Message-ID: <20071206161045.1054bbe7@gara>
Date:	Thu, 6 Dec 2007 16:10:45 -0600
From:	"Jose R. Santos" <jrs@...ibm.com>
To:	linux-ext4 <linux-ext4@...r.kernel.org>
Subject: [RFC] [PATCH] Flex_BG ialloc awareness V2.

Hi folks,

New version of the Flex_BG ialloc allocation patch.

Changes from the last version:

- Size of the FLEX_BG in now written to the super block at mke2fs time
  instead of calculating at mount time (testing patch for e2fsprog's
  next branch attached).

- Rename a lots of the confusing "meta" terms as suggested by Andreas.

- Use the Orlov if the FLEX_BG size is 0.

- Use shift instead of divide in ext4_meta_group() as suggested by
  Andreas.

- Use sb_bgl_lock() instead of one spinlock per FLEX_BG as suggested by
  Andreas.  (Needs more perf testing)

- Remove some dead/prototype code.

Signed-off-by: Jose R. Santos <jrs@...ibm.com>
--

diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index b102b0e..7ef9787 100644
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -600,6 +600,7 @@ void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb,
 	struct ext4_sb_info *sbi;
 	int err = 0, ret;
 	ext4_grpblk_t group_freed;
+	ext4_group_t flex_group;
 
 	*pdquot_freed_blocks = 0;
 	sbi = EXT4_SB(sb);
@@ -745,6 +746,14 @@ do_more:
 	spin_unlock(sb_bgl_lock(sbi, block_group));
 	percpu_counter_add(&sbi->s_freeblocks_counter, count);
 
+	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
+	    sbi->s_groups_per_flex_shift) {
+		flex_group = ext4_flex_group(sbi, block_group);
+		spin_lock(sb_bgl_lock(sbi, flex_group));
+		sbi->s_flex_groups[flex_group].free_blocks += count;
+		spin_unlock(sb_bgl_lock(sbi, flex_group));
+	}
+
 	/* We dirtied the bitmap block */
 	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 	err = ext4_journal_dirty_metadata(handle, bitmap_bh);
@@ -1610,6 +1619,7 @@ ext4_fsblk_t ext4_new_blocks_old(handle_t *handle, struct inode *inode,
 	unsigned short windowsz = 0;
 	ext4_group_t ngroups;
 	unsigned long num = *count;
+	ext4_group_t flex_group;
 
 	*errp = -ENOSPC;
 	sb = inode->i_sb;
@@ -1815,6 +1825,14 @@ allocated:
 	spin_unlock(sb_bgl_lock(sbi, group_no));
 	percpu_counter_sub(&sbi->s_freeblocks_counter, num);
 
+	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
+	    sbi->s_groups_per_flex_shift) {
+		flex_group = ext4_flex_group(sbi, group_no);
+		spin_lock(sb_bgl_lock(sbi, flex_group));
+		sbi->s_flex_groups[flex_group].free_blocks -= num;
+		spin_unlock(sb_bgl_lock(sbi, flex_group));
+	}
+
 	BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
 	err = ext4_journal_dirty_metadata(handle, gdp_bh);
 	if (!fatal)
diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index 17b5df1..35ab8ff 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -158,6 +158,7 @@ void ext4_free_inode (handle_t *handle, struct inode * inode)
 	struct ext4_super_block * es;
 	struct ext4_sb_info *sbi;
 	int fatal = 0, err;
+	ext4_group_t flex_group;
 
 	if (atomic_read(&inode->i_count) > 1) {
 		printk ("ext4_free_inode: inode has count=%d\n",
@@ -235,6 +236,13 @@ void ext4_free_inode (handle_t *handle, struct inode * inode)
 			if (is_directory)
 				percpu_counter_dec(&sbi->s_dirs_counter);
 
+			if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
+			    sbi->s_groups_per_flex_shift) {
+				flex_group = ext4_flex_group(sbi, block_group);
+				spin_lock(sb_bgl_lock(sbi, flex_group));
+				sbi->s_flex_groups[flex_group].free_inodes++;
+				spin_unlock(sb_bgl_lock(sbi, flex_group));
+			}
 		}
 		BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
 		err = ext4_journal_dirty_metadata(handle, bh2);
@@ -289,6 +297,71 @@ static int find_group_dir(struct super_block *sb, struct inode *parent,
 	return ret;
 }
 
