[RFC][PATCH 8/9]ext4: Add the EXT4_IOC_MOVE_VICTIM ioctl

[RFC][PATCH 8/9]ext4: Add the EXT4_IOC_MOVE_VICTIM ioctl

[Akira Fujita]

ext4: online defrag -- Add the EXT4_IOC_MOVE_VICTIM ioctl.

From: Akira Fujita <a-fujita@rs.jp.nec.com>

The EXT4_IOC_MOVE_VICTIM moves the victim extents into other block group.
Therefore the contiguous free space is made in the target block group.
This ioctl is used only in the force defrag (-f).

Signed-off-by: Akira Fujita <a-fujita@rs.jp.nec.com>
Signed-off-by: Takashi Sato <t-sato@yk.jp.nec.com>
---
 fs/ext4/balloc.c       |    1 +
 fs/ext4/defrag.c       |  264 ++++++++++++++++++++++++++++++++++++++++++-----
 fs/ext4/ext4.h         |   20 ++++-
 fs/ext4/ext4_extents.h |    3 +
 fs/ext4/extents.c      |   52 ++++++++--
 fs/ext4/ioctl.c        |   16 +++-
 fs/ext4/mballoc.c      |    5 +
 fs/ext4/mballoc.h      |    1 +
 8 files changed, 322 insertions(+), 40 deletions(-)

diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index b9821be..e791484 100644
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -705,6 +705,7 @@ static ext4_fsblk_t do_blk_alloc(handle_t *handle, struct inode *inode,
 	ar.goal = goal;
 	ar.len = *count;
 	ar.logical = iblock;
+	ar.excepted_group = -1;

 	if (S_ISREG(inode->i_mode) && !(flags & EXT4_META_BLOCK))
 		/* enable in-core preallocation for data block allocation */
diff --git a/fs/ext4/defrag.c b/fs/ext4/defrag.c
index 67030bc..20cb847 100644
--- a/fs/ext4/defrag.c
+++ b/fs/ext4/defrag.c
@@ -204,6 +204,92 @@ out:
 }

 /**
+ * ext4_defrag_move_victim - Create free space for defrag
+ *
+ * @target_filp:	target file
+ * @ext_info:		target extents array to move
+ *
+ * This function returns 0 if succeed, otherwise
+ * returns error value.
+ */
+int
+ext4_defrag_move_victim(struct file *target_filp,
+			struct ext4_extents_info *ext_info)
+{
+	struct inode *org_inode = target_filp->f_dentry->d_inode;
+	struct super_block *sb = org_inode->i_sb;
+	struct file victim_file;
+	struct dentry victim_dent;
+	struct inode *victim_inode;
+	struct ext4_extent_data ext;
+	ext4_fsblk_t goal = ext_info->goal;
+	ext4_group_t group;
+	ext4_grpblk_t grp_off;
+	int ret, i;
+
+	/* Setup dummy extent data */
+	ext.len = 0;
+
+	/* Get the inode of the victim file */
+	victim_inode = ext4_iget(sb, ext_info->ino);
+	if (IS_ERR(victim_inode))
+		return PTR_ERR(victim_inode);
+
+	/*
+	 * If the permission of victim inode has been changed
+	 * since we checked before, defrag will fail.
+	 * Because defrag won't be albe to move the extents of victim inode
+	 * to other block group.
+	 */
+	if (!capable(CAP_DAC_OVERRIDE)) {
+		if ((victim_inode->i_mode & S_IRUSR) != S_IRUSR)
+			return -EACCES;
+		if (current->fsuid != victim_inode->i_uid)
+			return -EACCES;
+	}
+
+	/* Setup file for the victim file */
+	victim_dent.d_inode = victim_inode;
+	victim_file.f_dentry = &victim_dent;
+	victim_file.f_mapping = victim_inode->i_mapping;
+
+	/*
+	 * If goal has not been set in the user space,
+	 * set the start offset of block group where victim inode is
+	 * located to goal.
+	 */
+	if (goal == -1) {
+		ext4_get_group_no_and_offset(victim_inode->i_sb,
+				ext_info->ext[0].start, &group, &grp_off);
+		goal = ext4_group_first_block_no(sb, group + 1);
+	}
+
+	for (i = 0; i < ext_info->entries; i++) {
+		/* Move original blocks to another block group */
+		ret = ext4_defrag(&victim_file, ext_info->ext[i].block,
+			ext_info->ext[i].len, goal, DEFRAG_FORCE_VICTIM, &ext);
+		if (ret < 0) {
+			printk(KERN_ERR "ext4 defrag: "
+				"Moving victim file failed. ino [%llu]\n",
+				ext_info->ino);
+			goto out;
+		}
+
+		/* Sync journal blocks */
+		ret = ext4_force_commit(sb);
+		if (ret) {
+			printk(KERN_ERR "ext4 defrag: "
+				"ext4_force_commit failed(%d)\n", ret);
+			goto out;
+		}
+	}
+
+out:
+	iput(victim_inode);
+	return ret;
+}
+
+/**
  * ext4_defrag_merge_across_blocks - Merge extents across leaf block
  *
  * @handle:		journal handle
@@ -213,6 +299,7 @@ out:
  * @start_ext:		first new extent to be merged
  * @new_ext:		middle of new extent to be merged
  * @end_ext:		last new extent to be merged
+ * @phase:		phase of the force defrag mode
  *
  * This function returns 0 if succeed, otherwise returns error value.
  */
@@ -220,14 +307,20 @@ static int
 ext4_defrag_merge_across_blocks(handle_t *handle, struct inode *org_inode,
 		struct ext4_extent *o_start, struct ext4_extent *o_end,
 		struct ext4_extent *start_ext, struct ext4_extent *new_ext,
-		struct ext4_extent *end_ext)
+		struct ext4_extent *end_ext, int phase)
 {
 	struct ext4_ext_path *org_path = NULL;
 	ext4_lblk_t eblock = 0;
 	int new_flag = 0;
 	int end_flag = 0;
+	int defrag_flag;
 	int err;

