Ocfs2 updates for 2.6.28

index ec2ed15..60232b1 100644
@@ -2236,6 +2236,10 @@ const struct inode_operations ocfs2_special_file_iops = {
 	.permission	= ocfs2_permission,
 };
 
+/*
+ * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
+ * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
+ */
 const struct file_operations ocfs2_fops = {
 	.llseek		= generic_file_llseek,
 	.read		= do_sync_read,

@@ -2250,6 +2254,7 @@ const struct file_operations ocfs2_fops = {
 #ifdef CONFIG_COMPAT
 	.compat_ioctl   = ocfs2_compat_ioctl,
 #endif
+	.lock		= ocfs2_lock,
 	.flock		= ocfs2_flock,
 	.splice_read	= ocfs2_file_splice_read,
 	.splice_write	= ocfs2_file_splice_write,

@@ -2266,5 +2271,51 @@ const struct file_operations ocfs2_dops = {
 #ifdef CONFIG_COMPAT
 	.compat_ioctl   = ocfs2_compat_ioctl,
 #endif
+	.lock		= ocfs2_lock,
+	.flock		= ocfs2_flock,
+};
+
+/*
+ * POSIX-lockless variants of our file_operations.
+ *
+ * These will be used if the underlying cluster stack does not support
+ * posix file locking, if the user passes the "localflocks" mount
+ * option, or if we have a local-only fs.
+ *
+ * ocfs2_flock is in here because all stacks handle UNIX file locks,
+ * so we still want it in the case of no stack support for
+ * plocks. Internally, it will do the right thing when asked to ignore
+ * the cluster.
+ */
+const struct file_operations ocfs2_fops_no_plocks = {
+	.llseek		= generic_file_llseek,
+	.read		= do_sync_read,
+	.write		= do_sync_write,
+	.mmap		= ocfs2_mmap,
+	.fsync		= ocfs2_sync_file,
+	.release	= ocfs2_file_release,
+	.open		= ocfs2_file_open,
+	.aio_read	= ocfs2_file_aio_read,
+	.aio_write	= ocfs2_file_aio_write,
+	.unlocked_ioctl	= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
+	.flock		= ocfs2_flock,
+	.splice_read	= ocfs2_file_splice_read,
+	.splice_write	= ocfs2_file_splice_write,
+};
+
+const struct file_operations ocfs2_dops_no_plocks = {
+	.llseek		= generic_file_llseek,
+	.read		= generic_read_dir,
+	.readdir	= ocfs2_readdir,
+	.fsync		= ocfs2_sync_file,
+	.release	= ocfs2_dir_release,
+	.open		= ocfs2_dir_open,
+	.unlocked_ioctl	= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
 	.flock		= ocfs2_flock,
 };

index 1e27b4d..5a6d3e4 100644
@@ -28,6 +28,8 @@
 
 extern const struct file_operations ocfs2_fops;
 extern const struct file_operations ocfs2_dops;
+extern const struct file_operations ocfs2_fops_no_plocks;
+extern const struct file_operations ocfs2_dops_no_plocks;
 extern const struct inode_operations ocfs2_file_iops;
 extern const struct inode_operations ocfs2_special_file_iops;
 struct ocfs2_alloc_context;

index 7e9e4c7..99f012a 100644
@@ -219,6 +219,7 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 	struct super_block *sb;
 	struct ocfs2_super *osb;
 	int status = -EINVAL;
+	int use_plocks = 1;
 
 	mlog_entry("(0x%p, size:%llu)\n", inode,
 		   (unsigned long long)le64_to_cpu(fe->i_size));

@@ -226,6 +227,10 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 	sb = inode->i_sb;
 	osb = OCFS2_SB(sb);
 
+	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
+	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
+		use_plocks = 0;
+
 	/* this means that read_inode cannot create a superblock inode
 	 * today.  change if needed. */
 	if (!OCFS2_IS_VALID_DINODE(fe) ||

@@ -295,13 +300,19 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 
 	switch (inode->i_mode & S_IFMT) {
 	    case S_IFREG:
-		    inode->i_fop = &ocfs2_fops;
+		    if (use_plocks)
+			    inode->i_fop = &ocfs2_fops;
+		    else
+			    inode->i_fop = &ocfs2_fops_no_plocks;
 		    inode->i_op = &ocfs2_file_iops;
 		    i_size_write(inode, le64_to_cpu(fe->i_size));
 		    break;
 	    case S_IFDIR:
 		    inode->i_op = &ocfs2_dir_iops;
-		    inode->i_fop = &ocfs2_dops;
+		    if (use_plocks)
+			    inode->i_fop = &ocfs2_dops;
+		    else
+			    inode->i_fop = &ocfs2_dops_no_plocks;
 		    i_size_write(inode, le64_to_cpu(fe->i_size));
 		    break;
 	    case S_IFLNK:

index 203f871..544ac62 100644
@@ -24,6 +24,7 @@
  */
 
 #include <linux/fs.h>
+#include <linux/fcntl.h>
 
 #define MLOG_MASK_PREFIX ML_INODE
 #include <cluster/masklog.h>

@@ -32,6 +33,7 @@
 
 #include "dlmglue.h"
 #include "file.h"
+#include "inode.h"
 #include "locks.h"
 
 static int ocfs2_do_flock(struct file *file, struct inode *inode,

@@ -123,3 +125,16 @@ int ocfs2_flock(struct file *file, int cmd, struct file_lock *fl)
 	else
 		return ocfs2_do_flock(file, inode, cmd, fl);
 }
+
+int ocfs2_lock(struct file *file, int cmd, struct file_lock *fl)
+{
+	struct inode *inode = file->f_mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (!(fl->fl_flags & FL_POSIX))
+		return -ENOLCK;
+	if (__mandatory_lock(inode))
+		return -ENOLCK;
+
+	return ocfs2_plock(osb->cconn, OCFS2_I(inode)->ip_blkno, file, cmd, fl);
+}

index 9743ef2..496d488 100644
@@ -27,5 +27,6 @@
 #define OCFS2_LOCKS_H
 
 int ocfs2_flock(struct file *file, int cmd, struct file_lock *fl);
+int ocfs2_lock(struct file *file, int cmd, struct file_lock *fl);
 
 #endif /* OCFS2_LOCKS_H */

index 353fc35..faec2d8 100644
@@ -28,6 +28,7 @@
 #include "ocfs2.h"  /* For struct ocfs2_lock_res */
 #include "stackglue.h"
 
+#include <linux/dlm_plock.h>
 
 /*
  * The control protocol starts with a handshake.  Until the handshake

@@ -746,6 +747,37 @@ static void user_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
 {
 }
 
+static int user_plock(struct ocfs2_cluster_connection *conn,
+		      u64 ino,
+		      struct file *file,
+		      int cmd,
+		      struct file_lock *fl)
+{
+	/*
+	 * This more or less just demuxes the plock request into any
+	 * one of three dlm calls.
+	 *
+	 * Internally, fs/dlm will pass these to a misc device, which
+	 * a userspace daemon will read and write to.
+	 *
+	 * For now, cancel requests (which happen internally only),
+	 * are turned into unlocks. Most of this function taken from
+	 * gfs2_lock.
+	 */
+
+	if (cmd == F_CANCELLK) {
+		cmd = F_SETLK;
+		fl->fl_type = F_UNLCK;
+	}
+
+	if (IS_GETLK(cmd))
+		return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
+	else if (fl->fl_type == F_UNLCK)
+		return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
+	else
+		return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
+}
+
 /*
  * Compare a requested locking protocol version against the current one.
  *

@@ -839,6 +871,7 @@ static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
 	.dlm_unlock	= user_dlm_unlock,
 	.lock_status	= user_dlm_lock_status,
 	.lock_lvb	= user_dlm_lvb,
+	.plock		= user_plock,
 	.dump_lksb	= user_dlm_dump_lksb,
 };
 

index 07f348b..7150f5d 100644
@@ -288,6 +288,26 @@ void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
 }
 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
 
+int ocfs2_stack_supports_plocks(void)
+{
+	return !!(active_stack && active_stack->sp_ops->plock);
+}
+EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
+
+/*
+ * ocfs2_plock() can only be safely called if
+ * ocfs2_stack_supports_plocks() returned true
+ */
+int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
+		struct file *file, int cmd, struct file_lock *fl)
+{
+	WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
+	if (active_stack->sp_ops->plock)
+		return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
+	return -EOPNOTSUPP;
+}
+EXPORT_SYMBOL_GPL(ocfs2_plock);
+
 int ocfs2_cluster_connect(const char *stack_name,
 			  const char *group,
 			  int grouplen,

index db56281..c571af3 100644
@@ -28,6 +28,10 @@
 #include "dlm/dlmapi.h"
 #include <linux/dlm.h>
 
+/* Needed for plock-related prototypes */
+struct file;
+struct file_lock;
+
 /*
  * dlmconstants.h does not have a LOCAL flag.  We hope to remove it
  * some day, but right now we need it.  Let's fake it.  This value is larger

@@ -187,6 +191,17 @@ struct ocfs2_stack_operations {
 	void *(*lock_lvb)(union ocfs2_dlm_lksb *lksb);
 
 	/*
+	 * Cluster-aware posix locks
+	 *
+	 * This is NULL for stacks which do not support posix locks.
+	 */
+	int (*plock)(struct ocfs2_cluster_connection *conn,
+		     u64 ino,
+		     struct file *file,
+		     int cmd,
+		     struct file_lock *fl);
+
+	/*
 	 * This is an optoinal debugging hook.  If provided, the
 	 * stack can dump debugging information about this lock.
 	 */

@@ -240,6 +255,10 @@ int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb);
 void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb);
 void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb);
 
+int ocfs2_stack_supports_plocks(void);
+int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
+		struct file *file, int cmd, struct file_lock *fl);
+
 void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto);
 
 

index 7150f5d..68b668b 100644
@@ -290,7 +290,7 @@ EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
 
 int ocfs2_stack_supports_plocks(void)
 {
-	return !!(active_stack && active_stack->sp_ops->plock);
+	return active_stack && active_stack->sp_ops->plock;
 }
 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
 

index 28e492e..b05ce66 100644
@@ -47,8 +47,6 @@
 
 #define OCFS2_LOCAL_ALLOC(dinode)	(&((dinode)->id2.i_lab))
 
-static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb);
-
 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
 
 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,

@@ -75,21 +73,13 @@ static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 					  struct inode *local_alloc_inode);
 
-static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb)
-{
-	BUG_ON(osb->s_clustersize_bits > 20);
-
-	/* Size local alloc windows by the megabyte */
-	return osb->local_alloc_size << (20 - osb->s_clustersize_bits);
-}
-
 /*
  * Tell us whether a given allocation should use the local alloc
  * file. Otherwise, it has to go to the main bitmap.
  */
 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
 {
-	int la_bits = ocfs2_local_alloc_window_bits(osb);
+	int la_bits = osb->local_alloc_bits;
 	int ret = 0;
 
 	if (osb->local_alloc_state != OCFS2_LA_ENABLED)

@@ -120,14 +110,16 @@ int ocfs2_load_local_alloc(struct ocfs2_super *osb)
 
 	mlog_entry_void();
 
-	if (osb->local_alloc_size == 0)
+	if (osb->local_alloc_bits == 0)
 		goto bail;
 
-	if (ocfs2_local_alloc_window_bits(osb) >= osb->bitmap_cpg) {
+	if (osb->local_alloc_bits >= osb->bitmap_cpg) {
 		mlog(ML_NOTICE, "Requested local alloc window %d is larger "
 		     "than max possible %u. Using defaults.\n",
-		     ocfs2_local_alloc_window_bits(osb), (osb->bitmap_cpg - 1));
-		osb->local_alloc_size = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
+		     osb->local_alloc_bits, (osb->bitmap_cpg - 1));
+		osb->local_alloc_bits =
+			ocfs2_megabytes_to_clusters(osb->sb,
+						    OCFS2_DEFAULT_LOCAL_ALLOC_SIZE);
 	}
 
 	/* read the alloc off disk */

@@ -190,8 +182,7 @@ bail:
 	if (inode)
 		iput(inode);
 
-	mlog(0, "Local alloc window bits = %d\n",
-	     ocfs2_local_alloc_window_bits(osb));
+	mlog(0, "Local alloc window bits = %d\n", osb->local_alloc_bits);
 
 	mlog_exit(status);
 	return status;

@@ -490,7 +481,7 @@ int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	if (bits_wanted > ocfs2_local_alloc_window_bits(osb)) {
+	if (bits_wanted > osb->local_alloc_bits) {
 		mlog(0, "Asking for more than my max window size!\n");
 		status = -ENOSPC;
 		goto bail;

@@ -803,7 +794,7 @@ static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	(*ac)->ac_bits_wanted = ocfs2_local_alloc_window_bits(osb);
+	(*ac)->ac_bits_wanted = osb->local_alloc_bits;
 
 	status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
 	if (status < 0) {

@@ -849,7 +840,7 @@ static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
 		     "one\n");
 
 	mlog(0, "Allocating %u clusters for a new window.\n",
-	     ocfs2_local_alloc_window_bits(osb));
+	     osb->local_alloc_bits);
 
