ext3-mballoc3-2.6.18.patch

非常经典的一个分布式系统

Index: linux-2.6.18.8/include/linux/ext3_fs_i.h
===================================================================
--- linux-2.6.18.8.orig/include/linux/ext3_fs_i.h	2007-07-17 09:18:14.000000000 +0200
+++ linux-2.6.18.8/include/linux/ext3_fs_i.h	2007-07-17 09:18:53.000000000 +0200
@@ -154,6 +154,10 @@ struct ext3_inode_info {
 	struct inode vfs_inode;
 
  	struct ext3_ext_cache i_cached_extent;
+
+	/* mballoc */
+	struct list_head i_prealloc_list;
+	spinlock_t i_prealloc_lock;
 };
 
 #endif	/* _LINUX_EXT3_FS_I */
Index: linux-2.6.18.8/include/linux/ext3_fs_sb.h
===================================================================
--- linux-2.6.18.8.orig/include/linux/ext3_fs_sb.h	2007-07-17 09:18:14.000000000 +0200
+++ linux-2.6.18.8/include/linux/ext3_fs_sb.h	2007-07-17 09:18:53.000000000 +0200
@@ -21,8 +21,15 @@
 #include <linux/wait.h>
 #include <linux/blockgroup_lock.h>
 #include <linux/percpu_counter.h>
+#include <linux/list.h>
 #endif
 #include <linux/rbtree.h>
+#include <linux/proc_fs.h>
+
+struct ext3_buddy_group_blocks;
+struct ext3_locality_group;
+struct ext3_mb_history;
+#define EXT3_BB_MAX_BLOCKS
 
 /*
  * third extended-fs super-block data in memory
Index: linux-2.6.18.8/include/linux/ext3_fs.h
===================================================================
--- linux-2.6.18.8.orig/include/linux/ext3_fs.h	2007-07-17 09:18:14.000000000 +0200
+++ linux-2.6.18.8/include/linux/ext3_fs.h	2007-07-17 09:18:53.000000000 +0200
@@ -17,6 +17,7 @@
 #define _LINUX_EXT3_FS_H
 
 #include <linux/types.h>
+#include <linux/ext3_fs_i.h>
 
 /*
  * The second extended filesystem constants/structures
@@ -67,12 +68,12 @@
 
 struct ext3_allocation_request {
 	struct inode *inode;	/* target inode for block we're allocating */
-	unsigned long logical;	/* logical block in target inode */
-	unsigned long goal;	/* phys. target (a hint) */
-	unsigned long lleft;	/* the closest logical allocated block to the left */
-	unsigned long pleft;	/* phys. block for ^^^ */
-	unsigned long lright;	/* the closest logical allocated block to the right */
-	unsigned long pright;	/* phys. block for ^^^ */
+	ext3_fsblk_t logical;	/* logical block in target inode */
+	ext3_fsblk_t goal;	/* phys. target (a hint) */
+	ext3_fsblk_t lleft;	/* the closest logical allocated block to the left */
+	ext3_fsblk_t pleft;	/* phys. block for ^^^ */
+	ext3_fsblk_t lright;	/* the closest logical allocated block to the right */
+	ext3_fsblk_t pright;	/* phys. block for ^^^ */
 	unsigned long len;	/* how many blocks we want to allocate */
 	unsigned long flags;	/* flags. see above EXT3_MB_HINT_* */
 };
@@ -400,6 +401,7 @@ struct ext3_inode {
 #define EXT3_MOUNT_IOPEN_NOPRIV		0x800000/* Make iopen world-readable */
 #define EXT3_MOUNT_EXTENTS		0x2000000/* Extents support */
 #define EXT3_MOUNT_EXTDEBUG             0x4000000/* Extents debug */
+#define EXT3_MOUNT_MBALLOC		0x8000000/* Buddy allocation support */
 