+	if (phase == DEFRAG_FORCE_VICTIM)
+		defrag_flag = 1;
+	else
+		defrag_flag = 0;
+
 	if (le16_to_cpu(start_ext->ee_len) &&
 		le16_to_cpu(new_ext->ee_len) &&
 		le16_to_cpu(end_ext->ee_len)) {
@@ -304,8 +397,8 @@ ext4_defrag_merge_across_blocks(handle_t *handle, struct inode *org_inode,
 			org_path = NULL;
 			goto out;
 		}
-		err = ext4_ext_insert_extent(handle, org_inode,
-					org_path, new_ext);
+		err = ext4_ext_insert_extent_defrag(handle, org_inode,
+					org_path, new_ext, defrag_flag);
 		if (err)
 			goto out;
 	}
@@ -318,8 +411,8 @@ ext4_defrag_merge_across_blocks(handle_t *handle, struct inode *org_inode,
 			org_path = NULL;
 			goto out;
 		}
-		err = ext4_ext_insert_extent(handle, org_inode,
-					org_path, end_ext);
+		err = ext4_ext_insert_extent_defrag(handle, org_inode,
+					org_path, end_ext, defrag_flag);
 		if (err)
 			goto out;
 	}
@@ -397,6 +490,7 @@ ext4_defrag_merge_inside_block(struct ext4_extent *o_start,
  * @new_ext:	middle of new extent to be merged
  * @end_ext:	last new extent to be merged
  * @replaced:	the number of blocks which will be replaced with new_ext
+ * @phase:	phase of the force defrag mode
  *
  * This function returns 0 if succeed, otherwise returns error value.
  */
@@ -405,7 +499,7 @@ ext4_defrag_merge_extents(handle_t *handle, struct inode *org_inode,
 		struct ext4_ext_path *org_path,
 		struct ext4_extent *o_start, struct ext4_extent *o_end,
 		struct ext4_extent *start_ext, struct ext4_extent *new_ext,
-		struct ext4_extent *end_ext, ext4_fsblk_t replaced)
+		struct ext4_extent *end_ext, ext4_fsblk_t replaced, int phase)
 {
 	struct  ext4_extent_header *eh;
 	unsigned need_slots, slots_range;
@@ -443,7 +537,7 @@ ext4_defrag_merge_extents(handle_t *handle, struct inode *org_inode,

 		ret = ext4_defrag_merge_across_blocks(handle, org_inode,
 					o_start, o_end, start_ext, new_ext,
-					end_ext);
+					end_ext, phase);
 		if (ret < 0)
 			return ret;
 	} else {
@@ -476,13 +570,14 @@ ext4_defrag_merge_extents(handle_t *handle, struct inode *org_inode,
  * @org_path:		path indicates first extent to be defraged
  * @dext:		destination extent
  * @from:		start offset on the target file
+ * @phase:		phase of the force defrag mode
  *
  * This function returns 0 if succeed, otherwise returns error value.
  */
 static int
 ext4_defrag_leaf_block(handle_t *handle, struct inode *org_inode,
 		struct ext4_ext_path *org_path, struct ext4_extent *dext,
-		ext4_lblk_t *from)
+		ext4_lblk_t *from, int phase)
 {
 	struct ext4_extent *oext, *o_start = NULL, *o_end = NULL, *prev_ext;
 	struct ext4_extent new_ext, start_ext, end_ext;
@@ -583,7 +678,7 @@ ext4_defrag_leaf_block(handle_t *handle, struct inode *org_inode,
 				+ le16_to_cpu(oext->ee_len) - 1) {
 			ret = ext4_defrag_merge_extents(handle, org_inode,
 					org_path, o_start, o_end, &start_ext,
-					&new_ext, &end_ext, replaced);
+					&new_ext, &end_ext, replaced, phase);
 			if (ret < 0)
 				return ret;

