[RFC][PATCH 3/6] delalloc ENOSPC: reserve blocks

[RFC][PATCH 3/6] delalloc ENOSPC: reserve blocks

[Mingming Cao]

ext4: delalloc ENOSPC handling core

From: Mingming cao <cmm@us.ibm.com>

Core part of delaloc ENOSPC (block reservation.) 
data/meta blocks are reserved on write_begin(), and per-inode reserved counters
are updated  after block allocation.
 
Signed-off-by: Mingming cao <cmm@us.ibm.com>

---
 fs/ext4/inode.c |   69 +++++++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 66 insertions(+), 3 deletions(-)

Index: linux-2.6.26-rc4/fs/ext4/inode.c
===================================================================
--- linux-2.6.26-rc4.orig/fs/ext4/inode.c	2008-06-01 14:26:13.000000000 -0700
+++ linux-2.6.26-rc4/fs/ext4/inode.c	2008-06-01 15:04:06.000000000 -0700
@@ -38,6 +38,7 @@
 #include "ext4_jbd2.h"
 #include "xattr.h"
 #include "acl.h"
+#include "ext4_extents.h"
 
 static void ext4_invalidatepage(struct page *page, unsigned long offset);
 
@@ -1410,6 +1411,61 @@ static int ext4_journalled_write_end(str
 	return ret ? ret : copied;
 }
 
+static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
+{
+       struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+       unsigned long md_needed, mdblocks, total = 0;
+
+       /*
+        * calculate the amount of metadata blocks to reserve
+	* in order to allocate nrblocks
+	* worse case is one extent per block
+	*/
+	total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
+	mdblocks = ext4_ext_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;
+	total = md_needed + nrblocks;
+
+	if (ext4_has_free_blocks(sbi, total) < total)
+		return -ENOSPC;
+
+	/* reduce fs free blocks counter */
+	percpu_counter_sub(&sbi->s_freeblocks_counter, total);
+
+	EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
+	EXT4_I(inode)->i_reserved_meta_blocks += md_needed;
+
+	return 0;       /* success */
+}
+
+static void ext4_da_release_space(struct inode *inode, int used, int to_free)
+{
+	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+	int total, mdb, release;
+
+	/* calculate 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);
+
+	/* figure out how many metablocks to release */
+	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
+	mdb = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
+
+	release = to_free + mdb;
+
+	/* update fs free blocks counter for truncate case */
+	percpu_counter_add(&sbi->s_freeblocks_counter, release);
+
+	/* update per-inode reservations */
+	BUG_ON(used + to_free > EXT4_I(inode)->i_reserved_data_blocks);
+	EXT4_I(inode)->i_reserved_data_blocks -= used + to_free;
+
+	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
+	EXT4_I(inode)->i_reserved_meta_blocks -= mdb;
+}
+
 /*
  * this is a special callback for ->write_begin() only
  * it's intention is to return mapped block or reserve space
@@ -1428,13 +1484,17 @@ static int ext4_da_get_block_prep(struct
 	 * the same as allocated blocks.
 	 */
 	ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1,  bh_result, 0, 0);
-	if (ret == 0) {
-		/* the block isn't allocated yet, let's reserve space */
-		/* XXX: call reservation here */
+	if ((ret == 0)&& !buffer_delay(bh_result)) {
+		/* the block isn't (pre)allocated yet, let's reserve space */
 		/*
 		 * XXX: __block_prepare_write() unmaps passed block,
 		 * is it OK?
 		 */
+		ret = ext4_da_reserve_space(inode, 1);
+		if (ret)
+			/* not enough space to reserve */
+			return ret;
+
 		map_bh(bh_result, inode->i_sb, 0);
 		set_buffer_new(bh_result);
 		set_buffer_delay(bh_result);
@@ -1463,6 +1523,9 @@ static int ext4_da_get_block_write(struc
 	if (ret > 0) {
 		bh_result->b_size = (ret << inode->i_blkbits);
 
+		/* release reserved-but-unused meta blocks */
+		ext4_da_release_space(inode, ret, 0);
+
 		/*
 		 * Update on-disk size along with block allocation
 		 * we don't use 'extend_disksize' as size may change

Re: [RFC][PATCH 3/6] delalloc ENOSPC: reserve blocks

[Aneesh Kumar K.V]