 /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
 #ifndef clear_opt
@@ -787,12 +789,12 @@ ext3_group_first_block_no(struct super_b
 /* balloc.c */
 extern int ext3_bg_has_super(struct super_block *sb, int group);
 extern unsigned long ext3_bg_num_gdb(struct super_block *sb, int group);
-extern ext3_fsblk_t ext3_new_block (handle_t *handle, struct inode *inode,
+extern ext3_fsblk_t ext3_new_block_old (handle_t *handle, struct inode *inode,
 			ext3_fsblk_t goal, int *errp);
-extern ext3_fsblk_t ext3_new_blocks (handle_t *handle, struct inode *inode,
+extern ext3_fsblk_t ext3_new_blocks_old (handle_t *handle, struct inode *inode,
 			ext3_fsblk_t goal, unsigned long *count, int *errp);
 extern void ext3_free_blocks (handle_t *handle, struct inode *inode,
-			ext3_fsblk_t block, unsigned long count);
+			ext3_fsblk_t block, unsigned long count, int metadata);
 extern void ext3_free_blocks_sb (handle_t *handle, struct super_block *sb,
 				 ext3_fsblk_t block, unsigned long count,
 				unsigned long *pdquot_freed_blocks);
@@ -836,15 +838,45 @@ extern long ext3_mb_stats;
 extern long ext3_mb_max_to_scan;
 extern int ext3_mb_init(struct super_block *, int);
 extern int ext3_mb_release(struct super_block *);
-extern unsigned long ext3_mb_new_blocks(handle_t *, struct ext3_allocation_request *, int *);
+extern ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
+				   ext3_fsblk_t goal, int *errp);
+extern ext3_fsblk_t ext3_mb_new_blocks(handle_t *,
+				struct ext3_allocation_request *, int *);
 extern int ext3_mb_reserve_blocks(struct super_block *, int);
 extern void ext3_mb_release_blocks(struct super_block *, int);
 extern void ext3_mb_release_blocks(struct super_block *, int);
 extern void ext3_mb_discard_inode_preallocations(struct inode *);
 extern int __init init_ext3_proc(void);
 extern void exit_ext3_proc(void);
-extern void ext3_mb_free_blocks(handle_t *, struct inode *, unsigned long, unsigned long, int, int *);
+extern void ext3_mb_free_blocks(handle_t *, struct inode *, unsigned long,
+				unsigned long, int, unsigned long *);
+
+static inline ext3_fsblk_t ext3_new_blocks(handle_t *handle,
+					   struct inode *inode,
+					   ext3_fsblk_t goal,
+					   unsigned long *count, int *errp)
+{
+	struct ext3_allocation_request ar;
+	ext3_fsblk_t ret;
 
+	if (!test_opt(inode->i_sb, MBALLOC)) {
+		ret = ext3_new_blocks_old(handle, inode, goal, count, errp);
+		return ret;
+	}
+
+	ar.inode = inode;
+	ar.goal = goal;
+	ar.len = *count;
+	ar.logical = 0;
+	ar.lleft = 0;
+	ar.pleft = 0;
+	ar.lright = 0;
+	ar.pright = 0;
+	ar.flags = 0;
+	ret = ext3_mb_new_blocks(handle, &ar, errp);
+	*count = ar.len;
+	return ret;
+}
 
 /* inode.c */
 int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
Index: linux-2.6.18.8/fs/ext3/super.c
===================================================================
--- linux-2.6.18.8.orig/fs/ext3/super.c	2007-07-17 09:18:14.000000000 +0200
+++ linux-2.6.18.8/fs/ext3/super.c	2007-07-17 09:18:53.000000000 +0200
@@ -688,6 +688,7 @@ enum {
 	Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
 	Opt_grpquota,
 	Opt_extents, Opt_noextents, Opt_extdebug,
+	Opt_mballoc, Opt_nomballoc, Opt_stripe,
 };
 
 static match_table_t tokens = {
@@ -743,6 +744,9 @@ static match_table_t tokens = {
 	{Opt_extents, "extents"},
 	{Opt_noextents, "noextents"},
 	{Opt_extdebug, "extdebug"},
+	{Opt_mballoc, "mballoc"},
+	{Opt_nomballoc, "nomballoc"},
+	{Opt_stripe, "stripe=%u"},
 	{Opt_err, NULL},
 	{Opt_resize, "resize"},
 };
@@ -1096,6 +1100,19 @@ clear_qf_name:
 		case Opt_extdebug:
 			set_opt (sbi->s_mount_opt, EXTDEBUG);
 			break;
+		case Opt_mballoc:
+			set_opt(sbi->s_mount_opt, MBALLOC);
+			break;
+		case Opt_nomballoc:
+			clear_opt(sbi->s_mount_opt, MBALLOC);
+			break;
+		case Opt_stripe:
+			if (match_int(&args[0], &option))
+				return 0;
+			if (option < 0)
+				return 0;
+			sbi->s_stripe = option;
+			break;
 		default:
 			printk (KERN_ERR
 				"EXT3-fs: Unrecognized mount option \"%s\" "
@@ -1826,6 +1843,7 @@ static int ext3_fill_super (struct super
 		"writeback");
 
 	ext3_ext_init(sb);
+	ext3_mb_init(sb, needs_recovery);
 
 	lock_kernel();
 	return 0;
Index: linux-2.6.18.8/fs/ext3/extents.c
===================================================================
--- linux-2.6.18.8.orig/fs/ext3/extents.c	2007-07-17 09:18:14.000000000 +0200
+++ linux-2.6.18.8/fs/ext3/extents.c	2007-07-17 09:18:53.000000000 +0200
@@ -795,7 +795,7 @@ cleanup:
 		for (i = 0; i < depth; i++) {
 			if (!ablocks[i])
 				continue;
-			ext3_free_blocks(handle, inode, ablocks[i], 1);
+			ext3_free_blocks(handle, inode, ablocks[i], 1, 1);
 		}
 	}
 	kfree(ablocks);
@@ -1613,7 +1613,7 @@ int ext3_ext_rm_idx(handle_t *handle, st
 	ext_debug(inode, "index is empty, remove it, free block %lu\n", leaf);
 	bh = sb_find_get_block(inode->i_sb, leaf);
 	ext3_forget(handle, 1, inode, bh, leaf);
-	ext3_free_blocks(handle, inode, leaf, 1);
+	ext3_free_blocks(handle, inode, leaf, 1, 1);
 	return err;
 }
 