@@ -635,6 +730,7 @@ ext4_defrag_leaf_block(handle_t *handle, struct inode *org_inode,
  * @from:		block offset of org_inode
  * @dest_off:		block offset of dest_inode
  * @count:		block count to be replaced
+ * @phase:              phase of the force defrag mode
  *
  * This function returns 0 if succeed, otherwise returns error value.
  * Replace extents for blocks from "from" to "from + count - 1".
@@ -642,7 +738,7 @@ ext4_defrag_leaf_block(handle_t *handle, struct inode *org_inode,
 static int
 ext4_defrag_replace_branches(handle_t *handle, struct inode *org_inode,
 			struct inode *dest_inode, ext4_lblk_t from,
-			ext4_lblk_t dest_off, ext4_lblk_t count)
+			ext4_lblk_t dest_off, ext4_lblk_t count, int phase)
 {
 	struct ext4_ext_path *org_path = NULL;
 	struct ext4_ext_path *dest_path = NULL;
@@ -703,7 +799,7 @@ ext4_defrag_replace_branches(handle_t *handle, struct inode *org_inode,

 		/* Loop for the original extent blocks */
 		err = ext4_defrag_leaf_block(handle, org_inode,
-						org_path, dext, &from);
+						org_path, dext, &from, phase);
 		if (err < 0)
 			goto out;

@@ -713,7 +809,7 @@ ext4_defrag_replace_branches(handle_t *handle, struct inode *org_inode,
 		 * e.g. ext4_defrag_merge_extents()
 		 */
 		err = ext4_defrag_leaf_block(handle, dest_inode,
-					dest_path, swap_ext, &dest_off);
+					dest_path, swap_ext, &dest_off, -1);
 		if (err < 0)
 			goto out;

@@ -809,6 +905,7 @@ out:
  * @req_blocks:		contiguous blocks count we need
  * @iblock:		target file offset
  * @goal:		goal offset
+ * @phase:              phase of the force defrag mode
  *
  */
 static void
@@ -817,8 +914,22 @@ ext4_defrag_fill_ar(struct inode *org_inode, struct inode *dest_inode,
 			struct ext4_ext_path *org_path,
 			struct ext4_ext_path *dest_path,
 			ext4_fsblk_t req_blocks, ext4_lblk_t iblock,
-			ext4_fsblk_t goal)
+			ext4_fsblk_t goal, int phase)
 {
+	ext4_group_t org_grp_no;
+	ext4_grpblk_t org_blk_off;
+	int org_depth = ext_depth(org_inode);
+
+	if (phase == DEFRAG_FORCE_VICTIM) {
+		ext4_get_group_no_and_offset(org_inode->i_sb,
+				ext_pblock(org_path[org_depth].p_ext),
+				&org_grp_no, &org_blk_off);
+		ar->excepted_group = org_grp_no;
+	} else {
+		/* Allocate contiguous blocks to any block group */
+		ar->excepted_group = -1;
+	}
+
 	ar->inode = dest_inode;
 	ar->len = req_blocks;
 	ar->logical = iblock;
@@ -882,6 +993,61 @@ ext4_defrag_alloc_blocks(handle_t *handle, struct inode *org_inode,
 }