On Sun, Jun 01, 2008 at 04:35:55PM -0700, Mingming Cao wrote:
> ext4: delalloc ENOSPC handling core
> 
> From: Mingming cao <cmm@us.ibm.com>
> 
> Core part of delaloc ENOSPC (block reservation.) 
> data/meta blocks are reserved on write_begin(), and per-inode reserved counters
> are updated  after block allocation.
> 
> Signed-off-by: Mingming cao <cmm@us.ibm.com>
> 
> ---
>  fs/ext4/inode.c |   69 +++++++++++++++++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 66 insertions(+), 3 deletions(-)
> 
> Index: linux-2.6.26-rc4/fs/ext4/inode.c
> ===================================================================
> --- linux-2.6.26-rc4.orig/fs/ext4/inode.c	2008-06-01 14:26:13.000000000 -0700
> +++ linux-2.6.26-rc4/fs/ext4/inode.c	2008-06-01 15:04:06.000000000 -0700
> @@ -38,6 +38,7 @@
>  #include "ext4_jbd2.h"
>  #include "xattr.h"
>  #include "acl.h"
> +#include "ext4_extents.h"
> 
>  static void ext4_invalidatepage(struct page *page, unsigned long offset);
> 
> @@ -1410,6 +1411,61 @@ static int ext4_journalled_write_end(str
>  	return ret ? ret : copied;
>  }
> 
> +static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
> +{
> +       struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
> +       unsigned long md_needed, mdblocks, total = 0;
> +
> +       /*
> +        * calculate the amount of metadata blocks to reserve

	 recalculate the amount of metadata blocks to reserve.




> +	* in order to allocate nrblocks
> +	* worse case is one extent per block
> +	*/
> +	total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
> +	mdblocks = ext4_ext_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;
> +	total = md_needed + nrblocks;
> +
> +	if (ext4_has_free_blocks(sbi, total) < total)
> +		return -ENOSPC;
> +
> +	/* reduce fs free blocks counter */
> +	percpu_counter_sub(&sbi->s_freeblocks_counter, total);
> +
> +	EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
> +	EXT4_I(inode)->i_reserved_meta_blocks += md_needed;

Better written as
	EXT4_I(inode)->i_reserved_meta_blocks += mdblocks;

> +
> +	return 0;       /* success */
> +}
> +
> +static void ext4_da_release_space(struct inode *inode, int used, int to_free)
> +{
> +	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
> +	int total, mdb, release;

	int total, mdb, release, mdb_to_free;

> +
> +	/* calculate 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);
> +
> +	/* figure out how many metablocks to release */
> +	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
> +	mdb = EXT4_I(inode)->i_reserved_meta_blocks - mdb;

	mdb_to_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;


> +
> +	release = to_free + mdb;
> +
	release = to_free + mdb_to_free;


> +	/* update fs free blocks counter for truncate case */
> +	percpu_counter_add(&sbi->s_freeblocks_counter, release);
> +
> +	/* update per-inode reservations */
> +	BUG_ON(used + to_free > EXT4_I(inode)->i_reserved_data_blocks);
> +	EXT4_I(inode)->i_reserved_data_blocks -= used + to_free;
> +
> +	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
> +	EXT4_I(inode)->i_reserved_meta_blocks -= mdb;

	/*
	 * update the meta block reservation 
	 */
	EXT4_I(inode)->i_reserved_meta_blocks = mdb;

> +}
> +
>  /*
>   * this is a special callback for ->write_begin() only
>   * it's intention is to return mapped block or reserve space
> @@ -1428,13 +1484,17 @@ static int ext4_da_get_block_prep(struct
>  	 * the same as allocated blocks.
>  	 */
>  	ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1,  bh_result, 0, 0);
> -	if (ret == 0) {
> -		/* the block isn't allocated yet, let's reserve space */
> -		/* XXX: call reservation here */
> +	if ((ret == 0)&& !buffer_delay(bh_result)) {
> +		/* the block isn't (pre)allocated yet, let's reserve space */
>  		/*
>  		 * XXX: __block_prepare_write() unmaps passed block,
>  		 * is it OK?
>  		 */
> +		ret = ext4_da_reserve_space(inode, 1);
> +		if (ret)
> +			/* not enough space to reserve */
> +			return ret;
> +
>  		map_bh(bh_result, inode->i_sb, 0);
>  		set_buffer_new(bh_result);
>  		set_buffer_delay(bh_result);
> @@ -1463,6 +1523,9 @@ static int ext4_da_get_block_write(struc
>  	if (ret > 0) {
>  		bh_result->b_size = (ret << inode->i_blkbits);
> 
> +		/* release reserved-but-unused meta blocks */
> +		ext4_da_release_space(inode, ret, 0);
> +
>  		/*
>  		 * Update on-disk size along with block allocation
>  		 * we don't use 'extend_disksize' as size may change
> 
> 