 	/* Instruct the allocation code to try the most recently used
 	 * cluster group. We'll re-record the group used this pass

@@ -859,8 +850,7 @@ static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
 	/* we used the generic suballoc reserve function, but we set
 	 * everything up nicely, so there's no reason why we can't use
 	 * the more specific cluster api to claim bits. */
-	status = ocfs2_claim_clusters(osb, handle, ac,
-				      ocfs2_local_alloc_window_bits(osb),
+	status = ocfs2_claim_clusters(osb, handle, ac, osb->local_alloc_bits,
 				      &cluster_off, &cluster_count);
 	if (status < 0) {
 		if (status != -ENOSPC)

index 7f625f2..43dd42e 100644
@@ -252,7 +252,7 @@ struct ocfs2_super
 	struct ocfs2_journal *journal;
 	unsigned long osb_commit_interval;
 
-	int local_alloc_size;
+	unsigned int local_alloc_bits;
 	enum ocfs2_local_alloc_state local_alloc_state;
 	struct buffer_head *local_alloc_bh;
 	u64 la_last_gd;

@@ -554,6 +554,14 @@ static inline unsigned int ocfs2_pages_per_cluster(struct super_block *sb)
 	return pages_per_cluster;
 }
 
+static inline unsigned int ocfs2_megabytes_to_clusters(struct super_block *sb,
+						       unsigned int megs)
+{
+	BUILD_BUG_ON(OCFS2_MAX_CLUSTERSIZE > 1048576);
+
+	return megs << (20 - OCFS2_SB(sb)->s_clustersize_bits);
+}
+
 static inline void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
 {
 	spin_lock(&osb->osb_lock);

index 88255d3..c7bb4b8 100644
@@ -637,7 +637,7 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 	osb->s_atime_quantum = parsed_options.atime_quantum;
 	osb->preferred_slot = parsed_options.slot;
 	osb->osb_commit_interval = parsed_options.commit_interval;
-	osb->local_alloc_size = parsed_options.localalloc_opt;
+	osb->local_alloc_bits = ocfs2_megabytes_to_clusters(sb, parsed_options.localalloc_opt);
 
 	status = ocfs2_verify_userspace_stack(osb, &parsed_options);
 	if (status)

@@ -938,6 +938,7 @@ static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
 {
 	struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
 	unsigned long opts = osb->s_mount_opt;
+	unsigned int local_alloc_megs;
 
 	if (opts & OCFS2_MOUNT_HB_LOCAL)
 		seq_printf(s, ",_netdev,heartbeat=local");

@@ -970,8 +971,9 @@ static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
 		seq_printf(s, ",commit=%u",
 			   (unsigned) (osb->osb_commit_interval / HZ));
 
-	if (osb->local_alloc_size != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
-		seq_printf(s, ",localalloc=%d", osb->local_alloc_size);
+	local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
+	if (local_alloc_megs != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
+		seq_printf(s, ",localalloc=%d", local_alloc_megs);
 
 	if (opts & OCFS2_MOUNT_LOCALFLOCKS)
 		seq_printf(s, ",localflocks,");

index b05ce66..f71658a 100644
@@ -73,16 +73,51 @@ static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 					  struct inode *local_alloc_inode);
 
+static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb)
+{
+	return (osb->local_alloc_state == OCFS2_LA_THROTTLED ||
+		osb->local_alloc_state == OCFS2_LA_ENABLED);
+}
+
+void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
+				      unsigned int num_clusters)
+{
+	spin_lock(&osb->osb_lock);
+	if (osb->local_alloc_state == OCFS2_LA_DISABLED ||
+	    osb->local_alloc_state == OCFS2_LA_THROTTLED)
+		if (num_clusters >= osb->local_alloc_default_bits) {
+			cancel_delayed_work(&osb->la_enable_wq);
+			osb->local_alloc_state = OCFS2_LA_ENABLED;
+		}
+	spin_unlock(&osb->osb_lock);
+}
+
+void ocfs2_la_enable_worker(struct work_struct *work)
+{
+	struct ocfs2_super *osb =
+		container_of(work, struct ocfs2_super,
+			     la_enable_wq.work);
+	spin_lock(&osb->osb_lock);
+	osb->local_alloc_state = OCFS2_LA_ENABLED;
+	spin_unlock(&osb->osb_lock);
+}
+
 /*
  * Tell us whether a given allocation should use the local alloc
  * file. Otherwise, it has to go to the main bitmap.
+ *
+ * This function does semi-dirty reads of local alloc size and state!
+ * This is ok however, as the values are re-checked once under mutex.
  */
 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
 {
-	int la_bits = osb->local_alloc_bits;
 	int ret = 0;
+	int la_bits;
+
+	spin_lock(&osb->osb_lock);
+	la_bits = osb->local_alloc_bits;
 
-	if (osb->local_alloc_state != OCFS2_LA_ENABLED)
+	if (!ocfs2_la_state_enabled(osb))
 		goto bail;
 
 	/* la_bits should be at least twice the size (in clusters) of

@@ -96,6 +131,7 @@ int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
 bail:
 	mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n",
 	     osb->local_alloc_state, (unsigned long long)bits, la_bits, ret);
+	spin_unlock(&osb->osb_lock);
 	return ret;
 }
 

@@ -208,6 +244,9 @@ void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
 
 	mlog_entry_void();
 
+	cancel_delayed_work(&osb->la_enable_wq);
+	flush_workqueue(ocfs2_wq);
+
 	if (osb->local_alloc_state == OCFS2_LA_UNUSED)
 		goto out;
 

@@ -445,7 +484,7 @@ out:
 }
 
 /*
- * make sure we've got at least bitswanted contiguous bits in the
+ * make sure we've got at least bits_wanted contiguous bits in the
  * local alloc. You lose them when you drop i_mutex.
  *
  * We will add ourselves to the transaction passed in, but may start

@@ -476,16 +515,18 @@ int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
 
 	mutex_lock(&local_alloc_inode->i_mutex);
 
-	if (osb->local_alloc_state != OCFS2_LA_ENABLED) {
-		status = -ENOSPC;
-		goto bail;
-	}
-
-	if (bits_wanted > osb->local_alloc_bits) {
-		mlog(0, "Asking for more than my max window size!\n");
+	/*
+	 * We must double check state and allocator bits because
+	 * another process may have changed them while holding i_mutex.
+	 */
+	spin_lock(&osb->osb_lock);
+	if (!ocfs2_la_state_enabled(osb) ||
+	    (bits_wanted > osb->local_alloc_bits)) {
+		spin_unlock(&osb->osb_lock);
 		status = -ENOSPC;
 		goto bail;
 	}
+	spin_unlock(&osb->osb_lock);
 
 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
 

@@ -513,6 +554,21 @@ int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
 				mlog_errno(status);
 			goto bail;
 		}
+
+		/*
+		 * Under certain conditions, the window slide code
+		 * might have reduced the number of bits available or
+		 * disabled the the local alloc entirely. Re-check
+		 * here and return -ENOSPC if necessary.
+		 */
+		status = -ENOSPC;
+		if (!ocfs2_la_state_enabled(osb))
+			goto bail;
+
+		free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
+			le32_to_cpu(alloc->id1.bitmap1.i_used);
+		if (bits_wanted > free_bits)
+			goto bail;
 	}
 
 	ac->ac_inode = local_alloc_inode;

@@ -780,6 +836,85 @@ bail:
 	return status;
 }
 
+enum ocfs2_la_event {
+	OCFS2_LA_EVENT_SLIDE,		/* Normal window slide. */
+	OCFS2_LA_EVENT_FRAGMENTED,	/* The global bitmap has
+					 * enough bits theoretically
+					 * free, but a contiguous
+					 * allocation could not be
+					 * found. */
+	OCFS2_LA_EVENT_ENOSPC,		/* Global bitmap doesn't have
+					 * enough bits free to satisfy
+					 * our request. */
+};
+#define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
+/*
+ * Given an event, calculate the size of our next local alloc window.
+ *
+ * This should always be called under i_mutex of the local alloc inode
+ * so that local alloc disabling doesn't race with processes trying to
+ * use the allocator.
+ *
+ * Returns the state which the local alloc was left in. This value can
+ * be ignored by some paths.
+ */
+static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
+				  enum ocfs2_la_event event)
+{
+	unsigned int bits;
+	int state;
+
+	spin_lock(&osb->osb_lock);
+	if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
+		WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
+		goto out_unlock;
+	}
+
+	/*
+	 * ENOSPC and fragmentation are treated similarly for now.
+	 */
+	if (event == OCFS2_LA_EVENT_ENOSPC ||
+	    event == OCFS2_LA_EVENT_FRAGMENTED) {
+		/*
+		 * We ran out of contiguous space in the primary
+		 * bitmap. Drastically reduce the number of bits used
+		 * by local alloc until we have to disable it.
+		 */
+		bits = osb->local_alloc_bits >> 1;
+		if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
+			/*
+			 * By setting state to THROTTLED, we'll keep
+			 * the number of local alloc bits used down
+			 * until an event occurs which would give us
+			 * reason to assume the bitmap situation might
+			 * have changed.
+			 */
+			osb->local_alloc_state = OCFS2_LA_THROTTLED;
+			osb->local_alloc_bits = bits;
+		} else {
+			osb->local_alloc_state = OCFS2_LA_DISABLED;
+		}
+		queue_delayed_work(ocfs2_wq, &osb->la_enable_wq,
+				   OCFS2_LA_ENABLE_INTERVAL);
+		goto out_unlock;
+	}
+
+	/*
+	 * Don't increase the size of the local alloc window until we
+	 * know we might be able to fulfill the request. Otherwise, we
+	 * risk bouncing around the global bitmap during periods of
+	 * low space.
+	 */
+	if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
+		osb->local_alloc_bits = osb->local_alloc_default_bits;
+
+out_unlock:
+	state = osb->local_alloc_state;
+	spin_unlock(&osb->osb_lock);
+
+	return state;
+}
+
 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
 						struct ocfs2_alloc_context **ac,
 						struct inode **bitmap_inode,

@@ -794,12 +929,21 @@ static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
 		goto bail;
 	}
 
+retry_enospc:
 	(*ac)->ac_bits_wanted = osb->local_alloc_bits;
 
 	status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
+	if (status == -ENOSPC) {
+		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
+		    OCFS2_LA_DISABLED)
+			goto bail;
+
+		ocfs2_free_ac_resource(*ac);
+		memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
+		goto retry_enospc;
+	}
 	if (status < 0) {
-		if (status != -ENOSPC)
-			mlog_errno(status);
+		mlog_errno(status);
 		goto bail;
 	}
 

@@ -852,6 +996,34 @@ static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
 	 * the more specific cluster api to claim bits. */
 	status = ocfs2_claim_clusters(osb, handle, ac, osb->local_alloc_bits,
 				      &cluster_off, &cluster_count);
+	if (status == -ENOSPC) {
+retry_enospc:
+		/*
+		 * Note: We could also try syncing the journal here to
+		 * allow use of any free bits which the current
+		 * transaction can't give us access to. --Mark
+		 */
+		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
+		    OCFS2_LA_DISABLED)
+			goto bail;
+
+		status = ocfs2_claim_clusters(osb, handle, ac,
+					      osb->local_alloc_bits,
+					      &cluster_off,
+					      &cluster_count);
+		if (status == -ENOSPC)
+			goto retry_enospc;
+		/*
+		 * We only shrunk the *minimum* number of in our
+		 * request - it's entirely possible that the allocator
+		 * might give us more than we asked for.
+		 */
+		if (status == 0) {
+			spin_lock(&osb->osb_lock);
+			osb->local_alloc_bits = cluster_count;
+			spin_unlock(&osb->osb_lock);
+		}
+	}
 	if (status < 0) {
 		if (status != -ENOSPC)
 			mlog_errno(status);

@@ -895,6 +1067,8 @@ static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 
 	mlog_entry_void();
 
+	ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
+
 	/* This will lock the main bitmap for us. */
 	status = ocfs2_local_alloc_reserve_for_window(osb,
 						      &ac,

index 3f76631..ac5ea9f 100644
@@ -52,4 +52,8 @@ int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
 				 u32 *bit_off,
 				 u32 *num_bits);
 
+void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
+				      unsigned int num_clusters);
+void ocfs2_la_enable_worker(struct work_struct *work);
+
 #endif /* OCFS2_LOCALALLOC_H */

index 43dd42e..4d6e200 100644
@@ -171,9 +171,13 @@ struct ocfs2_alloc_stats
 
 enum ocfs2_local_alloc_state
 {
-	OCFS2_LA_UNUSED = 0,
-	OCFS2_LA_ENABLED,
-	OCFS2_LA_DISABLED
+	OCFS2_LA_UNUSED = 0,	/* Local alloc will never be used for
+				 * this mountpoint. */
+	OCFS2_LA_ENABLED,	/* Local alloc is in use. */
+	OCFS2_LA_THROTTLED,	/* Local alloc is in use, but number
+				 * of bits has been reduced. */
+	OCFS2_LA_DISABLED	/* Local alloc has temporarily been
+				 * disabled. */
 };
 
 enum ocfs2_mount_options

@@ -252,9 +256,20 @@ struct ocfs2_super
 	struct ocfs2_journal *journal;
 	unsigned long osb_commit_interval;
 