 /**
+ * ext4_defrag_check_phase
+ * 	- Check condition of the allocated blocks (only force defrag mode)
+ *
+ * @ar:			allocation request for multiple block allocation
+ * @dest_grp_no:	block group num of the allocated blocks
+ * @goal_grp_no:	block group num of the destination of block allocation
+ * @alloc_total:	sum total of the allocated blocks
+ * @req_blocks:		contiguous blocks count we need
+ * @phase:              phase of the force defrag mode
+ *
+ * This function returns 0 if succeed, otherwise returns error value.
+ */
+static int
+ext4_defrag_check_phase(struct ext4_allocation_request *ar,
+			ext4_group_t dest_grp_no, ext4_group_t goal_grp_no,
+			ext4_fsblk_t alloc_total, ext4_lblk_t req_blocks,
+			int phase)
+{
+	int err = 0;
+
+	switch (phase) {
+	case DEFRAG_FORCE_TRY:
+		/* If there is not enough space, return -ENOSPC. */
+		if (ar->len != req_blocks)
+			/* -ENOSPC triggers DEFRAG_FORCE_VICTIM phase. */
+			err = -ENOSPC;
+		break;
+	case DEFRAG_FORCE_VICTIM:
+		/* We can't allocate new blocks in the same block group. */
+		if (dest_grp_no == ar->excepted_group) {
+			printk(KERN_ERR "ext4 defrag: Failed to allocate"
+					" victim file to other block group\n");
+			err = -ENOSPC;
+		}
+		break;
+	case DEFRAG_FORCE_GATHER:
+		/*
+		 * If the target free area made by moving victim
+		 * extents has been already used by other process,
+		 * defrag returns error value and fails.
+		 */
+		if (dest_grp_no != goal_grp_no
+		    || alloc_total != req_blocks) {
+			printk(KERN_ERR "ext4 defrag: The target contiguous"
+					" free area has been already used"
+					" by other process\n");
+			err = -EIO;
+		}
+		break;
+	}
+
+	return err;
+}
+
+/**
  * ext4_defrag_partial - Defrag a file per page
  *
  * @tmp_inode:			temporary inode
@@ -890,13 +1056,15 @@ ext4_defrag_alloc_blocks(handle_t *handle, struct inode *org_inode,
  * @dest_blk_offset:		block index on temporary file
  * @data_offset_in_page:	block index where data swapping starts
  * @block_len_in_page:		the number of blocks to be swapped
+ * @phase:			phase of the force defrag mode
  *
  * This function returns 0 if succeed, otherwise returns error value.
  */
 static int
 ext4_defrag_partial(struct inode *tmp_inode, struct file *filp,
 			pgoff_t org_page_offset, ext4_lblk_t dest_blk_offset,
-			int data_offset_in_page, int block_len_in_page)
+			int data_offset_in_page, int block_len_in_page,
+			int phase)
 {
 	struct inode *org_inode = filp->f_dentry->d_inode;
 	struct address_space *mapping = org_inode->i_mapping;
@@ -979,7 +1147,7 @@ ext4_defrag_partial(struct inode *tmp_inode, struct file *filp,
 	try_to_release_page(page, 0);
 	ret = ext4_defrag_replace_branches(handle, org_inode, tmp_inode,
 						org_blk_offset, dest_blk_offset,
-						block_len_in_page);
+						block_len_in_page, phase);
 	if (ret < 0)
 		goto out;

@@ -1030,6 +1198,7 @@ out:
  * @tar_end:		the last block number of the allocated blocks
  * @sum_tmp:		the extents count  in the allocated blocks
  * @goal:		block offset for allocation
+ * @phase:		phase of the force defrag mode
  *
  * This function returns the values as below.
  *	0 (improved)
@@ -1039,7 +1208,7 @@ out:
 static int
 ext4_defrag_comp_ext_count(struct inode *org_inode,
 			struct ext4_ext_path *org_path, ext4_lblk_t tar_end,
-			int sum_tmp, ext4_fsblk_t goal)
+			int sum_tmp, ext4_fsblk_t goal, int phase)
 {
 	struct ext4_extent *ext = NULL;
 	int depth = ext_depth(org_inode);
@@ -1066,7 +1235,8 @@ ext4_defrag_comp_ext_count(struct inode *org_inode,
 			if (sum_org == sum_tmp && !goal) {
 				/* Not improved */
 				ret = 1;
-			} else if (sum_org < sum_tmp) {
+			} else if (sum_org < sum_tmp &&
+					phase != DEFRAG_FORCE_VICTIM) {
 				/* Fragment increased */
 				ret = -ENOSPC;
 				printk(KERN_ERR "ext4 defrag: "
@@ -1095,6 +1265,7 @@ ext4_defrag_comp_ext_count(struct inode *org_inode,
  * @req_blocks:		the number of blocks to allocate
  * @iblock:		file related offset
  * @goal:		block offset for allocation
+ * @phase:		phase of the force defrag mode
  *
  * This function returns the value as below:
  *	0 (succeed)
@@ -1105,7 +1276,7 @@ static int
 ext4_defrag_new_extent_tree(struct inode *org_inode, struct inode *tmp_inode,
 			struct ext4_ext_path *org_path, ext4_lblk_t req_start,
 			ext4_lblk_t req_blocks, ext4_lblk_t iblock,
-			ext4_fsblk_t goal)
+			ext4_fsblk_t goal, int phase)
 {
 	handle_t *handle;
 	struct ext4_sb_info *sbi = EXT4_SB(org_inode->i_sb);
@@ -1117,6 +1288,8 @@ ext4_defrag_new_extent_tree(struct inode *org_inode, struct inode *tmp_inode,
 	ext4_fsblk_t newblock = 0;
 	ext4_lblk_t req_end = req_start + req_blocks - 1;
 	ext4_lblk_t rest_blocks = 0;
+	ext4_group_t dest_group_no, goal_group_no;
+	ext4_grpblk_t dest_blk_off, goal_blk_off;
 	int sum_tmp = 0;
 	int metadata = 1;
 	int ret;
@@ -1133,7 +1306,7 @@ ext4_defrag_new_extent_tree(struct inode *org_inode, struct inode *tmp_inode,