-aneesh

Re: [RFC][PATCH 3/6] delalloc ENOSPC: reserve blocks

[Aneesh Kumar K.V]

On Mon, Jun 02, 2008 at 12:14:33PM +0530, Aneesh Kumar K.V wrote:
> On Sun, Jun 01, 2008 at 04:35:55PM -0700, Mingming Cao wrote:
> > ext4: delalloc ENOSPC handling core
> > 
> > From: Mingming cao <cmm@us.ibm.com>
> > 
> > Core part of delaloc ENOSPC (block reservation.) 
> > data/meta blocks are reserved on write_begin(), and per-inode reserved counters
> > are updated  after block allocation.
> > 
> > Signed-off-by: Mingming cao <cmm@us.ibm.com>
> > 
> > ---
> >  fs/ext4/inode.c |   69 +++++++++++++++++++++++++++++++++++++++++++++++++++++---
> >  1 file changed, 66 insertions(+), 3 deletions(-)
> > 
> > Index: linux-2.6.26-rc4/fs/ext4/inode.c
> > ===================================================================
> > --- linux-2.6.26-rc4.orig/fs/ext4/inode.c	2008-06-01 14:26:13.000000000 -0700
> > +++ linux-2.6.26-rc4/fs/ext4/inode.c	2008-06-01 15:04:06.000000000 -0700
> > @@ -38,6 +38,7 @@
> >  #include "ext4_jbd2.h"
> >  #include "xattr.h"
> >  #include "acl.h"
> > +#include "ext4_extents.h"
> > 
> >  static void ext4_invalidatepage(struct page *page, unsigned long offset);
> > 
> > @@ -1410,6 +1411,61 @@ static int ext4_journalled_write_end(str
> >  	return ret ? ret : copied;
> >  }
> > 
> > +static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
> > +{
> > +       struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
> > +       unsigned long md_needed, mdblocks, total = 0;
> > +
> > +       /*
> > +        * calculate the amount of metadata blocks to reserve
> 
> 	 recalculate the amount of metadata blocks to reserve.
> 
> 
> 
> 
> > +	* in order to allocate nrblocks
> > +	* worse case is one extent per block
> > +	*/
> > +	total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
> > +	mdblocks = ext4_ext_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;
> > +	total = md_needed + nrblocks;
> > +
> > +	if (ext4_has_free_blocks(sbi, total) < total)
> > +		return -ENOSPC;
> > +
> > +	/* reduce fs free blocks counter */
> > +	percpu_counter_sub(&sbi->s_freeblocks_counter, total);
> > +
> > +	EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
> > +	EXT4_I(inode)->i_reserved_meta_blocks += md_needed;
> 
> Better written as
> 	EXT4_I(inode)->i_reserved_meta_blocks += mdblocks;
> 
> 
	EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;

-aneesh

Re: [RFC][PATCH 3/6] delalloc ENOSPC: reserve blocks

[Aneesh Kumar K.V]

On Sun, Jun 01, 2008 at 04:35:55PM -0700, Mingming Cao wrote:
> ext4: delalloc ENOSPC handling core
> 
> From: Mingming cao <cmm@us.ibm.com>
> 
> Core part of delaloc ENOSPC (block reservation.) 
> data/meta blocks are reserved on write_begin(), and per-inode reserved counters
> are updated  after block allocation.


I am not clear about ext4_da_release_space. Are we accounting for the
blocks related to meta-data blocks allocated ? . In
ext4_da_get_block_write we get the number of data blocks allocated.
and we don't update sbi->s_freeblocks_counter in the block allocator
any more. So request for meta-data blocks are not updating the
sbi->s_freeblocks_counter. In ext4_da_release_space we  update
the  sbi->s_freeblocks_counter without considering the above. Am  I
missing something ?


-aneesh