+	struct delayed_work		la_enable_wq;
+
+	/*
+	 * Must hold local alloc i_mutex and osb->osb_lock to change
+	 * local_alloc_bits. Reads can be done under either lock.
+	 */
 	unsigned int local_alloc_bits;
-	enum ocfs2_local_alloc_state local_alloc_state;
+	unsigned int local_alloc_default_bits;
+
+	enum ocfs2_local_alloc_state local_alloc_state; /* protected
+							 * by osb_lock */
+
 	struct buffer_head *local_alloc_bh;
+
 	u64 la_last_gd;
 
 	/* Next two fields are for local node slot recovery during

index d2d278f..de7b93d 100644
@@ -111,7 +111,7 @@ static inline void ocfs2_block_to_cluster_group(struct inode *inode,
 						u64 *bg_blkno,
 						u16 *bg_bit_off);
 
-static void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
+void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
 {
 	struct inode *inode = ac->ac_inode;
 

@@ -686,15 +686,6 @@ int ocfs2_reserve_clusters(struct ocfs2_super *osb,
 		if ((status < 0) && (status != -ENOSPC)) {
 			mlog_errno(status);
 			goto bail;
-		} else if (status == -ENOSPC) {
-			/* reserve_local_bits will return enospc with
-			 * the local alloc inode still locked, so we
-			 * can change this safely here. */
-			mlog(0, "Disabling local alloc\n");
-			/* We set to OCFS2_LA_DISABLED so that umount
-			 * can clean up what's left of the local
-			 * allocation */
-			osb->local_alloc_state = OCFS2_LA_DISABLED;
 		}
 	}
 

@@ -1005,6 +996,7 @@ static int ocfs2_cluster_group_search(struct inode *inode,
 	int search = -ENOSPC;
 	int ret;
 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 	u16 tmp_off, tmp_found;
 	unsigned int max_bits, gd_cluster_off;
 

@@ -1045,6 +1037,12 @@ static int ocfs2_cluster_group_search(struct inode *inode,
 			*bit_off = tmp_off;
 			*bits_found = tmp_found;
 			search = 0; /* success */
+		} else if (tmp_found) {
+			/*
+			 * Don't show bits which we'll be returning
+			 * for allocation to the local alloc bitmap.
+			 */
+			ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
 		}
 	}
 

@@ -1203,9 +1201,8 @@ static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
 	status = -ENOSPC;
 	/* for now, the chain search is a bit simplistic. We just use
 	 * the 1st group with any empty bits. */
-	while ((status = ac->ac_group_search(alloc_inode, group_bh,
-					     bits_wanted, min_bits, bit_off,
-					     &tmp_bits)) == -ENOSPC) {
+	while ((status = ac->ac_group_search(alloc_inode, group_bh, bits_wanted,
+					     min_bits, bit_off, &tmp_bits)) == -ENOSPC) {
 		if (!bg->bg_next_group)
 			break;
 

@@ -1838,9 +1835,15 @@ int ocfs2_free_clusters(handle_t *handle,
 	status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
 					  bg_start_bit, bg_blkno,
 					  num_clusters);
-	if (status < 0)
+	if (status < 0) {
 		mlog_errno(status);
+		goto out;
+	}
+
+	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
+					 num_clusters);
 
+out:
 	mlog_exit(status);
 	return status;
 }

index 544c600..40d51da 100644
@@ -147,6 +147,7 @@ static inline int ocfs2_is_cluster_bitmap(struct inode *inode)
  * apis above. */
 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
 				      struct ocfs2_alloc_context *ac);
+void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac);
 
 /* given a cluster offset, calculate which block group it belongs to
  * and return that block offset. */

index c7bb4b8..e58a525 100644
@@ -637,7 +637,8 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 	osb->s_atime_quantum = parsed_options.atime_quantum;
 	osb->preferred_slot = parsed_options.slot;
 	osb->osb_commit_interval = parsed_options.commit_interval;
-	osb->local_alloc_bits = ocfs2_megabytes_to_clusters(sb, parsed_options.localalloc_opt);
+	osb->local_alloc_default_bits = ocfs2_megabytes_to_clusters(sb, parsed_options.localalloc_opt);
+	osb->local_alloc_bits = osb->local_alloc_default_bits;
 
 	status = ocfs2_verify_userspace_stack(osb, &parsed_options);
 	if (status)

@@ -1425,6 +1426,7 @@ static int ocfs2_initialize_super(struct super_block *sb,
 
 	osb->local_alloc_state = OCFS2_LA_UNUSED;
 	osb->local_alloc_bh = NULL;
+	INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
 
 	init_waitqueue_head(&osb->osb_mount_event);
 

index f71658a..b889f10 100644
@@ -28,6 +28,7 @@
 #include <linux/slab.h>
 #include <linux/highmem.h>
 #include <linux/bitops.h>
+#include <linux/debugfs.h>
 
 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
 #include <cluster/masklog.h>

@@ -73,6 +74,85 @@ static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 					  struct inode *local_alloc_inode);
 
+#ifdef CONFIG_OCFS2_FS_STATS
+
+DEFINE_MUTEX(la_debug_mutex);
+
+static int ocfs2_la_debug_open(struct inode *inode, struct file *file)
+{
+	file->private_data = inode->i_private;
+	return 0;
+}
+
+#define LA_DEBUG_BUF_SZ	PAGE_CACHE_SIZE
+#define LA_DEBUG_VER	1
+static ssize_t ocfs2_la_debug_read(struct file *file, char __user *userbuf,
+				   size_t count, loff_t *ppos)
+{
+	struct ocfs2_super *osb = file->private_data;
+	int written, ret;
+	char *buf = osb->local_alloc_debug_buf;
+
+	mutex_lock(&la_debug_mutex);
+	memset(buf, 0, LA_DEBUG_BUF_SZ);
+
+	written = snprintf(buf, LA_DEBUG_BUF_SZ,
+			   "0x%x\t0x%llx\t%u\t%u\t0x%x\n",
+			   LA_DEBUG_VER,
+			   (unsigned long long)osb->la_last_gd,
+			   osb->local_alloc_default_bits,
+			   osb->local_alloc_bits, osb->local_alloc_state);
+
+	ret = simple_read_from_buffer(userbuf, count, ppos, buf, written);
+
+	mutex_unlock(&la_debug_mutex);
+	return ret;
+}
+
+static const struct file_operations ocfs2_la_debug_fops = {
+	.open =		ocfs2_la_debug_open,
+	.read =		ocfs2_la_debug_read,
+};
+
+static void ocfs2_init_la_debug(struct ocfs2_super *osb)
+{
+	osb->local_alloc_debug_buf = kmalloc(LA_DEBUG_BUF_SZ, GFP_NOFS);
+	if (!osb->local_alloc_debug_buf)
+		return;
+
+	osb->local_alloc_debug = debugfs_create_file("local_alloc_stats",
+						     S_IFREG|S_IRUSR,
+						     osb->osb_debug_root,
+						     osb,
+						     &ocfs2_la_debug_fops);
+	if (!osb->local_alloc_debug) {
+		kfree(osb->local_alloc_debug_buf);
+		osb->local_alloc_debug_buf = NULL;
+	}
+}
+
+static void ocfs2_shutdown_la_debug(struct ocfs2_super *osb)
+{
+	if (osb->local_alloc_debug)
+		debugfs_remove(osb->local_alloc_debug);
+
+	if (osb->local_alloc_debug_buf)
+		kfree(osb->local_alloc_debug_buf);
+
+	osb->local_alloc_debug_buf = NULL;
+	osb->local_alloc_debug = NULL;
+}
+#else	/* CONFIG_OCFS2_FS_STATS */
+static void ocfs2_init_la_debug(struct ocfs2_super *osb)
+{
+	return;
+}
+static void ocfs2_shutdown_la_debug(struct ocfs2_super *osb)
+{
+	return;
+}
+#endif
+
 static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb)
 {
 	return (osb->local_alloc_state == OCFS2_LA_THROTTLED ||

@@ -146,6 +226,8 @@ int ocfs2_load_local_alloc(struct ocfs2_super *osb)
 
 	mlog_entry_void();
 
+	ocfs2_init_la_debug(osb);
+
 	if (osb->local_alloc_bits == 0)
 		goto bail;
 

@@ -218,6 +300,9 @@ bail:
 	if (inode)
 		iput(inode);
 
+	if (status < 0)
+		ocfs2_shutdown_la_debug(osb);
+
 	mlog(0, "Local alloc window bits = %d\n", osb->local_alloc_bits);
 
 	mlog_exit(status);

@@ -247,6 +332,8 @@ void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
 	cancel_delayed_work(&osb->la_enable_wq);
 	flush_workqueue(ocfs2_wq);
 
+	ocfs2_shutdown_la_debug(osb);
+
 	if (osb->local_alloc_state == OCFS2_LA_UNUSED)
 		goto out;
 

index 4d6e200..1282799 100644
@@ -272,6 +272,11 @@ struct ocfs2_super
 
 	u64 la_last_gd;
 
+#ifdef CONFIG_OCFS2_FS_STATS
+	struct dentry *local_alloc_debug;
+	char *local_alloc_debug_buf;
+#endif
+
 	/* Next two fields are for local node slot recovery during
 	 * mount. */
 	int dirty;

index 02227c3..b1c634d 100644
@@ -76,8 +76,6 @@ static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 
 #ifdef CONFIG_OCFS2_FS_STATS
 
-DEFINE_MUTEX(la_debug_mutex);
-
 static int ocfs2_la_debug_open(struct inode *inode, struct file *file)
 {
 	file->private_data = inode->i_private;

@@ -89,6 +87,7 @@ static int ocfs2_la_debug_open(struct inode *inode, struct file *file)
 static ssize_t ocfs2_la_debug_read(struct file *file, char __user *userbuf,
 				   size_t count, loff_t *ppos)
 {
+	static DEFINE_MUTEX(la_debug_mutex);
 	struct ocfs2_super *osb = file->private_data;
 	int written, ret;
 	char *buf = osb->local_alloc_debug_buf;

index 10bfb46..c74711f 100644
@@ -368,12 +368,13 @@ struct ocfs2_merge_ctxt {
  */
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
 			   struct inode *inode,
-			   struct ocfs2_dinode *fe)
+			   struct buffer_head *bh)
 {
 	int retval;
 	struct ocfs2_extent_list *el;
 	struct ocfs2_extent_block *eb;
 	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *)bh->b_data;
 
 	mlog_entry_void();
 

index 42ff94b..758dbda 100644
@@ -47,7 +47,7 @@ int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
 			struct ocfs2_cached_dealloc_ctxt *dealloc);
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
 			   struct inode *inode,
-			   struct ocfs2_dinode *fe);
+			   struct buffer_head *bh);
 /* how many new metadata chunks would an allocation need at maximum? */
 static inline int ocfs2_extend_meta_needed(struct ocfs2_dinode *fe)
 {

index a53da14..e2008dc 100644
@@ -1712,8 +1712,9 @@ int ocfs2_write_begin_nolock(struct address_space *mapping,
 		 * ocfs2_lock_allocators(). It greatly over-estimates
 		 * the work to be done.
 		 */
-		ret = ocfs2_lock_allocators(inode, di, clusters_to_alloc,
-					    extents_to_split, &data_ac, &meta_ac);
+		ret = ocfs2_lock_allocators(inode, wc->w_di_bh,
+					    clusters_to_alloc, extents_to_split,
+					    &data_ac, &meta_ac);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;

index 9cce563..fda09c3 100644
@@ -1479,7 +1479,8 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
 	spin_lock(&OCFS2_I(dir)->ip_lock);
 	if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 		spin_unlock(&OCFS2_I(dir)->ip_lock);
-		num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
+		num_free_extents = ocfs2_num_free_extents(osb, dir,
+							  parent_fe_bh);
 		if (num_free_extents < 0) {
 			status = num_free_extents;
 			mlog_errno(status);

index 60232b1..92460f1 100644
@@ -521,7 +521,7 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 	if (mark_unwritten)
 		flags = OCFS2_EXT_UNWRITTEN;
 
-	free_extents = ocfs2_num_free_extents(osb, inode, fe);
+	free_extents = ocfs2_num_free_extents(osb, inode, fe_bh);
 	if (free_extents < 0) {
 		status = free_extents;
 		mlog_errno(status);

@@ -609,7 +609,7 @@ leave:
  * File systems which don't support holes call this from
  * ocfs2_extend_allocation().
  */
-int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
+int ocfs2_lock_allocators(struct inode *inode, struct buffer_head *di_bh,
 			  u32 clusters_to_add, u32 extents_to_split,
 			  struct ocfs2_alloc_context **data_ac,
 			  struct ocfs2_alloc_context **meta_ac)

@@ -617,6 +617,7 @@ int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
 	int ret = 0, num_free_extents;
 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 
 	*meta_ac = NULL;
 	if (data_ac)

@@ -629,7 +630,7 @@ int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, (long long)i_size_read(inode),
 	     le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
 
