[Ocfs2-devel] [PATCH 1/1] ocfs2: add extent block stealing for ocfs2

[Tiger Yang <tiger.yang@oracle.com>]

Sorry. I forgot set extent block stolen number to zero after stolen 1024 
extent blocks.

the fixed patch is enclosed.

Best regards,
tiger

Tiger Yang wrote:
> This patch add extent(metadata) block stealing mechanism for
> extent allocation. This mechanism is same as the inode stealing.
> ocfs2_steal_from_other_nodes was introducted to steal the
> resources(inode or extent block) from other slot.
>
> Signed-off-by: Tiger Yang <tiger.yang@oracle.com>
> ---
>  fs/ocfs2/alloc.c        |    5 +-
>  fs/ocfs2/dir.c          |    2 +-
>  fs/ocfs2/localalloc.c   |    2 +-
>  fs/ocfs2/ocfs2.h        |   38 ++++++++++++++---
>  fs/ocfs2/refcounttree.c |    6 +-
>  fs/ocfs2/suballoc.c     |  104 +++++++++++++++++++++++++++++-----------------
>  fs/ocfs2/super.c        |    9 +++-
>  fs/ocfs2/xattr.c        |    2 +-
>  8 files changed, 111 insertions(+), 57 deletions(-)
>
> diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
> index d17bdc7..cae72f0 100644
> --- a/fs/ocfs2/alloc.c
> +++ b/fs/ocfs2/alloc.c
> @@ -1050,7 +1050,8 @@ static int ocfs2_create_new_meta_bhs(handle_t *handle,
>  			strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
>  			eb->h_blkno = cpu_to_le64(first_blkno);
>  			eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
> -			eb->h_suballoc_slot = cpu_to_le16(osb->slot_num);
> +			eb->h_suballoc_slot =
> +				cpu_to_le16(meta_ac->ac_alloc_slot);
>  			eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
>  			eb->h_list.l_count =
>  				cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
> @@ -6037,7 +6038,7 @@ static void ocfs2_truncate_log_worker(struct work_struct *work)
>  	if (status < 0)
>  		mlog_errno(status);
>  	else
> -		ocfs2_init_inode_steal_slot(osb);
> +		ocfs2_init_all_steal_slot(osb);
>  
>  	mlog_exit(status);
>  }
> diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
> index 28c3ec2..765d66c 100644
> --- a/fs/ocfs2/dir.c
> +++ b/fs/ocfs2/dir.c
> @@ -2439,7 +2439,7 @@ static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
>  	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
>  	memset(dx_root, 0, osb->sb->s_blocksize);
>  	strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
> -	dx_root->dr_suballoc_slot = cpu_to_le16(osb->slot_num);
> +	dx_root->dr_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
>  	dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
>  	dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
>  	dx_root->dr_blkno = cpu_to_le64(dr_blkno);
> diff --git a/fs/ocfs2/localalloc.c b/fs/ocfs2/localalloc.c
> index ac10f83..70480ce 100644
> --- a/fs/ocfs2/localalloc.c
> +++ b/fs/ocfs2/localalloc.c
> @@ -476,7 +476,7 @@ out_mutex:
>  
>  out:
>  	if (!status)
> -		ocfs2_init_inode_steal_slot(osb);
> +		ocfs2_init_all_steal_slot(osb);
>  	mlog_exit(status);
>  	return status;
>  }
> diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
> index 9362eea..989570c 100644
> --- a/fs/ocfs2/ocfs2.h
> +++ b/fs/ocfs2/ocfs2.h
> @@ -301,7 +301,9 @@ struct ocfs2_super
>  	u32 s_next_generation;
>  	unsigned long osb_flags;
>  	s16 s_inode_steal_slot;
> +	s16 s_extent_steal_slot;
>  	atomic_t s_num_inodes_stolen;
> +	atomic_t s_num_extents_stolen;
>  
>  	unsigned long s_mount_opt;
>  	unsigned int s_atime_quantum;
> @@ -756,28 +758,50 @@ static inline unsigned int ocfs2_megabytes_to_clusters(struct super_block *sb,
>  	return megs << (20 - OCFS2_SB(sb)->s_clustersize_bits);
>  }
>  
> -static inline void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
> +static inline void ocfs2_init_all_steal_slot(struct ocfs2_super *osb)
>  {
>  	spin_lock(&osb->osb_lock);
>  	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
> +	osb->s_extent_steal_slot = OCFS2_INVALID_SLOT;
>  	spin_unlock(&osb->osb_lock);
>  	atomic_set(&osb->s_num_inodes_stolen, 0);
