+ /*
+ * The content of bg_free_blocks_count is previously
+ * assigned with out knowledge of the new allocation
+ * scheme. Need to update the number of free blocks
+ * per group descriptor or fsck will complain.
+ */
+
+ for (k=i; k<i+gpm; k++){
+ if (k > fs->group_desc_count)
+ break;
+ start = ext2fs_group_first_block(fs, k);
+ last = ext2fs_group_last_block(fs, k);
+ for (j=start; j<=last; j++) {
+ if( !ext2fs_fast_test_block_bitmap(fs->block_map, j))
+ blocks++;
+ }
+ fs->group_desc[k].bg_free_blocks_count = blocks;
+ blocks = 0;
+ }
+ }
+ return 0;
+}
+
errcode_t ext2fs_allocate_group_table(ext2_filsys fs, dgrp_t group,
ext2fs_block_bitmap bmap)
{
@@ -107,10 +231,16 @@ errcode_t ext2fs_allocate_tables(ext2_filsys fs)
errcode_t retval;
dgrp_t i;
- for (i = 0; i < fs->group_desc_count; i++) {
- retval = ext2fs_allocate_group_table(fs, i, fs->block_map);
- if (retval)
- return retval;
+ if (EXT2_HAS_INCOMPAT_FEATURE (fs->super,
+ EXT4_FEATURE_INCOMPAT_FLEX_BG))
+ ext2fs_allocate_flex_groups(fs);
+
+ else {
+ for (i = 0; i < fs->group_desc_count; i++) {
+ retval = ext2fs_allocate_group_table(fs, i, fs->block_map);
+ if (retval)
+ return retval;
+ }
}
return 0;
}
Jose R. Santos wrote:
> From: Jose R. Santos <[email protected]>
>
>
> e2fsck/super.c | 10 ++++++++--
> lib/ext2fs/check_desc.c | 10 ++++++++--
> 2 files changed, 16 insertions(+), 4 deletions(-)
>
> diff --git a/e2fsck/super.c b/e2fsck/super.c
> index 00a131c..8e58e5c 100644
> --- a/e2fsck/super.c
> +++ b/e2fsck/super.c
> @@ -578,8 +578,14 @@ void check_super_block(e2fsck_t ctx)
> for (i = 0, gd=fs->group_desc; i < fs->group_desc_count; i++, gd++) {
> pctx.group = i;
>
> - first_block = ext2fs_group_first_block(fs, i);
> - last_block = ext2fs_group_last_block(fs, i);
> + if (EXT2_HAS_INCOMPAT_FEATURE (fs->super,
> + EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
> + first_block = fs->super->s_first_data_block;
> + last_block = fs->super->s_blocks_count;
I guess this should be fs->super->s_blocks_count - 1 ;
> + } else {
> + first_block = ext2fs_group_first_block(fs, i);
> + last_block = ext2fs_group_last_block(fs, i);
> + }
>
> if ((gd->bg_block_bitmap < first_block) ||
> (gd->bg_block_bitmap > last_block)) {
> diff --git a/lib/ext2fs/check_desc.c b/lib/ext2fs/check_desc.c
> index 146f9e5..bb65c06 100644
> --- a/lib/ext2fs/check_desc.c
> +++ b/lib/ext2fs/check_desc.c
> @@ -38,8 +38,14 @@ errcode_t ext2fs_check_desc(ext2_filsys fs)
> EXT2_CHECK_MAGIC(fs, EXT2_ET_MAGIC_EXT2FS_FILSYS);
>
> for (i = 0; i < fs->group_desc_count; i++) {
> - first_block = ext2fs_group_first_block(fs, i);
> - last_block = ext2fs_group_last_block(fs, i);
> + if (EXT2_HAS_INCOMPAT_FEATURE (fs->super, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
> + first_block = fs->super->s_first_data_block;
> + last_block = fs->super->s_blocks_count;
> +
I guess this should be fs->super->s_blocks_count - 1 ;
-aneesh
Jose R. Santos wrote:
> From: Jose R. Santos <[email protected]>
>
> Ne
> + case ALLOC_INODE_TABLES:
> + for (i=0, blk=new_blk; i < count; i++, blk++) {
I guess you can drop the blk update in the for() loop above.