-	num_free_extents = ocfs2_num_free_extents(osb, inode, di);
+	num_free_extents = ocfs2_num_free_extents(osb, inode, di_bh);
 	if (num_free_extents < 0) {
 		ret = num_free_extents;
 		mlog_errno(ret);

@@ -724,7 +725,7 @@ static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 restart_all:
 	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 
-	status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
+	status = ocfs2_lock_allocators(inode, bh, clusters_to_add, 0, &data_ac,
 				       &meta_ac);
 	if (status) {
 		mlog_errno(status);

@@ -1395,7 +1396,7 @@ static int __ocfs2_remove_inode_range(struct inode *inode,
 	struct ocfs2_alloc_context *meta_ac = NULL;
 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 
-	ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
+	ret = ocfs2_lock_allocators(inode, di_bh, 0, 1, NULL, &meta_ac);
 	if (ret) {
 		mlog_errno(ret);
 		return ret;

index 5a6d3e4..c96b805 100644
@@ -57,7 +57,7 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 			       enum ocfs2_alloc_restarted *reason_ret);
 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size,
 			  u64 zero_to);
-int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
+int ocfs2_lock_allocators(struct inode *inode, struct buffer_head *fe,
 			  u32 clusters_to_add, u32 extents_to_split,
 			  struct ocfs2_alloc_context **data_ac,
 			  struct ocfs2_alloc_context **meta_ac);

index c74711f..dc844df 100644
@@ -4536,7 +4536,8 @@ static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
 	} else
 		rightmost_el = path_leaf_el(path);
 
-	credits += path->p_tree_depth + ocfs2_extend_meta_needed(di);
+	credits += path->p_tree_depth +
+		   ocfs2_extend_meta_needed(&di->id2.i_list);
 	ret = ocfs2_extend_trans(handle, credits);
 	if (ret) {
 		mlog_errno(ret);

index 758dbda..249e79e 100644
@@ -48,8 +48,14 @@ int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
 			   struct inode *inode,
 			   struct buffer_head *bh);
-/* how many new metadata chunks would an allocation need at maximum? */
-static inline int ocfs2_extend_meta_needed(struct ocfs2_dinode *fe)
+/*
+ * how many new metadata chunks would an allocation need at maximum?
+ *
+ * Please note that the caller must make sure that root_el is the root
+ * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
+ * the result may be wrong.
+ */
+static inline int ocfs2_extend_meta_needed(struct ocfs2_extent_list *root_el)
 {
 	/*
 	 * Rather than do all the work of determining how much we need

@@ -59,7 +65,7 @@ static inline int ocfs2_extend_meta_needed(struct ocfs2_dinode *fe)
 	 * new tree_depth==0 extent_block, and one block at the new
 	 * top-of-the tree.
 	 */
-	return le16_to_cpu(fe->id2.i_list.l_tree_depth) + 2;
+	return le16_to_cpu(root_el->l_tree_depth) + 2;
 }
 
 void ocfs2_dinode_new_extent_list(struct inode *inode, struct ocfs2_dinode *di);

index e2008dc..bbe3f8b 100644
@@ -1720,7 +1720,8 @@ int ocfs2_write_begin_nolock(struct address_space *mapping,
 			goto out;
 		}
 
-		credits = ocfs2_calc_extend_credits(inode->i_sb, di,
+		credits = ocfs2_calc_extend_credits(inode->i_sb,
+						    &di->id2.i_list,
 						    clusters_to_alloc);
 
 	}

index fda09c3..126aa21 100644
@@ -1430,6 +1430,7 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
 	int credits, num_free_extents, drop_alloc_sem = 0;
 	loff_t dir_i_size;
 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
+	struct ocfs2_extent_list *el = &fe->id2.i_list;
 	struct ocfs2_alloc_context *data_ac = NULL;
 	struct ocfs2_alloc_context *meta_ac = NULL;
 	handle_t *handle = NULL;

@@ -1488,7 +1489,7 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
 		}
 
 		if (!num_free_extents) {
-			status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
+			status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
 			if (status < 0) {
 				if (status != -ENOSPC)
 					mlog_errno(status);

@@ -1503,7 +1504,7 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
 			goto bail;
 		}
 
-		credits = ocfs2_calc_extend_credits(sb, fe, 1);
+		credits = ocfs2_calc_extend_credits(sb, el, 1);
 	} else {
 		spin_unlock(&OCFS2_I(dir)->ip_lock);
 		credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;

index 92460f1..65b5e65 100644
@@ -540,7 +540,7 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 		goto leave;
 	} else if ((!free_extents)
 		   && (ocfs2_alloc_context_bits_left(meta_ac)
-		       < ocfs2_extend_meta_needed(fe))) {
+		       < ocfs2_extend_meta_needed(&fe->id2.i_list))) {
 		mlog(0, "filesystem is really fragmented...\n");
 		status = -EAGAIN;
 		reason = RESTART_META;

@@ -652,7 +652,7 @@ int ocfs2_lock_allocators(struct inode *inode, struct buffer_head *di_bh,
 	 */
 	if (!num_free_extents ||
 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
-		ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
+		ret = ocfs2_reserve_new_metadata(osb, &di->id2.i_list, meta_ac);
 		if (ret < 0) {
 			if (ret != -ENOSPC)
 				mlog_errno(ret);

@@ -732,7 +732,8 @@ restart_all:
 		goto leave;
 	}
 
-	credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
+	credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
+					    clusters_to_add);
 	handle = ocfs2_start_trans(osb, credits);
 	if (IS_ERR(handle)) {
 		status = PTR_ERR(handle);

@@ -790,7 +791,7 @@ restarted_transaction:
 			mlog(0, "restarting transaction.\n");
 			/* TODO: This can be more intelligent. */
 			credits = ocfs2_calc_extend_credits(osb->sb,
-							    fe,
+							    &fe->id2.i_list,
 							    clusters_to_add);
 			status = ocfs2_extend_trans(handle, credits);
 			if (status < 0) {

index 2178ebf..9485f80 100644
@@ -340,11 +340,16 @@ int                  ocfs2_journal_dirty_data(handle_t *handle,
 #define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3              \
 			     + OCFS2_UNLINK_CREDITS)
 
+/*
+ * Please note that the caller must make sure that root_el is the root
+ * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
+ * the result may be wrong.
+ */
 static inline int ocfs2_calc_extend_credits(struct super_block *sb,
-					    struct ocfs2_dinode *fe,
+					    struct ocfs2_extent_list *root_el,
 					    u32 bits_wanted)
 {
-	int bitmap_blocks, sysfile_bitmap_blocks, dinode_blocks;
+	int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
 
 	/* bitmap dinode, group desc. + relinked group. */
 	bitmap_blocks = OCFS2_SUBALLOC_ALLOC;

@@ -355,16 +360,16 @@ static inline int ocfs2_calc_extend_credits(struct super_block *sb,
 	 * however many metadata chunks needed * a remaining suballoc
 	 * alloc. */
 	sysfile_bitmap_blocks = 1 +
-		(OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(fe);
+		(OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
 
 	/* this does not include *new* metadata blocks, which are
-	 * accounted for in sysfile_bitmap_blocks. fe +
+	 * accounted for in sysfile_bitmap_blocks. root_el +
 	 * prev. last_eb_blk + blocks along edge of tree.
 	 * calc_symlink_credits passes because we just need 1
 	 * credit for the dinode there. */
-	dinode_blocks = 1 + 1 + le16_to_cpu(fe->id2.i_list.l_tree_depth);
+	extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
 
-	return bitmap_blocks + sysfile_bitmap_blocks + dinode_blocks;
+	return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks;
 }
 
 static inline int ocfs2_calc_symlink_credits(struct super_block *sb)

index de7b93d..2a817bc 100644
@@ -494,7 +494,7 @@ bail:
 }
 
 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
-			       struct ocfs2_dinode *fe,
+			       struct ocfs2_extent_list *root_el,
 			       struct ocfs2_alloc_context **ac)
 {
 	int status;

@@ -507,7 +507,7 @@ int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	(*ac)->ac_bits_wanted = ocfs2_extend_meta_needed(fe);
+	(*ac)->ac_bits_wanted = ocfs2_extend_meta_needed(root_el);
 	(*ac)->ac_which = OCFS2_AC_USE_META;
 	slot = osb->slot_num;
 	(*ac)->ac_group_search = ocfs2_block_group_search;

index 40d51da..3f96c87 100644
@@ -59,8 +59,13 @@ static inline int ocfs2_alloc_context_bits_left(struct ocfs2_alloc_context *ac)
 	return ac->ac_bits_wanted - ac->ac_bits_given;
 }
 
+/*
+ * Please note that the caller must make sure that root_el is the root
+ * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
+ * the result may be wrong.
+ */
 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
-			       struct ocfs2_dinode *fe,
+			       struct ocfs2_extent_list *root_el,
 			       struct ocfs2_alloc_context **ac);
 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
 			    struct ocfs2_alloc_context **ac);

index dc844df..90cefc5 100644
@@ -49,6 +49,143 @@
 
 #include "buffer_head_io.h"
 
+/*
+ * ocfs2_extent_tree and ocfs2_extent_tree_operations are used to abstract
+ * the b-tree operations in ocfs2. Now all the b-tree operations are not
+ * limited to ocfs2_dinode only. Any data which need to allocate clusters
+ * to store can use b-tree. And it only needs to implement its ocfs2_extent_tree
+ * and operation.
+ *
+ * ocfs2_extent_tree contains info for the root of the b-tree, it must have a
+ * root ocfs2_extent_list and a root_bh so that they can be used in the b-tree
+ * functions.
+ * ocfs2_extent_tree_operations abstract the normal operations we do for
+ * the root of extent b-tree.
+ */
+struct ocfs2_extent_tree;
+
+struct ocfs2_extent_tree_operations {
+	void (*set_last_eb_blk) (struct ocfs2_extent_tree *et, u64 blkno);
+	u64 (*get_last_eb_blk) (struct ocfs2_extent_tree *et);
+	void (*update_clusters) (struct inode *inode,
+				 struct ocfs2_extent_tree *et,
+				 u32 new_clusters);
+	int (*sanity_check) (struct inode *inode, struct ocfs2_extent_tree *et);
+};
+
+struct ocfs2_extent_tree {
+	enum ocfs2_extent_tree_type type;
+	struct ocfs2_extent_tree_operations *eops;
+	struct buffer_head *root_bh;
+	struct ocfs2_extent_list *root_el;
+};
+
+static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					 u64 blkno)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)et->root_bh->b_data;
+
+	BUG_ON(et->type != OCFS2_DINODE_EXTENT);
+	di->i_last_eb_blk = cpu_to_le64(blkno);
+}
+
+static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)et->root_bh->b_data;
+
+	BUG_ON(et->type != OCFS2_DINODE_EXTENT);
+	return le64_to_cpu(di->i_last_eb_blk);
+}
+
+static void ocfs2_dinode_update_clusters(struct inode *inode,
+					 struct ocfs2_extent_tree *et,
+					 u32 clusters)
+{
+	struct ocfs2_dinode *di =
+			(struct ocfs2_dinode *)et->root_bh->b_data;
+
+	le32_add_cpu(&di->i_clusters, clusters);
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+	OCFS2_I(inode)->ip_clusters = le32_to_cpu(di->i_clusters);
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+}
+
+static int ocfs2_dinode_sanity_check(struct inode *inode,
+				     struct ocfs2_extent_tree *et)
+{
+	int ret = 0;
+	struct ocfs2_dinode *di;
+
+	BUG_ON(et->type != OCFS2_DINODE_EXTENT);
+
+	di = (struct ocfs2_dinode *)et->root_bh->b_data;
+	if (!OCFS2_IS_VALID_DINODE(di)) {
+		ret = -EIO;
+		ocfs2_error(inode->i_sb,
+			"Inode %llu has invalid path root",
+			(unsigned long long)OCFS2_I(inode)->ip_blkno);
+	}
+
+	return ret;
+}
+
+static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
+	.set_last_eb_blk	= ocfs2_dinode_set_last_eb_blk,
+	.get_last_eb_blk	= ocfs2_dinode_get_last_eb_blk,
+	.update_clusters	= ocfs2_dinode_update_clusters,
+	.sanity_check		= ocfs2_dinode_sanity_check,
+};
+
+static struct ocfs2_extent_tree*
+	 ocfs2_new_extent_tree(struct buffer_head *bh,
+			       enum ocfs2_extent_tree_type et_type)
+{
+	struct ocfs2_extent_tree *et;
+
+	et = kzalloc(sizeof(*et), GFP_NOFS);
+	if (!et)
+		return NULL;
+
+	et->type = et_type;
+	get_bh(bh);
+	et->root_bh = bh;
+
+	/* current we only support dinode extent. */
+	BUG_ON(et->type != OCFS2_DINODE_EXTENT);
+	if (et_type == OCFS2_DINODE_EXTENT) {
+		et->root_el = &((struct ocfs2_dinode *)bh->b_data)->id2.i_list;
+		et->eops = &ocfs2_dinode_et_ops;
+	}
+
+	return et;
+}
+
+static void ocfs2_free_extent_tree(struct ocfs2_extent_tree *et)
+{
+	if (et) {
+		brelse(et->root_bh);
+		kfree(et);
+	}
+}
+
+static inline void ocfs2_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					 u64 new_last_eb_blk)
+{
+	et->eops->set_last_eb_blk(et, new_last_eb_blk);
+}
+
+static inline u64 ocfs2_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	return et->eops->get_last_eb_blk(et);
+}
+
+static inline void ocfs2_update_clusters(struct inode *inode,
+					 struct ocfs2_extent_tree *et,
+					 u32 clusters)
+{
+	et->eops->update_clusters(inode, et, clusters);
+}
+
 static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
 static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
 					 struct ocfs2_extent_block *eb);