> +	atomic_set(&osb->s_num_extents_stolen, 0);
>  }
>  
> -static inline void ocfs2_set_inode_steal_slot(struct ocfs2_super *osb,
> -					      s16 slot)
> +static inline void ocfs2_init_steal_slot(struct ocfs2_super *osb, int type)
>  {
>  	spin_lock(&osb->osb_lock);
> -	osb->s_inode_steal_slot = slot;
> +	if (type == INODE_ALLOC_SYSTEM_INODE)
> +		osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
> +	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
> +		osb->s_extent_steal_slot = OCFS2_INVALID_SLOT;
>  	spin_unlock(&osb->osb_lock);
> +	if (type == INODE_ALLOC_SYSTEM_INODE)
> +		atomic_set(&osb->s_num_inodes_stolen, 0);
> +	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
> +		atomic_set(&osb->s_num_extents_stolen, 0);
>  }
>  
> -static inline s16 ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
> +static inline void ocfs2_set_steal_slot(struct ocfs2_super *osb,
> +					int slot, int type)
>  {
> -	s16 slot;
> +	spin_lock(&osb->osb_lock);
> +	if (type == INODE_ALLOC_SYSTEM_INODE)
> +		osb->s_inode_steal_slot = slot;
> +	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
> +		osb->s_extent_steal_slot = slot;
> +	spin_unlock(&osb->osb_lock);
> +}
> +
> +static inline int ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
> +{
> +	int slot = OCFS2_INVALID_SLOT;
>  
>  	spin_lock(&osb->osb_lock);
> -	slot = osb->s_inode_steal_slot;
> +	if (type == INODE_ALLOC_SYSTEM_INODE)
> +		slot = osb->s_inode_steal_slot;
> +	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
> +		slot = osb->s_extent_steal_slot;
>  	spin_unlock(&osb->osb_lock);
>  
>  	return slot;
> diff --git a/fs/ocfs2/refcounttree.c b/fs/ocfs2/refcounttree.c
> index 74db2be..56abab3 100644
> --- a/fs/ocfs2/refcounttree.c
> +++ b/fs/ocfs2/refcounttree.c
> @@ -626,7 +626,7 @@ static int ocfs2_create_refcount_tree(struct inode *inode,
>  	rb = (struct ocfs2_refcount_block *)new_bh->b_data;
>  	memset(rb, 0, inode->i_sb->s_blocksize);
>  	strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
> -	rb->rf_suballoc_slot = cpu_to_le16(osb->slot_num);
> +	rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
>  	rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
>  	rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
>  	rb->rf_blkno = cpu_to_le64(first_blkno);
> @@ -1330,7 +1330,7 @@ static int ocfs2_expand_inline_ref_root(handle_t *handle,
>  	memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
>  
>  	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
> -	new_rb->rf_suballoc_slot = cpu_to_le16(OCFS2_SB(sb)->slot_num);
> +	new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
>  	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
>  	new_rb->rf_blkno = cpu_to_le64(blkno);
>  	new_rb->rf_cpos = cpu_to_le32(0);
> @@ -1576,7 +1576,7 @@ static int ocfs2_new_leaf_refcount_block(handle_t *handle,
>  	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
>  	memset(new_rb, 0, sb->s_blocksize);
>  	strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
> -	new_rb->rf_suballoc_slot = cpu_to_le16(OCFS2_SB(sb)->slot_num);
> +	new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
>  	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
>  	new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
>  	new_rb->rf_blkno = cpu_to_le64(blkno);
> diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
> index c30b644..2b39ee8 100644
> --- a/fs/ocfs2/suballoc.c
> +++ b/fs/ocfs2/suballoc.c
> @@ -52,6 +52,7 @@
>  #define ALLOC_GROUPS_FROM_GLOBAL	0x2
>  
>  #define OCFS2_MAX_INODES_TO_STEAL	1024
> +#define OCFS2_MAX_EXTENTS_TO_STEAL	1024
>  
>  static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
>  static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
> @@ -637,12 +638,45 @@ bail:
>  	return status;
>  }
>  
> +static int ocfs2_steal_from_other_nodes(struct ocfs2_super *osb,
> +					struct ocfs2_alloc_context *ac,