 	/* Fill struct ext4_allocation_request with necessary info */
 	ext4_defrag_fill_ar(org_inode, tmp_inode, &ar, org_path,
-				dest_path, req_blocks, iblock, goal);
+				dest_path, req_blocks, iblock, goal, phase);

 	handle = ext4_journal_start(tmp_inode, 0);
 	if (IS_ERR(handle)) {
@@ -1141,6 +1314,9 @@ ext4_defrag_new_extent_tree(struct inode *org_inode, struct inode *tmp_inode,
 		goto out2;
 	}

+	ext4_get_group_no_and_offset(tmp_inode->i_sb, goal,
+				&goal_group_no, &goal_blk_off);
+
 	while (alloc_total != req_blocks) {
 		/* Allocate blocks */
 		ret = ext4_defrag_alloc_blocks(handle, org_inode, tmp_inode,
@@ -1150,9 +1326,21 @@ ext4_defrag_new_extent_tree(struct inode *org_inode, struct inode *tmp_inode,
 		/* Claimed blocks are already reserved */
 		EXT4_I(ar.inode)->i_delalloc_reserved_flag = 1;

+		ext4_get_group_no_and_offset(tmp_inode->i_sb, newblock,
+					&dest_group_no, &dest_blk_off);
+
 		alloc_total += ar.len;
 		rest_blocks = req_blocks - alloc_total;

+		/* the checks that done in force mode */
+		if (phase) {
+			ret = ext4_defrag_check_phase(&ar, dest_group_no,
+					goal_group_no, alloc_total,
+					req_blocks, phase);
+			if (ret < 0)
+				goto out;
+		}
+
 		newex.ee_block = cpu_to_le32(alloc_total - ar.len);
 		ext4_ext_store_pblock(&newex, newblock);
 		newex.ee_len = cpu_to_le16(ar.len);
@@ -1162,13 +1350,14 @@ ext4_defrag_new_extent_tree(struct inode *org_inode, struct inode *tmp_inode,
 		if (ret < 0)
 			goto out;

-		ar.goal = newblock + ar.len;
+		if (!phase)
+			ar.goal = newblock + ar.len;
 		ar.len = req_blocks - alloc_total;
 		sum_tmp++;
 	}

 	ret = ext4_defrag_comp_ext_count(org_inode, org_path, req_end,
-					sum_tmp, goal);
+					sum_tmp, goal, phase);

 out:
 	if (ret < 0 && ar.len)
@@ -1195,14 +1384,16 @@ out2:
  * ext4_defrag_check - Check the environment whether a defrag can be done
  *
  * @org_inode:		original inode
+ * @ext:		extent to be moved (only defrag force mode)
  * @defrag_size:	size of defrag in blocks
  * @goal:		pointer to block offset for allocation
+ * @phase:		phase of the force defrag mode
  *
  * This function returns 0 if succeed, otherwise returns error value.
  */
 static int
-ext4_defrag_check(struct inode *org_inode, ext4_lblk_t defrag_size,
-		ext4_fsblk_t *goal)
+ext4_defrag_check(struct inode *org_inode, struct ext4_extent_data *ext,
+		ext4_lblk_t defrag_size, ext4_fsblk_t *goal, int phase)
 {

 	/* Ext4 online defrag supports only extent based file */
@@ -1223,6 +1414,19 @@ ext4_defrag_check(struct inode *org_inode, ext4_lblk_t defrag_size,
 		return -EOPNOTSUPP;
 	}