@@ -205,17 +342,6 @@ static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
 }
 
 /*
- * Allocate and initialize a new path based on a disk inode tree.
- */
-static struct ocfs2_path *ocfs2_new_inode_path(struct buffer_head *di_bh)
-{
-	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
-	struct ocfs2_extent_list *el = &di->id2.i_list;
-
-	return ocfs2_new_path(di_bh, el);
-}
-
-/*
  * Convenience function to journal all components in a path.
  */
 static int ocfs2_journal_access_path(struct inode *inode, handle_t *handle,

@@ -368,24 +494,33 @@ struct ocfs2_merge_ctxt {
  */
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
 			   struct inode *inode,
-			   struct buffer_head *bh)
+			   struct buffer_head *root_bh,
+			   enum ocfs2_extent_tree_type type)
 {
 	int retval;
-	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_list *el = NULL;
 	struct ocfs2_extent_block *eb;
 	struct buffer_head *eb_bh = NULL;
-	struct ocfs2_dinode *fe = (struct ocfs2_dinode *)bh->b_data;
+	u64 last_eb_blk = 0;
 
 	mlog_entry_void();
 
-	if (!OCFS2_IS_VALID_DINODE(fe)) {
-		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
-		retval = -EIO;
-		goto bail;
+	if (type == OCFS2_DINODE_EXTENT) {
+		struct ocfs2_dinode *fe =
+				(struct ocfs2_dinode *)root_bh->b_data;
+		if (!OCFS2_IS_VALID_DINODE(fe)) {
+			OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
+			retval = -EIO;
+			goto bail;
+		}
+
+		if (fe->i_last_eb_blk)
+			last_eb_blk = le64_to_cpu(fe->i_last_eb_blk);
+		el = &fe->id2.i_list;
 	}
 
-	if (fe->i_last_eb_blk) {
-		retval = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
+	if (last_eb_blk) {
+		retval = ocfs2_read_block(osb, last_eb_blk,
 					  &eb_bh, OCFS2_BH_CACHED, inode);
 		if (retval < 0) {
 			mlog_errno(retval);

@@ -393,8 +528,7 @@ int ocfs2_num_free_extents(struct ocfs2_super *osb,
 		}
 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 		el = &eb->h_list;
-	} else
-		el = &fe->id2.i_list;
+	}
 
 	BUG_ON(el->l_tree_depth != 0);
 

@@ -532,7 +666,7 @@ static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list  *el)
 static int ocfs2_add_branch(struct ocfs2_super *osb,
 			    handle_t *handle,
 			    struct inode *inode,
-			    struct buffer_head *fe_bh,
+			    struct ocfs2_extent_tree *et,
 			    struct buffer_head *eb_bh,
 			    struct buffer_head **last_eb_bh,
 			    struct ocfs2_alloc_context *meta_ac)

@@ -541,7 +675,6 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 	u64 next_blkno, new_last_eb_blk;
 	struct buffer_head *bh;
 	struct buffer_head **new_eb_bhs = NULL;
-	struct ocfs2_dinode *fe;
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list  *eb_el;
 	struct ocfs2_extent_list  *el;

@@ -551,13 +684,11 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 
 	BUG_ON(!last_eb_bh || !*last_eb_bh);
 
-	fe = (struct ocfs2_dinode *) fe_bh->b_data;
-
 	if (eb_bh) {
 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 		el = &eb->h_list;
 	} else
-		el = &fe->id2.i_list;
+		el = et->root_el;
 
 	/* we never add a branch to a leaf. */
 	BUG_ON(!el->l_tree_depth);

@@ -647,7 +778,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 		mlog_errno(status);
 		goto bail;
 	}
-	status = ocfs2_journal_access(handle, inode, fe_bh,
+	status = ocfs2_journal_access(handle, inode, et->root_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);

@@ -663,7 +794,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 	}
 
 	/* Link the new branch into the rest of the tree (el will
-	 * either be on the fe, or the extent block passed in. */
+	 * either be on the root_bh, or the extent block passed in. */
 	i = le16_to_cpu(el->l_next_free_rec);
 	el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
 	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);

@@ -672,7 +803,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 
 	/* fe needs a new last extent block pointer, as does the
 	 * next_leaf on the previously last-extent-block. */
-	fe->i_last_eb_blk = cpu_to_le64(new_last_eb_blk);
+	ocfs2_set_last_eb_blk(et, new_last_eb_blk);
 
 	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
 	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);

@@ -680,7 +811,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 	status = ocfs2_journal_dirty(handle, *last_eb_bh);
 	if (status < 0)
 		mlog_errno(status);
-	status = ocfs2_journal_dirty(handle, fe_bh);
+	status = ocfs2_journal_dirty(handle, et->root_bh);
 	if (status < 0)
 		mlog_errno(status);
 	if (eb_bh) {

@@ -718,16 +849,15 @@ bail:
 static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 				  handle_t *handle,
 				  struct inode *inode,
-				  struct buffer_head *fe_bh,
+				  struct ocfs2_extent_tree *et,
 				  struct ocfs2_alloc_context *meta_ac,
 				  struct buffer_head **ret_new_eb_bh)
 {
 	int status, i;
 	u32 new_clusters;
 	struct buffer_head *new_eb_bh = NULL;
-	struct ocfs2_dinode *fe;
 	struct ocfs2_extent_block *eb;
-	struct ocfs2_extent_list  *fe_el;
+	struct ocfs2_extent_list  *root_el;
 	struct ocfs2_extent_list  *eb_el;
 
 	mlog_entry_void();

@@ -747,8 +877,7 @@ static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 	}
 
 	eb_el = &eb->h_list;
-	fe = (struct ocfs2_dinode *) fe_bh->b_data;
-	fe_el = &fe->id2.i_list;
+	root_el = et->root_el;
 
 	status = ocfs2_journal_access(handle, inode, new_eb_bh,
 				      OCFS2_JOURNAL_ACCESS_CREATE);

@@ -757,11 +886,11 @@ static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	/* copy the fe data into the new extent block */
-	eb_el->l_tree_depth = fe_el->l_tree_depth;
-	eb_el->l_next_free_rec = fe_el->l_next_free_rec;
-	for(i = 0; i < le16_to_cpu(fe_el->l_next_free_rec); i++)
-		eb_el->l_recs[i] = fe_el->l_recs[i];
+	/* copy the root extent list data into the new extent block */
+	eb_el->l_tree_depth = root_el->l_tree_depth;
+	eb_el->l_next_free_rec = root_el->l_next_free_rec;
+	for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
+		eb_el->l_recs[i] = root_el->l_recs[i];
 
 	status = ocfs2_journal_dirty(handle, new_eb_bh);
 	if (status < 0) {

@@ -769,7 +898,7 @@ static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	status = ocfs2_journal_access(handle, inode, fe_bh,
+	status = ocfs2_journal_access(handle, inode, et->root_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);

@@ -778,21 +907,21 @@ static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 
 	new_clusters = ocfs2_sum_rightmost_rec(eb_el);
 
-	/* update fe now */
-	le16_add_cpu(&fe_el->l_tree_depth, 1);
-	fe_el->l_recs[0].e_cpos = 0;
-	fe_el->l_recs[0].e_blkno = eb->h_blkno;
-	fe_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
-	for(i = 1; i < le16_to_cpu(fe_el->l_next_free_rec); i++)
-		memset(&fe_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
-	fe_el->l_next_free_rec = cpu_to_le16(1);
+	/* update root_bh now */
+	le16_add_cpu(&root_el->l_tree_depth, 1);
+	root_el->l_recs[0].e_cpos = 0;
+	root_el->l_recs[0].e_blkno = eb->h_blkno;
+	root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
+	for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
+		memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
+	root_el->l_next_free_rec = cpu_to_le16(1);
 
 	/* If this is our 1st tree depth shift, then last_eb_blk
 	 * becomes the allocated extent block */
-	if (fe_el->l_tree_depth == cpu_to_le16(1))
-		fe->i_last_eb_blk = eb->h_blkno;
+	if (root_el->l_tree_depth == cpu_to_le16(1))
+		ocfs2_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
 
-	status = ocfs2_journal_dirty(handle, fe_bh);
+	status = ocfs2_journal_dirty(handle, et->root_bh);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;

@@ -818,22 +947,21 @@ bail:
  * 1) a lowest extent block is found, then we pass it back in
  *    *lowest_eb_bh and return '0'
  *
- * 2) the search fails to find anything, but the dinode has room. We
+ * 2) the search fails to find anything, but the root_el has room. We
  *    pass NULL back in *lowest_eb_bh, but still return '0'
  *
- * 3) the search fails to find anything AND the dinode is full, in
+ * 3) the search fails to find anything AND the root_el is full, in
  *    which case we return > 0
  *
  * return status < 0 indicates an error.
  */
 static int ocfs2_find_branch_target(struct ocfs2_super *osb,
 				    struct inode *inode,
-				    struct buffer_head *fe_bh,
+				    struct ocfs2_extent_tree *et,
 				    struct buffer_head **target_bh)
 {
 	int status = 0, i;
 	u64 blkno;
-	struct ocfs2_dinode *fe;
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list  *el;
 	struct buffer_head *bh = NULL;

@@ -843,8 +971,7 @@ static int ocfs2_find_branch_target(struct ocfs2_super *osb,
 
 	*target_bh = NULL;
 
-	fe = (struct ocfs2_dinode *) fe_bh->b_data;
-	el = &fe->id2.i_list;
+	el = et->root_el;
 
 	while(le16_to_cpu(el->l_tree_depth) > 1) {
 		if (le16_to_cpu(el->l_next_free_rec) == 0) {

@@ -896,8 +1023,8 @@ static int ocfs2_find_branch_target(struct ocfs2_super *osb,
 
 	/* If we didn't find one and the fe doesn't have any room,
 	 * then return '1' */
-	if (!lowest_bh
-	    && (fe->id2.i_list.l_next_free_rec == fe->id2.i_list.l_count))
+	el = et->root_el;
+	if (!lowest_bh && (el->l_next_free_rec == el->l_count))
 		status = 1;
 
 	*target_bh = lowest_bh;

@@ -920,19 +1047,19 @@ bail:
  * *last_eb_bh will be updated by ocfs2_add_branch().
  */
 static int ocfs2_grow_tree(struct inode *inode, handle_t *handle,
-			   struct buffer_head *di_bh, int *final_depth,
+			   struct ocfs2_extent_tree *et, int *final_depth,
 			   struct buffer_head **last_eb_bh,
 			   struct ocfs2_alloc_context *meta_ac)
 {
 	int ret, shift;
-	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
-	int depth = le16_to_cpu(di->id2.i_list.l_tree_depth);
+	struct ocfs2_extent_list *el = et->root_el;
+	int depth = le16_to_cpu(el->l_tree_depth);
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 	struct buffer_head *bh = NULL;
 
 	BUG_ON(meta_ac == NULL);
 
-	shift = ocfs2_find_branch_target(osb, inode, di_bh, &bh);
+	shift = ocfs2_find_branch_target(osb, inode, et, &bh);
 	if (shift < 0) {
 		ret = shift;
 		mlog_errno(ret);

@@ -949,7 +1076,7 @@ static int ocfs2_grow_tree(struct inode *inode, handle_t *handle,
 		/* ocfs2_shift_tree_depth will return us a buffer with
 		 * the new extent block (so we can pass that to
 		 * ocfs2_add_branch). */
-		ret = ocfs2_shift_tree_depth(osb, handle, inode, di_bh,
+		ret = ocfs2_shift_tree_depth(osb, handle, inode, et,
 					     meta_ac, &bh);
 		if (ret < 0) {
 			mlog_errno(ret);

@@ -976,7 +1103,7 @@ static int ocfs2_grow_tree(struct inode *inode, handle_t *handle,
 	/* call ocfs2_add_branch to add the final part of the tree with
 	 * the new data. */
 	mlog(0, "add branch. bh = %p\n", bh);
-	ret = ocfs2_add_branch(osb, handle, inode, di_bh, bh, last_eb_bh,
+	ret = ocfs2_add_branch(osb, handle, inode, et, bh, last_eb_bh,
 			       meta_ac);
 	if (ret < 0) {
 		mlog_errno(ret);