> + retval = ext2fs_get_free_blocks(fs, start_blk, last_blk,
> + fs->inode_blocks_per_group,
> + bmap, &new_blk);
> + if (retval)
> + return retval;
> + for (j=0, blk = new_blk;
> + j < fs->inode_blocks_per_group; j++, blk++)
> + ext2fs_mark_block_bitmap(bmap, blk);
> + fs->group_desc[group+i].bg_inode_table = new_blk;
> + }
> + break;
> +
> + }
> + return 0;
> +}
> +
> +
> +
> +errcode_t ext2fs_allocate_flex_groups(ext2_filsys fs)
> +{
> + errcode_t retval;
> + blk_t start, last, j, blocks;
> + dgrp_t i, k;
> + int gpm;
> +
> + gpm = GROUPS_PER_FLEXBG;
> + blocks = 0;
> +
> + for (i = 0; i < fs->group_desc_count; i=i+gpm) {
> + if (i == 0 )
> + start = ext2fs_group_first_block(fs,
> + FIRST_METADATA_GROUP);
> + else
> + start = ext2fs_group_first_block(fs, i);
> +
> + if (i+gpm-1 > fs->group_desc_count) {
if (i+gpm >= fs->group_desc_count)
> + last = ext2fs_group_last_block(fs, fs->group_desc_count);
> + gpm = fs->group_desc_count - i;
> + }
-aneesh
On Fri, 03 Aug 2007 10:54:04 +0530
"Aneesh Kumar K.V" <[email protected]> wrote:
> Jose R. Santos wrote:
> > From: Jose R. Santos <[email protected]>
> >
> >
> > e2fsck/super.c | 10 ++++++++--
> > lib/ext2fs/check_desc.c | 10 ++++++++--
> > 2 files changed, 16 insertions(+), 4 deletions(-)
> >
> > diff --git a/e2fsck/super.c b/e2fsck/super.c
> > index 00a131c..8e58e5c 100644
> > --- a/e2fsck/super.c
> > +++ b/e2fsck/super.c
> > @@ -578,8 +578,14 @@ void check_super_block(e2fsck_t ctx)
> > for (i = 0, gd=fs->group_desc; i < fs->group_desc_count; i++, gd++) {
> > pctx.group = i;
> >
> > - first_block = ext2fs_group_first_block(fs, i);
> > - last_block = ext2fs_group_last_block(fs, i);
> > + if (EXT2_HAS_INCOMPAT_FEATURE (fs->super,
> > + EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
> > + first_block = fs->super->s_first_data_block;
> > + last_block = fs->super->s_blocks_count;
>
>
> I guess this should be fs->super->s_blocks_count - 1 ;
Updated.
> > + } else {
> > + first_block = ext2fs_group_first_block(fs, i);
> > + last_block = ext2fs_group_last_block(fs, i);
> > + }
> >
> > if ((gd->bg_block_bitmap < first_block) ||
> > (gd->bg_block_bitmap > last_block)) {
> > diff --git a/lib/ext2fs/check_desc.c b/lib/ext2fs/check_desc.c
> > index 146f9e5..bb65c06 100644
> > --- a/lib/ext2fs/check_desc.c
> > +++ b/lib/ext2fs/check_desc.c
> > @@ -38,8 +38,14 @@ errcode_t ext2fs_check_desc(ext2_filsys fs)
> > EXT2_CHECK_MAGIC(fs, EXT2_ET_MAGIC_EXT2FS_FILSYS);
> >
> > for (i = 0; i < fs->group_desc_count; i++) {
> > - first_block = ext2fs_group_first_block(fs, i);
> > - last_block = ext2fs_group_last_block(fs, i);
> > + if (EXT2_HAS_INCOMPAT_FEATURE (fs->super, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
> > + first_block = fs->super->s_first_data_block;
> > + last_block = fs->super->s_blocks_count;
> > +
>
> I guess this should be fs->super->s_blocks_count - 1 ;
Updated.