+	if (phase == DEFRAG_FORCE_GATHER) {
+		if (!ext->len || ext->len < defrag_size) {
+			printk(KERN_ERR "ext4_defrag: "
+				"Invalid extent length\n");
+			return -EINVAL;
+		}
+		/*
+		 * Set the goal offset in the start offset of contiguous
+		 * free space that was made because vicim inode moved.
+		 */
+		*goal = ext->start;
+	}
+
 	return 0;
 }

@@ -1304,13 +1508,16 @@ out:
  * @block_start:	starting offset to defrag in blocks
  * @defrag_size:	size of defrag in blocks
  * @goal:		block offset for allocation
+ * @phase:		phase of the force defrag mode
+ * @ext:		extent to be moved (only defrag force mode)
  *
  * This function returns the number of blocks if succeed, otherwise
  * returns error value.
  */
 int
 ext4_defrag(struct file *filp, ext4_lblk_t block_start,
-		ext4_lblk_t defrag_size, ext4_fsblk_t goal)
+		ext4_lblk_t defrag_size, ext4_fsblk_t goal, int phase,
+		struct ext4_extent_data *ext)
 {
 	struct inode *org_inode = filp->f_dentry->d_inode, *tmp_inode = NULL;
 	struct ext4_ext_path *org_path = NULL, *holecheck_path = NULL;
@@ -1325,7 +1532,7 @@ ext4_defrag(struct file *filp, ext4_lblk_t block_start,
 	int block_len_in_page;

 	/* Check the filesystem environment whether defrag can be done */
-	ret = ext4_defrag_check(org_inode, defrag_size, &goal);
+	ret = ext4_defrag_check(org_inode, ext, defrag_size, &goal, phase);
 	if (ret < 0)
 		return ret;

@@ -1442,11 +1649,11 @@ ext4_defrag(struct file *filp, ext4_lblk_t block_start,

 		ret = ext4_defrag_new_extent_tree(org_inode, tmp_inode,
 					org_path, seq_start, seq_blocks,
-					block_start, goal);
+					block_start, goal, phase);

 		if (ret < 0) {
 			break;
-		} else if (ret == 1) {
+		} else if (ret == 1 && (!goal || (goal && !phase))) {
 			ret = 0;
 			seq_start = le32_to_cpu(ext_cur->ee_block);
 			goto CLEANUP;
@@ -1491,7 +1698,8 @@ ext4_defrag(struct file *filp, ext4_lblk_t block_start,
 						org_page_offset,
 						dest_block_offset,
 						data_offset_in_page,
-						block_len_in_page);
+						block_len_in_page,
+						phase);
 			if (ret < 0)
 				goto out;

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index c72703f..e0b09dc 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -97,6 +97,11 @@ struct ext4_allocation_request {
 	unsigned long len;
 	/* flags. see above EXT4_MB_HINT_* */
 	unsigned long flags;
+	/*
+	 * for ext4 online defrag:
+	 * the block group which is excepted from allocation target
+	 */
+	long long excepted_group;
 };

 /*
@@ -306,6 +311,7 @@ struct ext4_new_group_data {
 #define EXT4_IOC_SUPER_INFO		_IOR('f', 16, struct ext4_super_block)
 #define EXT4_IOC_FREE_BLOCKS_INFO	_IOWR('f', 17, struct ext4_extents_info)
 #define EXT4_IOC_FIEMAP_INO		_IOWR('f', 18, struct fiemap_ino)
+#define EXT4_IOC_MOVE_VICTIM		_IOW('f', 19, struct ext4_extents_info)

 /*
  * ioctl commands in 32 bit emulation
@@ -328,8 +334,15 @@ struct ext4_new_group_data {
  *
  * DEFRAG_MAX_ENT:	the maximum number of extents for exchanging between
  *			kernel-space and user-space per an ioctl
+ * DEFRAG_FORCE_TRY:	check whether we have free space fragmentation or not
+ * DEFRAG_FORCE_VICTIM:	move victim extents to make sufficient space
+ * DEFRAG_FORCE_GATHER:	move the target file into the free space made in the
+ *			DEFRAG_FORCE_VICTIM phase
  */
 #define DEFRAG_MAX_ENT		32
+#define DEFRAG_FORCE_TRY	1
+#define DEFRAG_FORCE_VICTIM	2
+#define DEFRAG_FORCE_GATHER	3

 struct ext4_extent_data {
 	ext4_lblk_t block;		/* start logical block number */
@@ -341,6 +354,8 @@ struct ext4_ext_defrag_data {
 	ext4_lblk_t start_offset;	/* start offset to defrag in blocks */
 	ext4_lblk_t defrag_size;	/* size of defrag in blocks */
 	ext4_fsblk_t goal;		/* block offset for allocation */
+	int flag;			/* free space mode flag */
+	struct ext4_extent_data ext;
 };