@@ -2068,11 +2195,11 @@ static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle,
 				     struct ocfs2_path *right_path,
 				     int subtree_index,
 				     struct ocfs2_cached_dealloc_ctxt *dealloc,
-				     int *deleted)
+				     int *deleted,
+				     struct ocfs2_extent_tree *et)
 {
 	int ret, i, del_right_subtree = 0, right_has_empty = 0;
-	struct buffer_head *root_bh, *di_bh = path_root_bh(right_path);
-	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
 	struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
 	struct ocfs2_extent_block *eb;
 

@@ -2124,7 +2251,7 @@ static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle,
 		 * We have to update i_last_eb_blk during the meta
 		 * data delete.
 		 */
-		ret = ocfs2_journal_access(handle, inode, di_bh,
+		ret = ocfs2_journal_access(handle, inode, et_root_bh,
 					   OCFS2_JOURNAL_ACCESS_WRITE);
 		if (ret) {
 			mlog_errno(ret);

@@ -2199,7 +2326,7 @@ static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle,
 		ocfs2_update_edge_lengths(inode, handle, left_path);
 
 		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
-		di->i_last_eb_blk = eb->h_blkno;
+		ocfs2_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
 
 		/*
 		 * Removal of the extent in the left leaf was skipped

@@ -2209,7 +2336,7 @@ static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle,
 		if (right_has_empty)
 			ocfs2_remove_empty_extent(left_leaf_el);
 
-		ret = ocfs2_journal_dirty(handle, di_bh);
+		ret = ocfs2_journal_dirty(handle, et_root_bh);
 		if (ret)
 			mlog_errno(ret);
 

@@ -2332,7 +2459,8 @@ static int __ocfs2_rotate_tree_left(struct inode *inode,
 				    handle_t *handle, int orig_credits,
 				    struct ocfs2_path *path,
 				    struct ocfs2_cached_dealloc_ctxt *dealloc,
-				    struct ocfs2_path **empty_extent_path)
+				    struct ocfs2_path **empty_extent_path,
+				    struct ocfs2_extent_tree *et)
 {
 	int ret, subtree_root, deleted;
 	u32 right_cpos;

@@ -2405,7 +2533,7 @@ static int __ocfs2_rotate_tree_left(struct inode *inode,
 
 		ret = ocfs2_rotate_subtree_left(inode, handle, left_path,
 						right_path, subtree_root,
-						dealloc, &deleted);
+						dealloc, &deleted, et);
 		if (ret == -EAGAIN) {
 			/*
 			 * The rotation has to temporarily stop due to

@@ -2448,29 +2576,20 @@ out:
 }
 
 static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle,
-				       struct ocfs2_path *path,
-				       struct ocfs2_cached_dealloc_ctxt *dealloc)
+				struct ocfs2_path *path,
+				struct ocfs2_cached_dealloc_ctxt *dealloc,
+				struct ocfs2_extent_tree *et)
 {
 	int ret, subtree_index;
 	u32 cpos;
 	struct ocfs2_path *left_path = NULL;
-	struct ocfs2_dinode *di;
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list *el;
 
-	/*
-	 * XXX: This code assumes that the root is an inode, which is
-	 * true for now but may change as tree code gets generic.
-	 */
-	di = (struct ocfs2_dinode *)path_root_bh(path)->b_data;
-	if (!OCFS2_IS_VALID_DINODE(di)) {
-		ret = -EIO;
-		ocfs2_error(inode->i_sb,
-			    "Inode %llu has invalid path root",
-			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
-		goto out;
-	}
 
+	ret = et->eops->sanity_check(inode, et);
+	if (ret)
+		goto out;
 	/*
 	 * There's two ways we handle this depending on
 	 * whether path is the only existing one.

@@ -2527,7 +2646,7 @@ static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle,
 		ocfs2_update_edge_lengths(inode, handle, left_path);
 
 		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
-		di->i_last_eb_blk = eb->h_blkno;
+		ocfs2_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
 	} else {
 		/*
 		 * 'path' is also the leftmost path which

@@ -2538,12 +2657,12 @@ static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle,
 		 */
 		ocfs2_unlink_path(inode, handle, dealloc, path, 1);
 
-		el = &di->id2.i_list;
+		el = et->root_el;
 		el->l_tree_depth = 0;
 		el->l_next_free_rec = 0;
 		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
 
-		di->i_last_eb_blk = 0;
+		ocfs2_set_last_eb_blk(et, 0);
 	}
 
 	ocfs2_journal_dirty(handle, path_root_bh(path));

@@ -2571,7 +2690,8 @@ out:
  */
 static int ocfs2_rotate_tree_left(struct inode *inode, handle_t *handle,
 				  struct ocfs2_path *path,
-				  struct ocfs2_cached_dealloc_ctxt *dealloc)
+				  struct ocfs2_cached_dealloc_ctxt *dealloc,
+				  struct ocfs2_extent_tree *et)
 {
 	int ret, orig_credits = handle->h_buffer_credits;
 	struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;

@@ -2585,7 +2705,7 @@ static int ocfs2_rotate_tree_left(struct inode *inode, handle_t *handle,
 	if (path->p_tree_depth == 0) {
 rightmost_no_delete:
 		/*
-		 * In-inode extents. This is trivially handled, so do
+		 * Inline extents. This is trivially handled, so do
 		 * it up front.
 		 */
 		ret = ocfs2_rotate_rightmost_leaf_left(inode, handle,

@@ -2639,7 +2759,7 @@ rightmost_no_delete:
 		 */
 
 		ret = ocfs2_remove_rightmost_path(inode, handle, path,
-						  dealloc);
+						  dealloc, et);
 		if (ret)
 			mlog_errno(ret);
 		goto out;

@@ -2651,7 +2771,7 @@ rightmost_no_delete:
 	 */
 try_rotate:
 	ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, path,
-				       dealloc, &restart_path);
+				       dealloc, &restart_path, et);
 	if (ret && ret != -EAGAIN) {
 		mlog_errno(ret);
 		goto out;

@@ -2663,7 +2783,7 @@ try_rotate:
 
 		ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits,
 					       tmp_path, dealloc,
-					       &restart_path);
+					       &restart_path, et);
 		if (ret && ret != -EAGAIN) {
 			mlog_errno(ret);
 			goto out;

@@ -2949,6 +3069,7 @@ static int ocfs2_merge_rec_left(struct inode *inode,
 				handle_t *handle,
 				struct ocfs2_extent_rec *split_rec,
 				struct ocfs2_cached_dealloc_ctxt *dealloc,
+				struct ocfs2_extent_tree *et,
 				int index)
 {
 	int ret, i, subtree_index = 0, has_empty_extent = 0;

@@ -3069,7 +3190,8 @@ static int ocfs2_merge_rec_left(struct inode *inode,
 		    le16_to_cpu(el->l_next_free_rec) == 1) {
 
 			ret = ocfs2_remove_rightmost_path(inode, handle,
-							  right_path, dealloc);
+							  right_path,
+							  dealloc, et);
 			if (ret) {
 				mlog_errno(ret);
 				goto out;

@@ -3096,7 +3218,8 @@ static int ocfs2_try_to_merge_extent(struct inode *inode,
 				     int split_index,
 				     struct ocfs2_extent_rec *split_rec,
 				     struct ocfs2_cached_dealloc_ctxt *dealloc,
-				     struct ocfs2_merge_ctxt *ctxt)
+				     struct ocfs2_merge_ctxt *ctxt,
+				     struct ocfs2_extent_tree *et)
 
 {
 	int ret = 0;

@@ -3114,7 +3237,7 @@ static int ocfs2_try_to_merge_extent(struct inode *inode,
 		 * illegal.
 		 */
 		ret = ocfs2_rotate_tree_left(inode, handle, path,
-					     dealloc);
+					     dealloc, et);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;

@@ -3157,7 +3280,8 @@ static int ocfs2_try_to_merge_extent(struct inode *inode,
 		BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
 
 		/* The merge left us with an empty extent, remove it. */
-		ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc);
+		ret = ocfs2_rotate_tree_left(inode, handle, path,
+					     dealloc, et);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;

@@ -3171,7 +3295,7 @@ static int ocfs2_try_to_merge_extent(struct inode *inode,
 		 */
 		ret = ocfs2_merge_rec_left(inode, path,
 					   handle, rec,
-					   dealloc,
+					   dealloc, et,
 					   split_index);
 
 		if (ret) {

@@ -3180,7 +3304,7 @@ static int ocfs2_try_to_merge_extent(struct inode *inode,
 		}
 
 		ret = ocfs2_rotate_tree_left(inode, handle, path,
-					     dealloc);
+					     dealloc, et);
 		/*
 		 * Error from this last rotate is not critical, so
 		 * print but don't bubble it up.

@@ -3200,7 +3324,7 @@ static int ocfs2_try_to_merge_extent(struct inode *inode,
 			ret = ocfs2_merge_rec_left(inode,
 						   path,
 						   handle, split_rec,
-						   dealloc,
+						   dealloc, et,
 						   split_index);
 			if (ret) {
 				mlog_errno(ret);

@@ -3223,7 +3347,7 @@ static int ocfs2_try_to_merge_extent(struct inode *inode,
 			 * our leaf. Try to rotate it away.
 			 */
 			ret = ocfs2_rotate_tree_left(inode, handle, path,
-						     dealloc);
+						     dealloc, et);
 			if (ret)
 				mlog_errno(ret);
 			ret = 0;

@@ -3357,16 +3481,6 @@ rotate:
 	ocfs2_rotate_leaf(el, insert_rec);
 }
 
-static inline void ocfs2_update_dinode_clusters(struct inode *inode,
-						struct ocfs2_dinode *di,
-						u32 clusters)
-{
-	le32_add_cpu(&di->i_clusters, clusters);
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	OCFS2_I(inode)->ip_clusters = le32_to_cpu(di->i_clusters);
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-}
-
 static void ocfs2_adjust_rightmost_records(struct inode *inode,
 					   handle_t *handle,
 					   struct ocfs2_path *path,

@@ -3568,8 +3682,8 @@ static void ocfs2_split_record(struct inode *inode,
 }
 
 /*
- * This function only does inserts on an allocation b-tree. For dinode
- * lists, ocfs2_insert_at_leaf() is called directly.
+ * This function only does inserts on an allocation b-tree. For tree
+ * depth = 0, ocfs2_insert_at_leaf() is called directly.
  *
  * right_path is the path we want to do the actual insert
  * in. left_path should only be passed in if we need to update that

@@ -3666,7 +3780,7 @@ out:
 
 static int ocfs2_do_insert_extent(struct inode *inode,
 				  handle_t *handle,
-				  struct buffer_head *di_bh,
+				  struct ocfs2_extent_tree *et,
 				  struct ocfs2_extent_rec *insert_rec,
 				  struct ocfs2_insert_type *type)
 {

@@ -3674,13 +3788,11 @@ static int ocfs2_do_insert_extent(struct inode *inode,
 	u32 cpos;
 	struct ocfs2_path *right_path = NULL;
 	struct ocfs2_path *left_path = NULL;
-	struct ocfs2_dinode *di;
 	struct ocfs2_extent_list *el;
 
-	di = (struct ocfs2_dinode *) di_bh->b_data;
-	el = &di->id2.i_list;
+	el = et->root_el;
 
-	ret = ocfs2_journal_access(handle, inode, di_bh,
+	ret = ocfs2_journal_access(handle, inode, et->root_bh,
 				   OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);

@@ -3692,7 +3804,7 @@ static int ocfs2_do_insert_extent(struct inode *inode,
 		goto out_update_clusters;
 	}
 
-	right_path = ocfs2_new_inode_path(di_bh);
+	right_path = ocfs2_new_path(et->root_bh, et->root_el);
 	if (!right_path) {
 		ret = -ENOMEM;
 		mlog_errno(ret);

@@ -3742,7 +3854,7 @@ static int ocfs2_do_insert_extent(struct inode *inode,
 		 * ocfs2_rotate_tree_right() might have extended the
 		 * transaction without re-journaling our tree root.
 		 */
-		ret = ocfs2_journal_access(handle, inode, di_bh,
+		ret = ocfs2_journal_access(handle, inode, et->root_bh,
 					   OCFS2_JOURNAL_ACCESS_WRITE);
 		if (ret) {
 			mlog_errno(ret);

@@ -3767,10 +3879,10 @@ static int ocfs2_do_insert_extent(struct inode *inode,
 
 out_update_clusters:
 	if (type->ins_split == SPLIT_NONE)
-		ocfs2_update_dinode_clusters(inode, di,
-					     le16_to_cpu(insert_rec->e_leaf_clusters));
+		ocfs2_update_clusters(inode, et,
+				      le16_to_cpu(insert_rec->e_leaf_clusters));
 
-	ret = ocfs2_journal_dirty(handle, di_bh);
+	ret = ocfs2_journal_dirty(handle, et->root_bh);
 	if (ret)
 		mlog_errno(ret);
 

@@ -3924,8 +4036,8 @@ static void ocfs2_figure_contig_type(struct inode *inode,
  * ocfs2_figure_appending_type() will figure out whether we'll have to
  * insert at the tail of the rightmost leaf.
  *
- * This should also work against the dinode list for tree's with 0
- * depth. If we consider the dinode list to be the rightmost leaf node
+ * This should also work against the root extent list for tree's with 0
+ * depth. If we consider the root extent list to be the rightmost leaf node
  * then the logic here makes sense.
  */
 static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,