Thanks
> -aneesh
-JRS
On Fri, 03 Aug 2007 12:01:20 +0530
"Aneesh Kumar K.V" <[email protected]> wrote:
>
>
> Jose R. Santos wrote:
> > From: Jose R. Santos <[email protected]>
> >
> > Ne
> > + case ALLOC_INODE_TABLES:
> > + for (i=0, blk=new_blk; i < count; i++, blk++) {
>
> I guess you can drop the blk update in the for() loop above.
This was from the remains of the very first attempt at inode table
allocation. Thanks for catching.
>
> > + retval = ext2fs_get_free_blocks(fs, start_blk, last_blk,
> > + fs->inode_blocks_per_group,
> > + bmap, &new_blk);
> > + if (retval)
> > + return retval;
> > + for (j=0, blk = new_blk;
> > + j < fs->inode_blocks_per_group; j++, blk++)
> > + ext2fs_mark_block_bitmap(bmap, blk);
> > + fs->group_desc[group+i].bg_inode_table = new_blk;
> > + }
> > + break;
> > +
> > + }
> > + return 0;
> > +}
> > +
> > +
> > +
> > +errcode_t ext2fs_allocate_flex_groups(ext2_filsys fs)
> > +{
> > + errcode_t retval;
> > + blk_t start, last, j, blocks;
> > + dgrp_t i, k;
> > + int gpm;
> > +
> > + gpm = GROUPS_PER_FLEXBG;
> > + blocks = 0;
> > +
> > + for (i = 0; i < fs->group_desc_count; i=i+gpm) {
> > + if (i == 0 )
> > + start = ext2fs_group_first_block(fs,
> > + FIRST_METADATA_GROUP);
> > + else
> > + start = ext2fs_group_first_block(fs, i);
> > +
> > + if (i+gpm-1 > fs->group_desc_count) {
>
>
> if (i+gpm >= fs->group_desc_count)
Update.
>
> > + last = ext2fs_group_last_block(fs, fs->group_desc_count);
> > + gpm = fs->group_desc_count - i;
> > + }
>
>
> -aneesh
Thanks
-JRS
On Aug 02, 2007 23:00 -0500, Jose R. Santos wrote:
> Eventually, a more thorough check would restrict bitmaps and inode
> tables to be located at the beginning of a flex block group range.
> Since the super block does not currently know about the number of
> groups per flex group, this will do for now.
As with regular block groups, it would probably be better to limit the
bitmaps and inode tables to anywhere inside the flexbg instead of the
start. That would allow, for example, INCOMPAT_FLEXBG to be enabled
on an existing filesystem and the metadata could be moved together as
space becomes available.
Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.
On Fri, 3 Aug 2007 12:22:17 -0600
Andreas Dilger <[email protected]> wrote:
> On Aug 02, 2007 23:00 -0500, Jose R. Santos wrote:
> > Eventually, a more thorough check would restrict bitmaps and inode
> > tables to be located at the beginning of a flex block group range.
> > Since the super block does not currently know about the number of
> > groups per flex group, this will do for now.
>
> As with regular block groups, it would probably be better to limit the
> bitmaps and inode tables to anywhere inside the flexbg instead of the
> start. That would allow, for example, INCOMPAT_FLEXBG to be enabled
> on an existing filesystem and the metadata could be moved together as
> space becomes available.
This is something that would be simple to do if the ratio of block
groups per flex group known to the filesystem itself. This implies
adding another field to the super block as reliable way to obtain this
information. The only thing keeping me from doing so is the uncertainty
of backwards compatibility when changing the super block structure.
I agree though that one of the requirements for this feature is more
robust checking of the location of the bitmaps and inode tables within
the flex group. Checking of the descriptor in flexbg become a little
more complicated than in regular block groups because:
1. The block and inode bitmaps should be allocated a one big chunk for
each flex group.
2. The block and inode bitmaps should be located in the first block
group and the inode tables with in the first few groups of a flex group.
3. If the full range of bitmaps in not allocated contiguously, this
means that bad blocks caused us to move a particular bitmap out and
thus the bad block list should be checked to ensure that this was the
case.
If the above conditions are not met, this could point to possible
corruption in the block descriptors.
> Cheers, Andreas
> --
> Andreas Dilger
> Principal Software Engineer
> Cluster File Systems, Inc.
>
-JRS