 struct ext4_extents_info {
@@ -1190,8 +1205,11 @@ extern void ext4_inode_table_set(struct super_block *sb,
 extern int ext4_ext_journal_restart(handle_t *handle, int needed);
 /* defrag.c */
 extern int ext4_defrag(struct file *filp, ext4_lblk_t block_start,
-			ext4_lblk_t defrag_size, ext4_fsblk_t goal);
+			ext4_lblk_t defrag_size, ext4_fsblk_t goal,
+			int flag, struct ext4_extent_data *ext);
 extern int ext4_defrag_fiemap_ino(struct file *filp, unsigned long arg);
+extern int ext4_defrag_move_victim(struct file *target_filp,
+				struct ext4_extents_info *ext_info);

 static inline ext4_fsblk_t ext4_blocks_count(struct ext4_super_block *es)
 {
diff --git a/fs/ext4/ext4_extents.h b/fs/ext4/ext4_extents.h
index 39e043b..9b9dee9 100644
--- a/fs/ext4/ext4_extents.h
+++ b/fs/ext4/ext4_extents.h
@@ -250,6 +250,9 @@ extern ext4_fsblk_t ext_pblock(struct ext4_extent *ex);
 extern ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 					struct ext4_ext_path *path,
 					ext4_lblk_t block);
+extern int ext4_ext_insert_extent_defrag(handle_t *handle, struct inode *inode,
+					struct ext4_ext_path *path,
+					struct ext4_extent *newext, int defrag);

 #endif /* _EXT4_EXTENTS */

diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index dde234f..a17f53c 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -186,11 +186,17 @@ ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 static ext4_fsblk_t
 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 			struct ext4_ext_path *path,
-			struct ext4_extent *ex, int *err)
+			struct ext4_extent *ex, int *err,
+			ext4_fsblk_t defrag_goal)
 {
 	ext4_fsblk_t goal, newblock;

-	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
+	if (defrag_goal)
+		goal = defrag_goal;
+	else
+		goal = ext4_ext_find_goal(inode, path,
+					le32_to_cpu(ex->ee_block));
+
 	newblock = ext4_new_meta_block(handle, inode, goal, err);
 	return newblock;
 }
@@ -675,7 +681,8 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
  */
 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 				struct ext4_ext_path *path,
-				struct ext4_extent *newext, int at)
+				struct ext4_extent *newext, int at,
+				ext4_fsblk_t defrag_goal)
 {
 	struct buffer_head *bh = NULL;
 	int depth = ext_depth(inode);
@@ -726,7 +733,7 @@ static int ext4_ext_split(handle_t *handle, struct inode *inode,
 	ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 	for (a = 0; a < depth - at; a++) {
 		newblock = ext4_ext_new_meta_block(handle, inode, path,
-						   newext, &err);
+						   newext, &err, defrag_goal);
 		if (newblock == 0)
 			goto cleanup;
 		ablocks[a] = newblock;
@@ -913,7 +920,8 @@ cleanup:
  */
 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
 					struct ext4_ext_path *path,
-					struct ext4_extent *newext)
+					struct ext4_extent *newext,
+					ext4_fsblk_t defrag_goal)
 {
 	struct ext4_ext_path *curp = path;
 	struct ext4_extent_header *neh;
@@ -922,7 +930,8 @@ static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
 	ext4_fsblk_t newblock;
 	int err = 0;

-	newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
+	newblock = ext4_ext_new_meta_block(handle, inode, path,
+					   newext, &err, defrag_goal);
 	if (newblock == 0)
 		return err;

@@ -998,7 +1007,8 @@ out:
  */
 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
 					struct ext4_ext_path *path,
-					struct ext4_extent *newext)
+					struct ext4_extent *newext,
+					ext4_fsblk_t defrag_goal)
 {
 	struct ext4_ext_path *curp;
 	int depth, i, err = 0;
@@ -1018,7 +1028,8 @@ repeat:
 	if (EXT_HAS_FREE_INDEX(curp)) {
 		/* if we found index with free entry, then use that
 		 * entry: create all needed subtree and add new leaf */
-		err = ext4_ext_split(handle, inode, path, newext, i);
+		err = ext4_ext_split(handle, inode, path, newext, i,
+					defrag_goal);
 		if (err)
 			goto out;