@@ -3976,14 +4088,13 @@ set_tail_append:
  * structure.
  */
 static int ocfs2_figure_insert_type(struct inode *inode,
-				    struct buffer_head *di_bh,
+				    struct ocfs2_extent_tree *et,
 				    struct buffer_head **last_eb_bh,
 				    struct ocfs2_extent_rec *insert_rec,
 				    int *free_records,
 				    struct ocfs2_insert_type *insert)
 {
 	int ret;
-	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list *el;
 	struct ocfs2_path *path = NULL;

@@ -3991,7 +4102,7 @@ static int ocfs2_figure_insert_type(struct inode *inode,
 
 	insert->ins_split = SPLIT_NONE;
 
-	el = &di->id2.i_list;
+	el = et->root_el;
 	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
 
 	if (el->l_tree_depth) {

@@ -4002,7 +4113,7 @@ static int ocfs2_figure_insert_type(struct inode *inode,
 		 * may want it later.
 		 */
 		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
-				       le64_to_cpu(di->i_last_eb_blk), &bh,
+				       ocfs2_get_last_eb_blk(et), &bh,
 				       OCFS2_BH_CACHED, inode);
 		if (ret) {
 			mlog_exit(ret);

@@ -4029,7 +4140,7 @@ static int ocfs2_figure_insert_type(struct inode *inode,
 		return 0;
 	}
 
-	path = ocfs2_new_inode_path(di_bh);
+	path = ocfs2_new_path(et->root_bh, et->root_el);
 	if (!path) {
 		ret = -ENOMEM;
 		mlog_errno(ret);

@@ -4079,7 +4190,8 @@ static int ocfs2_figure_insert_type(struct inode *inode,
 	 * the case that we're doing a tail append, so maybe we can
 	 * take advantage of that information somehow.
 	 */
-	if (le64_to_cpu(di->i_last_eb_blk) == path_leaf_bh(path)->b_blocknr) {
+	if (ocfs2_get_last_eb_blk(et) ==
+	    path_leaf_bh(path)->b_blocknr) {
 		/*
 		 * Ok, ocfs2_find_path() returned us the rightmost
 		 * tree path. This might be an appending insert. There are

@@ -4109,21 +4221,30 @@ out:
 int ocfs2_insert_extent(struct ocfs2_super *osb,
 			handle_t *handle,
 			struct inode *inode,
-			struct buffer_head *fe_bh,
+			struct buffer_head *root_bh,
 			u32 cpos,
 			u64 start_blk,
 			u32 new_clusters,
 			u8 flags,
-			struct ocfs2_alloc_context *meta_ac)
+			struct ocfs2_alloc_context *meta_ac,
+			enum ocfs2_extent_tree_type et_type)
 {
 	int status;
 	int uninitialized_var(free_records);
 	struct buffer_head *last_eb_bh = NULL;
 	struct ocfs2_insert_type insert = {0, };
 	struct ocfs2_extent_rec rec;
+	struct ocfs2_extent_tree *et = NULL;
 
 	BUG_ON(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL);
 
+	et = ocfs2_new_extent_tree(root_bh, et_type);
+	if (!et) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
 	mlog(0, "add %u clusters at position %u to inode %llu\n",
 	     new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno);
 

@@ -4141,7 +4262,7 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
 	rec.e_flags = flags;
 
-	status = ocfs2_figure_insert_type(inode, fe_bh, &last_eb_bh, &rec,
+	status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec,
 					  &free_records, &insert);
 	if (status < 0) {
 		mlog_errno(status);

@@ -4155,7 +4276,7 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 	     free_records, insert.ins_tree_depth);
 
 	if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
-		status = ocfs2_grow_tree(inode, handle, fe_bh,
+		status = ocfs2_grow_tree(inode, handle, et,
 					 &insert.ins_tree_depth, &last_eb_bh,
 					 meta_ac);
 		if (status) {

@@ -4165,16 +4286,18 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 	}
 
 	/* Finally, we can add clusters. This might rotate the tree for us. */
-	status = ocfs2_do_insert_extent(inode, handle, fe_bh, &rec, &insert);
+	status = ocfs2_do_insert_extent(inode, handle, et, &rec, &insert);
 	if (status < 0)
 		mlog_errno(status);
-	else
+	else if (et->type == OCFS2_DINODE_EXTENT)
 		ocfs2_extent_map_insert_rec(inode, &rec);
 
 bail:
 	if (last_eb_bh)
 		brelse(last_eb_bh);
 
+	if (et)
+		ocfs2_free_extent_tree(et);
 	mlog_exit(status);
 	return status;
 }

@@ -4202,7 +4325,7 @@ static void ocfs2_make_right_split_rec(struct super_block *sb,
 static int ocfs2_split_and_insert(struct inode *inode,
 				  handle_t *handle,
 				  struct ocfs2_path *path,
-				  struct buffer_head *di_bh,
+				  struct ocfs2_extent_tree *et,
 				  struct buffer_head **last_eb_bh,
 				  int split_index,
 				  struct ocfs2_extent_rec *orig_split_rec,

@@ -4216,7 +4339,6 @@ static int ocfs2_split_and_insert(struct inode *inode,
 	struct ocfs2_extent_rec split_rec = *orig_split_rec;
 	struct ocfs2_insert_type insert;
 	struct ocfs2_extent_block *eb;
-	struct ocfs2_dinode *di;
 
 leftright:
 	/*

@@ -4225,8 +4347,7 @@ leftright:
 	 */
 	rec = path_leaf_el(path)->l_recs[split_index];
 
-	di = (struct ocfs2_dinode *)di_bh->b_data;
-	rightmost_el = &di->id2.i_list;
+	rightmost_el = et->root_el;
 
 	depth = le16_to_cpu(rightmost_el->l_tree_depth);
 	if (depth) {

@@ -4237,8 +4358,8 @@ leftright:
 
 	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
 	    le16_to_cpu(rightmost_el->l_count)) {
-		ret = ocfs2_grow_tree(inode, handle, di_bh, &depth, last_eb_bh,
-				      meta_ac);
+		ret = ocfs2_grow_tree(inode, handle, et,
+				      &depth, last_eb_bh, meta_ac);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;

@@ -4275,8 +4396,7 @@ leftright:
 		do_leftright = 1;
 	}
 
-	ret = ocfs2_do_insert_extent(inode, handle, di_bh, &split_rec,
-				     &insert);
+	ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;

@@ -4318,8 +4438,9 @@ out:
  * of the tree is required. All other cases will degrade into a less
  * optimal tree layout.
  *
- * last_eb_bh should be the rightmost leaf block for any inode with a
- * btree. Since a split may grow the tree or a merge might shrink it, the caller cannot trust the contents of that buffer after this call.
+ * last_eb_bh should be the rightmost leaf block for any extent
+ * btree. Since a split may grow the tree or a merge might shrink it,
+ * the caller cannot trust the contents of that buffer after this call.
  *
  * This code is optimized for readability - several passes might be
  * made over certain portions of the tree. All of those blocks will

@@ -4327,7 +4448,7 @@ out:
  * extra overhead is not expressed in terms of disk reads.
  */
 static int __ocfs2_mark_extent_written(struct inode *inode,
-				       struct buffer_head *di_bh,
+				       struct ocfs2_extent_tree *et,
 				       handle_t *handle,
 				       struct ocfs2_path *path,
 				       int split_index,

@@ -4367,10 +4488,9 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 	 */
 	if (path->p_tree_depth) {
 		struct ocfs2_extent_block *eb;
-		struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 
 		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
-				       le64_to_cpu(di->i_last_eb_blk),
+				       ocfs2_get_last_eb_blk(et),
 				       &last_eb_bh, OCFS2_BH_CACHED, inode);
 		if (ret) {
 			mlog_exit(ret);

@@ -4404,7 +4524,7 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 		if (ctxt.c_split_covers_rec)
 			el->l_recs[split_index] = *split_rec;
 		else
-			ret = ocfs2_split_and_insert(inode, handle, path, di_bh,
+			ret = ocfs2_split_and_insert(inode, handle, path, et,
 						     &last_eb_bh, split_index,
 						     split_rec, meta_ac);
 		if (ret)

@@ -4412,7 +4532,7 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 	} else {
 		ret = ocfs2_try_to_merge_extent(inode, handle, path,
 						split_index, split_rec,
-						dealloc, &ctxt);
+						dealloc, &ctxt, et);
 		if (ret)
 			mlog_errno(ret);
 	}

@@ -4430,16 +4550,18 @@ out:
  *
  * The caller is responsible for passing down meta_ac if we'll need it.
  */
-int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *di_bh,
+int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *root_bh,
 			      handle_t *handle, u32 cpos, u32 len, u32 phys,
 			      struct ocfs2_alloc_context *meta_ac,
-			      struct ocfs2_cached_dealloc_ctxt *dealloc)
+			      struct ocfs2_cached_dealloc_ctxt *dealloc,
+			      enum ocfs2_extent_tree_type et_type)
 {
 	int ret, index;
 	u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys);
 	struct ocfs2_extent_rec split_rec;
 	struct ocfs2_path *left_path = NULL;
 	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_tree *et = NULL;
 
 	mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n",
 	     inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno);

@@ -4453,13 +4575,21 @@ int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *di_bh,
 		goto out;
 	}
 
+	et = ocfs2_new_extent_tree(root_bh, et_type);
+	if (!et) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
 	/*
 	 * XXX: This should be fixed up so that we just re-insert the
 	 * next extent records.
 	 */
-	ocfs2_extent_map_trunc(inode, 0);
+	if (et_type == OCFS2_DINODE_EXTENT)
+		ocfs2_extent_map_trunc(inode, 0);
 
-	left_path = ocfs2_new_inode_path(di_bh);
+	left_path = ocfs2_new_path(et->root_bh, et->root_el);
 	if (!left_path) {
 		ret = -ENOMEM;
 		mlog_errno(ret);

@@ -4490,23 +4620,25 @@ int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *di_bh,
 	split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags;
 	split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN;
 
-	ret = __ocfs2_mark_extent_written(inode, di_bh, handle, left_path,
-					  index, &split_rec, meta_ac, dealloc);
+	ret = __ocfs2_mark_extent_written(inode, et, handle, left_path,
+					  index, &split_rec, meta_ac,
+					  dealloc);
 	if (ret)
 		mlog_errno(ret);
 
 out:
 	ocfs2_free_path(left_path);
+	if (et)
+		ocfs2_free_extent_tree(et);
 	return ret;
 }
 
-static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
+static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et,
 			    handle_t *handle, struct ocfs2_path *path,
 			    int index, u32 new_range,
 			    struct ocfs2_alloc_context *meta_ac)
 {
 	int ret, depth, credits = handle->h_buffer_credits;
-	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 	struct buffer_head *last_eb_bh = NULL;
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list *rightmost_el, *el;

@@ -4524,7 +4656,7 @@ static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
 	depth = path->p_tree_depth;
 	if (depth > 0) {
 		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
-				       le64_to_cpu(di->i_last_eb_blk),
+				       ocfs2_get_last_eb_blk(et),
 				       &last_eb_bh, OCFS2_BH_CACHED, inode);
 		if (ret < 0) {
 			mlog_errno(ret);

@@ -4537,7 +4669,7 @@ static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
 		rightmost_el = path_leaf_el(path);
 
 	credits += path->p_tree_depth +
-		   ocfs2_extend_meta_needed(&di->id2.i_list);
+		   ocfs2_extend_meta_needed(et->root_el);
 	ret = ocfs2_extend_trans(handle, credits);
 	if (ret) {
 		mlog_errno(ret);

@@ -4546,7 +4678,7 @@ static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
 
 	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
 	    le16_to_cpu(rightmost_el->l_count)) {
-		ret = ocfs2_grow_tree(inode, handle, di_bh, &depth, &last_eb_bh,
+		ret = ocfs2_grow_tree(inode, handle, et, &depth, &last_eb_bh,
 				      meta_ac);
 		if (ret) {
 			mlog_errno(ret);

@@ -4560,7 +4692,7 @@ static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
 	insert.ins_split = SPLIT_RIGHT;
 	insert.ins_tree_depth = depth;
 
-	ret = ocfs2_do_insert_extent(inode, handle, di_bh, &split_rec, &insert);
+	ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert);
 	if (ret)
 		mlog_errno(ret);
 

@@ -4572,7 +4704,8 @@ out:
 static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle,
 			      struct ocfs2_path *path, int index,
 			      struct ocfs2_cached_dealloc_ctxt *dealloc,
-			      u32 cpos, u32 len)
+			      u32 cpos, u32 len,
+			      struct ocfs2_extent_tree *et)
 {
 	int ret;
 	u32 left_cpos, rec_range, trunc_range;

@@ -4584,7 +4717,7 @@ static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle,
 	struct ocfs2_extent_block *eb;
 
 	if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
-		ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc);
+		ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;

@@ -4715,7 +4848,7 @@ static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle,
 
 	ocfs2_journal_dirty(handle, path_leaf_bh(path));
 
-	ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc);
+	ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;