> +					int type)
> +{
> +	int i, status = -ENOSPC;
> +	int slot = ocfs2_get_steal_slot(osb, type);
> +
> +	/* Start to steal resource from the first slot after ours. */
> +	if (slot == OCFS2_INVALID_SLOT)
> +		slot = osb->slot_num + 1;
> +
> +	for (i = 0; i < osb->max_slots; i++, slot++) {
> +		if (slot == osb->max_slots)
> +			slot = 0;
> +
> +		if (slot == osb->slot_num)
> +			continue;
> +
> +		status = ocfs2_reserve_suballoc_bits(osb, ac,
> +						     type,
> +						     (u32)slot, NULL,
> +						     NOT_ALLOC_NEW_GROUP);
> +		if (status >= 0) {
> +			ocfs2_set_steal_slot(osb, slot, type);
> +			break;
> +		}
> +
> +		ocfs2_free_ac_resource(ac);
> +	}
> +
> +	return status;
> +}
> +
>  int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
>  				      int blocks,
>  				      struct ocfs2_alloc_context **ac)
>  {
>  	int status;
> -	u32 slot;
> +	int slot = ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
>  
>  	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
>  	if (!(*ac)) {
> @@ -653,12 +687,35 @@ int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
>  
>  	(*ac)->ac_bits_wanted = blocks;
>  	(*ac)->ac_which = OCFS2_AC_USE_META;
> -	slot = osb->slot_num;
>  	(*ac)->ac_group_search = ocfs2_block_group_search;
>  
> +	if (slot != OCFS2_INVALID_SLOT &&
> +		atomic_read(&osb->s_num_extents_stolen) <
> +		OCFS2_MAX_EXTENTS_TO_STEAL)
> +		goto extent_steal;
> +
>  	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
>  					     EXTENT_ALLOC_SYSTEM_INODE,
> -					     slot, NULL, ALLOC_NEW_GROUP);
> +					     (u32)osb->slot_num, NULL,
> +					     ALLOC_NEW_GROUP);
> +
> +
> +	if (status >= 0) {
> +		status = 0;
> +		if (slot != OCFS2_INVALID_SLOT)
> +			ocfs2_init_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
> +		goto bail;
> +	} else if (status < 0 && status != -ENOSPC) {
> +		mlog_errno(status);
> +		goto bail;
> +	}
> +
> +	ocfs2_free_ac_resource(*ac);
> +
> +extent_steal:
> +	status = ocfs2_steal_from_other_nodes(osb, *ac,
> +					      EXTENT_ALLOC_SYSTEM_INODE);
> +	atomic_inc(&osb->s_num_extents_stolen);
>  	if (status < 0) {
>  		if (status != -ENOSPC)
>  			mlog_errno(status);
> @@ -685,43 +742,11 @@ int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
>  					ac);
>  }
>  
> -static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
> -					      struct ocfs2_alloc_context *ac)
> -{
> -	int i, status = -ENOSPC;
> -	s16 slot = ocfs2_get_inode_steal_slot(osb);
> -
> -	/* Start to steal inodes from the first slot after ours. */
> -	if (slot == OCFS2_INVALID_SLOT)
> -		slot = osb->slot_num + 1;
> -
> -	for (i = 0; i < osb->max_slots; i++, slot++) {
> -		if (slot == osb->max_slots)
> -			slot = 0;
> -
> -		if (slot == osb->slot_num)
> -			continue;
> -
> -		status = ocfs2_reserve_suballoc_bits(osb, ac,
> -						     INODE_ALLOC_SYSTEM_INODE,
> -						     slot, NULL,
> -						     NOT_ALLOC_NEW_GROUP);
> -		if (status >= 0) {
> -			ocfs2_set_inode_steal_slot(osb, slot);
> -			break;
> -		}
> -
> -		ocfs2_free_ac_resource(ac);
> -	}
> -
> -	return status;
> -}
> -
>  int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
>  			    struct ocfs2_alloc_context **ac)
>  {
>  	int status;
> -	s16 slot = ocfs2_get_inode_steal_slot(osb);
> +	int slot = ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
>  	u64 alloc_group;
>  
>  	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
> @@ -761,7 +786,7 @@ int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
>  	alloc_group = osb->osb_inode_alloc_group;
>  	status = ocfs2_reserve_suballoc_bits(osb, *ac,
>  					     INODE_ALLOC_SYSTEM_INODE,
> -					     osb->slot_num,
> +					     (u32)osb->slot_num,
>  					     &alloc_group,
>  					     ALLOC_NEW_GROUP |
>  					     ALLOC_GROUPS_FROM_GLOBAL);
> @@ -779,7 +804,7 @@ int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
>  		 * from our own next time.
