[PATCH] delalloc: Add block reservation estimate for non-extent files

[PATCH] delalloc: Add block reservation estimate for non-extent files

[Mingming]

Add support for full delayed allocation over ext3 format file.

This patch added a function to estimate the number of indirect blocks need
to reserve for non-extent based (ext3) file, in order to properly
reserve the need amount of indirect blocks for delayed allocation.


Signed-off-by: Mingming Cao <cmm@us.ibm.com>
---
 fs/ext4/inode.c |   34 ++++++++++++++++++++++++++++++++--
 1 file changed, 32 insertions(+), 2 deletions(-)

Index: linux-2.6.26-rc6/fs/ext4/inode.c
===================================================================
--- linux-2.6.26-rc6.orig/fs/ext4/inode.c	2008-06-20 16:50:09.000000000 -0700
+++ linux-2.6.26-rc6/fs/ext4/inode.c	2008-06-20 17:36:08.000000000 -0700
@@ -1426,6 +1426,36 @@ static int ext4_journalled_write_end(str
 
 	return ret ? ret : copied;
 }
+/*
+ * Calculate the number of metadata blocks need to reserve
+ * to allocate @blocks for non extent file based file
+ */
+static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
+{
+	int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
+	int ind_blks, dind_blks, tind_blks;
+
+	/* number of new indirect blocks needed */
+	ind_blks = (blocks + icap - 1) / icap;
+
+	dind_blks = (ind_blks + icap - 1) / icap;
+
+	tind_blks = 1;
+
+	return ind_blks + dind_blks + tind_blks;
+}
+
+/*
+ * Calculate the number of metadata blocks need to reserve
+ * to allocate given number of blocks
+ */
+static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
+{
+	if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
+		return ext4_ext_calc_metadata_amount(inode, blocks);
+
+	return ext4_indirect_calc_metadata_amount(inode, blocks);
+}
 
 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
 {
@@ -1439,7 +1469,7 @@ static int ext4_da_reserve_space(struct 
 	 */
 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 	total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
-	mdblocks = ext4_ext_calc_metadata_amount(inode, total);
+	mdblocks = ext4_calc_metadata_amount(inode, total);
 	BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
 
 	md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
@@ -1468,7 +1498,7 @@ void ext4_da_release_space(struct inode 
 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 	/* recalculate the number of metablocks still need to be reserved */
 	total = EXT4_I(inode)->i_reserved_data_blocks - used - to_free;
-	mdb = ext4_ext_calc_metadata_amount(inode, total);
+	mdb = ext4_calc_metadata_amount(inode, total);
 
 	/* figure out how many metablocks to release */
 	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);

Re: [PATCH] delalloc: Add block reservation estimate for non-extent files

[Aneesh Kumar K.V]

On Fri, Jun 20, 2008 at 06:11:31PM -0700, Mingming wrote:
> Add support for full delayed allocation over ext3 format file.
> 
> This patch added a function to estimate the number of indirect blocks need
> to reserve for non-extent based (ext3) file, in order to properly
> reserve the need amount of indirect blocks for delayed allocation.
> 
> 
> Signed-off-by: Mingming Cao <cmm@us.ibm.com>
> ---
>  fs/ext4/inode.c |   34 ++++++++++++++++++++++++++++++++--
>  1 file changed, 32 insertions(+), 2 deletions(-)
> 
> Index: linux-2.6.26-rc6/fs/ext4/inode.c
> ===================================================================
> --- linux-2.6.26-rc6.orig/fs/ext4/inode.c	2008-06-20 16:50:09.000000000 -0700
> +++ linux-2.6.26-rc6/fs/ext4/inode.c	2008-06-20 17:36:08.000000000 -0700
> @@ -1426,6 +1426,36 @@ static int ext4_journalled_write_end(str
> 
>  	return ret ? ret : copied;
>  }
> +/*
> + * Calculate the number of metadata blocks need to reserve
> + * to allocate @blocks for non extent file based file
> + */
> +static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
> +{
> +	int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
> +	int ind_blks, dind_blks, tind_blks;
> +
> +	/* number of new indirect blocks needed */
> +	ind_blks = (blocks + icap - 1) / icap;
> +
> +	dind_blks = (ind_blks + icap - 1) / icap;
> +
> +	tind_blks = 1;
> +
> +	return ind_blks + dind_blks + tind_blks;
> +}
> +

With block mapped files using logical block number we should be able to
find out the exact number of meta-data blocks we would need. I am not sure
whether we need to be that complex is calculating the same.

-aneesh

Re: [PATCH] delalloc: Add block reservation estimate for non-extent files

[Mingming]

On Sat, 2008-06-21 at 15:55 +0530, Aneesh Kumar K.V wrote:
> On Fri, Jun 20, 2008 at 06:11:31PM -0700, Mingming wrote:
> > Add support for full delayed allocation over ext3 format file.
> > 
> > This patch added a function to estimate the number of indirect blocks need
> > to reserve for non-extent based (ext3) file, in order to properly
> > reserve the need amount of indirect blocks for delayed allocation.
> > 
> > 
> > Signed-off-by: Mingming Cao <cmm@us.ibm.com>
> > ---
> >  fs/ext4/inode.c |   34 ++++++++++++++++++++++++++++++++--
> >  1 file changed, 32 insertions(+), 2 deletions(-)
> > 
> > Index: linux-2.6.26-rc6/fs/ext4/inode.c
> > ===================================================================
> > --- linux-2.6.26-rc6.orig/fs/ext4/inode.c	2008-06-20 16:50:09.000000000 -0700
> > +++ linux-2.6.26-rc6/fs/ext4/inode.c	2008-06-20 17:36:08.000000000 -0700
> > @@ -1426,6 +1426,36 @@ static int ext4_journalled_write_end(str
> > 
> >  	return ret ? ret : copied;
> >  }
> > +/*
> > + * Calculate the number of metadata blocks need to reserve
> > + * to allocate @blocks for non extent file based file
> > + */
> > +static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
> > +{
> > +	int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
> > +	int ind_blks, dind_blks, tind_blks;
> > +
> > +	/* number of new indirect blocks needed */
> > +	ind_blks = (blocks + icap - 1) / icap;
> > +
> > +	dind_blks = (ind_blks + icap - 1) / icap;
> > +
> > +	tind_blks = 1;
> > +
> > +	return ind_blks + dind_blks + tind_blks;
> > +}
> > +
> 
> With block mapped files using logical block number we should be able to
> find out the exact number of meta-data blocks we would need. I am not sure
> whether we need to be that complex is calculating the same.
> 

Yeah the complexity is the part I was trying to avoid, plus we can't get
the exact number unless lock the indirect block mapping tree to do the
look up, that's seems too expensive to me as this is called for every
write..

Mingming