@@ -1672,7 +1672,7 @@ static int ext3_remove_blocks(handle_t *
 				unsigned long from, unsigned long to)
 {
 	struct buffer_head *bh;
-	int i;
+	int i, metadata = 0;
 
 #ifdef EXTENTS_STATS
 	{
@@ -1690,6 +1690,8 @@ static int ext3_remove_blocks(handle_t *
 		spin_unlock(&sbi->s_ext_stats_lock);
 	}
 #endif
+	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
+		metadata = 1;
 	if (from >= le32_to_cpu(ex->ee_block)
 	    && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
 		/* tail removal */
@@ -1701,7 +1703,7 @@ static int ext3_remove_blocks(handle_t *
 			bh = sb_find_get_block(inode->i_sb, start + i);
 			ext3_forget(handle, 0, inode, bh, start + i);
 		}
-		ext3_free_blocks(handle, inode, start, num);
+		ext3_free_blocks(handle, inode, start, num, metadata);
 	} else if (from == le32_to_cpu(ex->ee_block)
 		   && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
 		printk("strange request: removal %lu-%lu from %u:%u\n",
@@ -2034,7 +2036,7 @@ int ext3_ext_get_blocks(handle_t *handle
 	struct ext3_extent newex, *ex;
 	int goal, newblock, err = 0, depth;
 	unsigned long allocated = 0;
-	unsigned long next;
+	struct ext3_allocation_request ar;
 
 	__clear_bit(BH_New, &bh_result->b_state);
 	ext_debug(inode, "blocks %d/%lu requested for inode %u\n", (int) iblock,
@@ -2116,18 +2118,33 @@ int ext3_ext_get_blocks(handle_t *handle
 	if (S_ISREG(inode->i_mode) && (!EXT3_I(inode)->i_block_alloc_info))
 		ext3_init_block_alloc_info(inode);
 
+	/* find neighbour allocated blocks */
+	ar.lleft = iblock;
+	err = ext3_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
+	if (err)
+		goto out2;
+	ar.lright = iblock;
+	err = ext3_ext_search_right(inode, path, &ar.lright, &ar.pright);
+	if (err)
+		goto out2;
+
 	/* find next allocated block so that we know how many
 	 * blocks we can allocate without ovelapping next extent */
-	BUG_ON(iblock < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
-	next = ext3_ext_next_allocated_block(path);
-	BUG_ON(next <= iblock);
-	allocated = next - iblock;
+	BUG_ON(ar.pright != 0 && ar.lright <= iblock);
+	if (ar.pright == 0)
+		allocated = EXT_MAX_BLOCK - iblock;
+	else
+		allocated = ar.lright - iblock;
 	if (allocated > max_blocks)
 		allocated = max_blocks;
 
 	/* allocate new block */
-	goal = ext3_ext_find_goal(inode, path, iblock);
-	newblock = ext3_new_blocks(handle, inode, goal, &allocated, &err);
+	ar.inode = inode;
+	ar.goal = ext3_ext_find_goal(inode, path, iblock);
+	ar.logical = iblock;
+	ar.len = allocated;
+	ar.flags = EXT3_MB_HINT_DATA;
+	newblock = ext3_mb_new_blocks(handle, &ar, &err);
 	if (!newblock)
 		goto out2;
 	ext_debug(inode, "allocate new block: goal %d, found %d/%lu\n",
@@ -2137,12 +2154,16 @@ int ext3_ext_get_blocks(handle_t *handle
 	newex.ee_block = cpu_to_le32(iblock);
 	newex.ee_start = cpu_to_le32(newblock);
 	newex.ee_start_hi = 0;
-	newex.ee_len = cpu_to_le16(allocated);
+	newex.ee_len = cpu_to_le16(ar.len);
 	err = ext3_ext_insert_extent(handle, inode, path, &newex);
 	if (err) {
 		/* free data blocks we just allocated */
-		ext3_free_blocks(handle, inode, le32_to_cpu(newex.ee_start),
-				le16_to_cpu(newex.ee_len));
+		/* not a good idea to call discard here directly,
+		 * but otherwise we'd need to call it every free() */
+		ext3_mb_discard_inode_preallocations(inode);
+		ext3_free_blocks(handle, inode, newex.ee_start,
+				 newex.ee_len, 0);
+