+#define free_block_ratio 10
+
+int find_group_flex(struct super_block *sb, struct inode *parent)
+{
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct ext4_group_desc *desc;
+	struct buffer_head *bh;
+	struct flex_groups *flex_group = sbi->s_flex_groups;
+	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
+	ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
+	ext4_group_t ngroups = sbi->s_groups_count;
+	int flex_size = ext4_flex_bg_size(sbi);
+	ext4_group_t best_flex = -1;
+	ext4_group_t best_group = -1;
+	int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
+	int flex_freeb_ratio;
+	ext4_group_t n_fbg_groups;
+	ext4_group_t i;
+
+	n_fbg_groups = (sbi->s_groups_count + flex_size - 1) / flex_size;
+	best_flex = parent_fbg_group;
+
+find_close_to_parent:
+	flex_freeb_ratio = flex_group[best_flex].free_blocks*100/blocks_per_flex;
+	if (flex_group[best_flex].free_inodes &&
+	    flex_freeb_ratio > free_block_ratio)
+		goto found_flexbg;
+
+	if (best_flex && best_flex == parent_fbg_group) {
+		best_flex--;
+		goto find_close_to_parent;
+	}
+
+	for (i = 0; i < n_fbg_groups; i++) {
+		if (i == parent_fbg_group || i == parent_fbg_group - 1)
+			continue;
+
+		flex_freeb_ratio = flex_group[i].free_blocks*100/blocks_per_flex;
+
+		if (flex_freeb_ratio > free_block_ratio &&
+		    flex_group[i].free_inodes) {
+			best_flex = i;
+			break;
+		}
+
+		if (best_flex < 0 ||
+		    (flex_group[i].free_blocks >
+		     flex_group[best_flex].free_blocks &&
+		     flex_group[i].free_inodes))
+			best_flex = i;
+	}
+
+found_flexbg:
+	for (i = best_flex * flex_size; i < ngroups &&
+		     i < (best_flex + 1) * flex_size; i++) {
+		desc = ext4_get_group_desc(sb, i, &bh);
+		if (le16_to_cpu(desc->bg_free_inodes_count)) {
+			best_group = i;
+			goto out;
+		}
+	}
+out:
+	return best_group;
+}
+
 /*
  * Orlov's allocator for directories.
  *
@@ -504,6 +577,7 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
 	struct inode *ret;
 	ext4_group_t i;
 	int free = 0;
+	ext4_group_t flex_group;
 
 	/* Cannot create files in a deleted directory */
 	if (!dir || !dir->i_nlink)
@@ -517,6 +591,13 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
 
 	sbi = EXT4_SB(sb);
 	es = sbi->s_es;
+
+	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
+		sbi->s_groups_per_flex_shift) {
+		group = find_group_flex(sb, dir);
+		goto got_group;
+	}
+
 	if (S_ISDIR(mode)) {
 		if (test_opt (sb, OLDALLOC))
 			ret2 = find_group_dir(sb, dir, &group);
@@ -525,6 +606,7 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
 	} else
 		ret2 = find_group_other(sb, dir, &group);
 
+got_group:
 	err = -ENOSPC;
 	if (ret2 == -1)
 		goto out;
@@ -681,6 +763,14 @@ got:
 		percpu_counter_inc(&sbi->s_dirs_counter);
 	sb->s_dirt = 1;
 
+	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
+	    sbi->s_groups_per_flex_shift) {
+		flex_group = ext4_flex_group(sbi, group);
+		spin_lock(sb_bgl_lock(sbi, flex_group));
+		sbi->s_flex_groups[flex_group].free_inodes--;
+		spin_unlock(sb_bgl_lock(sbi, flex_group));
+	}
+
 	inode->i_uid = current->fsuid;
 	if (test_opt (sb, GRPID))
 		inode->i_gid = dir->i_gid;
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index b626339..81ad9b1 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -518,6 +518,7 @@ static void ext4_put_super (struct super_block * sb)
 	for (i = 0; i < sbi->s_gdb_count; i++)
 		brelse(sbi->s_group_desc[i]);
 	kfree(sbi->s_group_desc);
+	kfree(sbi->s_flex_groups);
 	percpu_counter_destroy(&sbi->s_freeblocks_counter);
 	percpu_counter_destroy(&sbi->s_freeinodes_counter);
 	percpu_counter_destroy(&sbi->s_dirs_counter);
@@ -1423,6 +1424,61 @@ static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
 	return res;
 }
 