>  		 */
>  		if (slot != OCFS2_INVALID_SLOT)
> -			ocfs2_init_inode_steal_slot(osb);
> +			ocfs2_init_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
>  		goto bail;
>  	} else if (status < 0 && status != -ENOSPC) {
>  		mlog_errno(status);
> @@ -789,7 +814,8 @@ int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
>  	ocfs2_free_ac_resource(*ac);
>  
>  inode_steal:
> -	status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
> +	status = ocfs2_steal_from_other_nodes(osb, *ac,
> +					      INODE_ALLOC_SYSTEM_INODE);
>  	atomic_inc(&osb->s_num_inodes_stolen);
>  	if (status < 0) {
>  		if (status != -ENOSPC)
> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
> index 2606991..a87c1c5 100644
> --- a/fs/ocfs2/super.c
> +++ b/fs/ocfs2/super.c
> @@ -301,9 +301,12 @@ static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
>  
>  	spin_lock(&osb->osb_lock);
>  	out += snprintf(buf + out, len - out,
> -			"%10s => Slot: %d  NumStolen: %d\n", "Steal",
> +			"%10s => I_Slot: %d  I_NumStolen : %d, EB_Slot: %d  "
> +			"EB_NumStolen : %d\n", "Steal",
>  			osb->s_inode_steal_slot,
> -			atomic_read(&osb->s_num_inodes_stolen));
> +			atomic_read(&osb->s_num_inodes_stolen),
> +			osb->s_extent_steal_slot,
> +			atomic_read(&osb->s_num_extents_stolen));
>  	spin_unlock(&osb->osb_lock);
>  
>  	out += snprintf(buf + out, len - out, "OrphanScan => ");
> @@ -1997,7 +2000,7 @@ static int ocfs2_initialize_super(struct super_block *sb,
>  	osb->blocked_lock_count = 0;
>  	spin_lock_init(&osb->osb_lock);
>  	spin_lock_init(&osb->osb_xattr_lock);
> -	ocfs2_init_inode_steal_slot(osb);
> +	ocfs2_init_all_steal_slot(osb);
>  
>  	atomic_set(&osb->alloc_stats.moves, 0);
>  	atomic_set(&osb->alloc_stats.local_data, 0);
> diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
> index 8fc6fb0..8ae4e5d 100644
> --- a/fs/ocfs2/xattr.c
> +++ b/fs/ocfs2/xattr.c
> @@ -2282,7 +2282,7 @@ static int ocfs2_create_xattr_block(handle_t *handle,
>  	xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
>  	memset(xblk, 0, inode->i_sb->s_blocksize);
>  	strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
> -	xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
> +	xblk->xb_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
>  	xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
>  	xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
>  	xblk->xb_blkno = cpu_to_le64(first_blkno);
>   

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-ocfs2-add-extent-block-stealing-for-ocfs2.patch
Type: text/x-patch
Size: 13287 bytes
Desc: not available
Url : http://oss.oracle.com/pipermail/ocfs2-devel/attachments/20100118/884e7d91/attachment.bin