@@ -1031,7 +1042,8 @@ repeat:
 			err = PTR_ERR(path);
 	} else {
 		/* tree is full, time to grow in depth */
-		err = ext4_ext_grow_indepth(handle, inode, path, newext);
+		err = ext4_ext_grow_indepth(handle, inode, path,
+						newext, defrag_goal);
 		if (err)
 			goto out;

@@ -1477,6 +1489,19 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
 				struct ext4_ext_path *path,
 				struct ext4_extent *newext)
 {
+	return ext4_ext_insert_extent_defrag(handle, inode, path, newext, 0);
+}
+
+/*
+ * ext4_ext_insert_extent_defrag:
+ * The difference from ext4_ext_insert_extent is to use the first block
+ * in newext as the goal of the new index block.
+ */
+int
+ext4_ext_insert_extent_defrag(handle_t *handle, struct inode *inode,
+				struct ext4_ext_path *path,
+				struct ext4_extent *newext, int defrag)
+{
 	struct ext4_extent_header *eh;
 	struct ext4_extent *ex, *fex;
 	struct ext4_extent *nearex; /* nearest extent */
@@ -1484,6 +1509,7 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
 	int depth, len, err;
 	ext4_lblk_t next;
 	unsigned uninitialized = 0;
+	ext4_fsblk_t defrag_goal;

 	BUG_ON(ext4_ext_get_actual_len(newext) == 0);
 	depth = ext_depth(inode);
@@ -1544,11 +1570,16 @@ repeat:
 			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 	}

+	if (defrag)
+		defrag_goal = ext_pblock(newext);
+	else
+		defrag_goal = 0;
 	/*
 	 * There is no free space in the found leaf.
 	 * We're gonna add a new leaf in the tree.
 	 */
-	err = ext4_ext_create_new_leaf(handle, inode, path, newext);
+	err = ext4_ext_create_new_leaf(handle, inode, path,
+					newext, defrag_goal);
 	if (err)
 		goto cleanup;
 	depth = ext_depth(inode);
@@ -2845,6 +2876,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
 	ar.goal = ext4_ext_find_goal(inode, path, iblock);
 	ar.logical = iblock;
 	ar.len = allocated;
+	ar.excepted_group = -1;
 	if (S_ISREG(inode->i_mode))
 		ar.flags = EXT4_MB_HINT_DATA;
 	else
diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
index b69b54a..15050d3 100644
--- a/fs/ext4/ioctl.c
+++ b/fs/ext4/ioctl.c
@@ -242,7 +242,8 @@ setversion_out:
 		}

 		err = ext4_defrag(filp, defrag.start_offset,
-			defrag.defrag_size, defrag.goal);
+			defrag.defrag_size, defrag.goal, defrag.flag,
+			&defrag.ext);
 		return err;
 	}

@@ -283,6 +284,19 @@ setversion_out:
 		return err;
 	}

+	case EXT4_IOC_MOVE_VICTIM: {
+		struct ext4_extents_info ext_info;
+		int err;
+
+		if (copy_from_user(&ext_info,
+			(struct ext4_extents_info __user *)arg,
+						sizeof(ext_info)))
+			return -EFAULT;
+
+		err = ext4_defrag_move_victim(filp, &ext_info);
+		return err;
+	}
+
 	case EXT4_IOC_GROUP_ADD: {
 		struct ext4_new_group_data input;
 		struct super_block *sb = inode->i_sb;
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index eebcf7c..d523e42 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -1774,6 +1774,10 @@ repeat:
 			if (group == EXT4_SB(sb)->s_groups_count)
 				group = 0;

+			if (ac->ac_excepted_group != -1 &&
+			    group == ac->ac_excepted_group)
+				continue;
+
 			/* quick check to skip empty groups */
 			grp = ext4_get_group_info(ac->ac_sb, group);
 			if (grp->bb_free == 0)
@@ -4069,6 +4073,7 @@ ext4_mb_initialize_context(struct ext4_allocation_context *ac,
 	ac->ac_bitmap_page = NULL;
 	ac->ac_buddy_page = NULL;
 	ac->ac_lg = NULL;
+	ac->ac_excepted_group = ar->excepted_group;

 	/* we have to define context: we'll we work with a file or
 	 * locality group. this is a policy, actually */
diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h
index b5dff1f..557308a 100644
--- a/fs/ext4/mballoc.h
+++ b/fs/ext4/mballoc.h
@@ -219,6 +219,7 @@ struct ext4_allocation_context {
 	struct page *ac_buddy_page;
 	struct ext4_prealloc_space *ac_pa;
 	struct ext4_locality_group *ac_lg;
+	long long ac_excepted_group;
 };

 #define AC_STATUS_CONTINUE	1