+static int ext4_fill_flex_info(struct super_block *sb)
+{
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct ext4_group_desc *gdp = NULL;
+	struct buffer_head *bh;
+	ext4_group_t flex_group_count;
+	ext4_group_t flex_group;
+	unsigned int shift;
+	int groups_per_flex = 0;
+	int tmp = 0;
+	__u64 block_bitmap = 0, cur_block_bitmap;
+	int i;
+
+	groups_per_flex = le16_to_cpu(sbi->s_es->s_flex_bg_size);
+
+	if (!groups_per_flex) {
+		sbi->s_groups_per_flex_shift = 0;
+		return 1;
+	}
+
+	shift = 0;
+	tmp = groups_per_flex;
+	while ((tmp >>= 1UL) != 0UL)
+		shift++;
+
+	sbi->s_groups_per_flex_shift = shift;
+	flex_group_count = (sbi->s_groups_count + groups_per_flex - 1) /
+		groups_per_flex;
+	sbi->s_flex_groups = kmalloc(flex_group_count *
+				     sizeof(struct flex_groups), GFP_KERNEL);
+	if (sbi->s_flex_groups == NULL) {
+		printk(KERN_ERR "EXT4-fs: not enough memory\n");
+		goto failed;
+	}
+	memset(sbi->s_flex_groups, 0, flex_group_count *
+	       sizeof(struct flex_groups));
+
+	gdp = ext4_get_group_desc(sb, 1, &bh);
+	block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
+
+	for (i = 0; i < sbi->s_groups_count; i++) {
+		gdp = ext4_get_group_desc(sb, i, &bh);
+
+		flex_group = ext4_flex_group(sbi, i);
+		sbi->s_flex_groups[flex_group].free_inodes +=
+			le16_to_cpu(gdp->bg_free_inodes_count);
+		sbi->s_flex_groups[flex_group].free_blocks +=
+			le16_to_cpu(gdp->bg_free_blocks_count);
+	}
+
+	return 1;
+failed:
+	return 0;
+}
+
 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
 			    struct ext4_group_desc *gdp)
 {
@@ -2037,6 +2093,13 @@ static int ext4_fill_super (struct super_block *sb, void *data, int silent)
 		printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
 		goto failed_mount2;
 	}
+	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
+		if (!ext4_fill_flex_info(sb)) {
+			printk(KERN_ERR
+			       "EXT4-fs: unable to initialize flex_bg meta info!\n");
+			goto failed_mount2;
+		}
+
 	sbi->s_gdb_count = db_count;
 	get_random_bytes(&sbi->s_next_generation, sizeof(u32));
 	spin_lock_init(&sbi->s_next_gen_lock);
diff --git a/include/linux/ext4_fs.h b/include/linux/ext4_fs.h
index bcdb59d..3b94dbf 100644
--- a/include/linux/ext4_fs.h
+++ b/include/linux/ext4_fs.h
@@ -152,6 +152,15 @@ struct ext4_group_desc
 	__u32	bg_reserved2[3];
 };
 
+/*
+ * Structure of a flex block group info
+ */
+
+struct flex_groups {
+	__u32 free_inodes;
+	__u32 free_blocks;
+};
+
 #define EXT4_BG_INODE_UNINIT	0x0001 /* Inode table/bitmap not in use */
 #define EXT4_BG_BLOCK_UNINIT	0x0002 /* Block bitmap not in use */
 #define EXT4_BG_INODE_ZEROED	0x0004 /* On-disk itable initialized to zero */
@@ -622,7 +631,9 @@ struct ext4_super_block {
 	__le16  s_mmp_interval;         /* # seconds to wait in MMP checking */
 	__le64  s_mmp_block;            /* Block for multi-mount protection */
 	__le32  s_raid_stripe_width;    /* blocks on all data disks (N*stride)*/
-	__u32   s_reserved[163];        /* Padding to the end of the block */
+	__le16	s_flex_bg_size;		/* FLEX_BG group size */
+	__le16	padding;		/* Padding to next 32bits */
+	__u32   s_reserved[162];        /* Padding to the end of the block */
 };
 
 #ifdef __KERNEL__
@@ -1120,6 +1131,17 @@ static inline void ext4_isize_set(struct ext4_inode *raw_inode, loff_t i_size)
 	raw_inode->i_size_high = cpu_to_le32(i_size >> 32);
 }
 
+static inline ext4_group_t ext4_flex_group(struct ext4_sb_info *sbi,
+					     ext4_group_t block_group)
+{
+	return block_group >> sbi->s_groups_per_flex_shift;
+}
+
+static inline unsigned int ext4_flex_bg_size(struct ext4_sb_info *sbi)
+{
+	return 1 << sbi->s_groups_per_flex_shift;
+}
+
 #define ext4_std_error(sb, errno)				\
 do {								\
 	if ((errno))						\
diff --git a/include/linux/ext4_fs_sb.h b/include/linux/ext4_fs_sb.h
index 744e746..ac7af1b 100644
--- a/include/linux/ext4_fs_sb.h
+++ b/include/linux/ext4_fs_sb.h
@@ -147,6 +147,9 @@ struct ext4_sb_info {
 
 	/* locality groups */
 	struct ext4_locality_group *s_locality_groups;
+
+	unsigned int s_groups_per_flex_shift;
+	struct flex_groups *s_flex_groups;
 };
 #define EXT4_GROUP_INFO(sb, group)					   \
 	EXT4_SB(sb)->s_group_info[(group) >> EXT4_DESC_PER_BLOCK_BITS(sb)] \

View attachment "e2fsprogs-flexbg-grouping.patch" of type "text/x-patch" (7927 bytes)

Powered by blists - more mailing lists