@@ -4726,20 +4859,29 @@ out:
 	return ret;
 }
 
-int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
+int ocfs2_remove_extent(struct inode *inode, struct buffer_head *root_bh,
 			u32 cpos, u32 len, handle_t *handle,
 			struct ocfs2_alloc_context *meta_ac,
-			struct ocfs2_cached_dealloc_ctxt *dealloc)
+			struct ocfs2_cached_dealloc_ctxt *dealloc,
+			enum ocfs2_extent_tree_type et_type)
 {
 	int ret, index;
 	u32 rec_range, trunc_range;
 	struct ocfs2_extent_rec *rec;
 	struct ocfs2_extent_list *el;
-	struct ocfs2_path *path;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_tree *et = NULL;
+
+	et = ocfs2_new_extent_tree(root_bh, et_type);
+	if (!et) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
 
 	ocfs2_extent_map_trunc(inode, 0);
 
-	path = ocfs2_new_inode_path(di_bh);
+	path = ocfs2_new_path(et->root_bh, et->root_el);
 	if (!path) {
 		ret = -ENOMEM;
 		mlog_errno(ret);

@@ -4792,13 +4934,13 @@ int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
 
 	if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
 		ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
-					 cpos, len);
+					 cpos, len, et);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
 		}
 	} else {
-		ret = ocfs2_split_tree(inode, di_bh, handle, path, index,
+		ret = ocfs2_split_tree(inode, et, handle, path, index,
 				       trunc_range, meta_ac);
 		if (ret) {
 			mlog_errno(ret);

@@ -4847,7 +4989,7 @@ int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
 		}
 
 		ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
-					 cpos, len);
+					 cpos, len, et);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;

@@ -4856,6 +4998,8 @@ int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
 
 out:
 	ocfs2_free_path(path);
+	if (et)
+		ocfs2_free_extent_tree(et);
 	return ret;
 }
 

@@ -6364,7 +6508,8 @@ int ocfs2_convert_inline_data_to_extents(struct inode *inode,
 		 * the in-inode data from our pages.
 		 */
 		ret = ocfs2_insert_extent(osb, handle, inode, di_bh,
-					  0, block, 1, 0, NULL);
+					  0, block, 1, 0,
+					  NULL, OCFS2_DINODE_EXTENT);
 		if (ret) {
 			mlog_errno(ret);
 			goto out_commit;

@@ -6406,13 +6551,14 @@ int ocfs2_commit_truncate(struct ocfs2_super *osb,
 	handle_t *handle = NULL;
 	struct inode *tl_inode = osb->osb_tl_inode;
 	struct ocfs2_path *path = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
 
 	mlog_entry_void();
 
 	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
 						     i_size_read(inode));
 
-	path = ocfs2_new_inode_path(fe_bh);
+	path = ocfs2_new_path(fe_bh, &di->id2.i_list);
 	if (!path) {
 		status = -ENOMEM;
 		mlog_errno(status);

index 249e79e..5a460a9 100644
@@ -26,28 +26,37 @@
 #ifndef OCFS2_ALLOC_H
 #define OCFS2_ALLOC_H
 
+enum ocfs2_extent_tree_type {
+	OCFS2_DINODE_EXTENT = 0,
+};
+
 struct ocfs2_alloc_context;
 int ocfs2_insert_extent(struct ocfs2_super *osb,
 			handle_t *handle,
 			struct inode *inode,
-			struct buffer_head *fe_bh,
+			struct buffer_head *root_bh,
 			u32 cpos,
 			u64 start_blk,
 			u32 new_clusters,
 			u8 flags,
-			struct ocfs2_alloc_context *meta_ac);
+			struct ocfs2_alloc_context *meta_ac,
+			enum ocfs2_extent_tree_type et_type);
 struct ocfs2_cached_dealloc_ctxt;
-int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *di_bh,
+int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *root_bh,
 			      handle_t *handle, u32 cpos, u32 len, u32 phys,
 			      struct ocfs2_alloc_context *meta_ac,
-			      struct ocfs2_cached_dealloc_ctxt *dealloc);
-int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
+			      struct ocfs2_cached_dealloc_ctxt *dealloc,
+			      enum ocfs2_extent_tree_type et_type);
+int ocfs2_remove_extent(struct inode *inode, struct buffer_head *root_bh,
 			u32 cpos, u32 len, handle_t *handle,
 			struct ocfs2_alloc_context *meta_ac,
-			struct ocfs2_cached_dealloc_ctxt *dealloc);
+			struct ocfs2_cached_dealloc_ctxt *dealloc,
+			enum ocfs2_extent_tree_type et_type);
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
 			   struct inode *inode,
-			   struct buffer_head *bh);
+			   struct buffer_head *root_bh,
+			   enum ocfs2_extent_tree_type et_type);
+
 /*
  * how many new metadata chunks would an allocation need at maximum?
  *

index bbe3f8b..44ea5eb 100644
@@ -1278,7 +1278,8 @@ static int ocfs2_write_cluster(struct address_space *mapping,
 	} else if (unwritten) {
 		ret = ocfs2_mark_extent_written(inode, wc->w_di_bh,
 						wc->w_handle, cpos, 1, phys,
-						meta_ac, &wc->w_dealloc);
+						meta_ac, &wc->w_dealloc,
+						OCFS2_DINODE_EXTENT);
 		if (ret < 0) {
 			mlog_errno(ret);
 			goto out;

@@ -1712,7 +1713,13 @@ int ocfs2_write_begin_nolock(struct address_space *mapping,
 		 * ocfs2_lock_allocators(). It greatly over-estimates
 		 * the work to be done.
 		 */
-		ret = ocfs2_lock_allocators(inode, wc->w_di_bh,
+		mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
+		     " clusters_to_add = %u, extents_to_split = %u\n",
+		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+		     (long long)i_size_read(inode), le32_to_cpu(di->i_clusters),
+		     clusters_to_alloc, extents_to_split);
+
+		ret = ocfs2_lock_allocators(inode, wc->w_di_bh, &di->id2.i_list,
 					    clusters_to_alloc, extents_to_split,
 					    &data_ac, &meta_ac);
 		if (ret) {

index 126aa21..ba0fb9e 100644
@@ -1306,7 +1306,7 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 	 * related blocks have been journaled already.
 	 */
 	ret = ocfs2_insert_extent(osb, handle, dir, di_bh, 0, blkno, len, 0,
-				  NULL);
+				  NULL, OCFS2_DINODE_EXTENT);
 	if (ret) {
 		mlog_errno(ret);
 		goto out_commit;

@@ -1338,7 +1338,7 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 		blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
 
 		ret = ocfs2_insert_extent(osb, handle, dir, di_bh, 1, blkno,
-					  len, 0, NULL);
+					  len, 0, NULL, OCFS2_DINODE_EXTENT);
 		if (ret) {
 			mlog_errno(ret);
 			goto out_commit;

@@ -1481,7 +1481,8 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
 	if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 		spin_unlock(&OCFS2_I(dir)->ip_lock);
 		num_free_extents = ocfs2_num_free_extents(osb, dir,
-							  parent_fe_bh);
+							  parent_fe_bh,
+							  OCFS2_DINODE_EXTENT);
 		if (num_free_extents < 0) {
 			status = num_free_extents;
 			mlog_errno(status);

140 lines not on this page · raw message

index c96b805..18e5c80 100644
@@ -57,10 +57,6 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 			       enum ocfs2_alloc_restarted *reason_ret);
 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size,
 			  u64 zero_to);
-int ocfs2_lock_allocators(struct inode *inode, struct buffer_head *fe,
-			  u32 clusters_to_add, u32 extents_to_split,
-			  struct ocfs2_alloc_context **data_ac,
-			  struct ocfs2_alloc_context **meta_ac);
 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr);
 int ocfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
 		  struct kstat *stat);

85 lines not on this page · raw message

index 3f96c87..a3e531e 100644
@@ -162,4 +162,9 @@ u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster);
 int ocfs2_check_group_descriptor(struct super_block *sb,
 				 struct ocfs2_dinode *di,
 				 struct ocfs2_group_desc *gd);
+int ocfs2_lock_allocators(struct inode *inode, struct buffer_head *root_bh,
+			  struct ocfs2_extent_list *root_el,
+			  u32 clusters_to_add, u32 extents_to_split,
+			  struct ocfs2_alloc_context **data_ac,
+			  struct ocfs2_alloc_context **meta_ac);
 #endif /* _CHAINALLOC_H_ */

index 90cefc5..1332309 100644
@@ -4302,6 +4302,116 @@ bail:
 	return status;
 }
 
+/*
+ * Allcate and add clusters into the extent b-tree.
+ * The new clusters(clusters_to_add) will be inserted at logical_offset.
+ * The extent b-tree's root is root_el and it should be in root_bh, and
+ * it is not limited to the file storage. Any extent tree can use this
+ * function if it implements the proper ocfs2_extent_tree.
+ */
+int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb,
+				struct inode *inode,
+				u32 *logical_offset,
+				u32 clusters_to_add,
+				int mark_unwritten,
+				struct buffer_head *root_bh,
+				struct ocfs2_extent_list *root_el,
+				handle_t *handle,
+				struct ocfs2_alloc_context *data_ac,
+				struct ocfs2_alloc_context *meta_ac,
+				enum ocfs2_alloc_restarted *reason_ret,
+				enum ocfs2_extent_tree_type type)
+{
+	int status = 0;
+	int free_extents;
+	enum ocfs2_alloc_restarted reason = RESTART_NONE;
+	u32 bit_off, num_bits;
+	u64 block;
+	u8 flags = 0;
+
+	BUG_ON(!clusters_to_add);
+
+	if (mark_unwritten)
+		flags = OCFS2_EXT_UNWRITTEN;
+
+	free_extents = ocfs2_num_free_extents(osb, inode, root_bh, type);
+	if (free_extents < 0) {
+		status = free_extents;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* there are two cases which could cause us to EAGAIN in the
+	 * we-need-more-metadata case:
+	 * 1) we haven't reserved *any*
+	 * 2) we are so fragmented, we've needed to add metadata too
+	 *    many times. */
+	if (!free_extents && !meta_ac) {
+		mlog(0, "we haven't reserved any metadata!\n");
+		status = -EAGAIN;
+		reason = RESTART_META;
+		goto leave;
+	} else if ((!free_extents)
+		   && (ocfs2_alloc_context_bits_left(meta_ac)
+		       < ocfs2_extend_meta_needed(root_el))) {
+		mlog(0, "filesystem is really fragmented...\n");
+		status = -EAGAIN;
+		reason = RESTART_META;
+		goto leave;
+	}
+
+	status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
+					clusters_to_add, &bit_off, &num_bits);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	BUG_ON(num_bits > clusters_to_add);
+
+	/* reserve our write early -- insert_extent may update the inode */
+	status = ocfs2_journal_access(handle, inode, root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
+	mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
+	     num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
+	status = ocfs2_insert_extent(osb, handle, inode, root_bh,
+				     *logical_offset, block, num_bits,
+				     flags, meta_ac, type);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_journal_dirty(handle, root_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	clusters_to_add -= num_bits;
+	*logical_offset += num_bits;
+
+	if (clusters_to_add) {
+		mlog(0, "need to alloc once more, wanted = %u\n",
+		     clusters_to_add);
+		status = -EAGAIN;
+		reason = RESTART_TRANS;
+	}
+
+leave:
+	mlog_exit(status);
+	if (reason_ret)
+		*reason_ret = reason;
+	return status;
+}
+
 static void ocfs2_make_right_split_rec(struct super_block *sb,
 				       struct ocfs2_extent_rec *split_rec,
 				       u32 cpos,

index 5a460a9..5d51232 100644
@@ -41,6 +41,23 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 			u8 flags,
 			struct ocfs2_alloc_context *meta_ac,
 			enum ocfs2_extent_tree_type et_type);
+enum ocfs2_alloc_restarted {
+	RESTART_NONE = 0,
+	RESTART_TRANS,
+	RESTART_META
+};
+int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb,
+				struct inode *inode,
+				u32 *logical_offset,
+				u32 clusters_to_add,
+				int mark_unwritten,
+				struct buffer_head *root_bh,
+				struct ocfs2_extent_list *root_el,
+				handle_t *handle,
+				struct ocfs2_alloc_context *data_ac,
+				struct ocfs2_alloc_context *meta_ac,
+				enum ocfs2_alloc_restarted *reason_ret,
+				enum ocfs2_extent_tree_type type);
 struct ocfs2_cached_dealloc_ctxt;
 int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *root_bh,
 			      handle_t *handle, u32 cpos, u32 len, u32 phys,

index 44ea5eb..e7acd28 100644
@@ -1255,10 +1255,10 @@ static int ocfs2_write_cluster(struct address_space *mapping,
 		 * any additional semaphores or cluster locks.
 		 */
 		tmp_pos = cpos;
-		ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode,
-						 &tmp_pos, 1, 0, wc->w_di_bh,
-						 wc->w_handle, data_ac,
-						 meta_ac, NULL);
+		ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
+					   &tmp_pos, 1, 0, wc->w_di_bh,
+					   wc->w_handle, data_ac,
+					   meta_ac, NULL);
 		/*
 		 * This shouldn't happen because we must have already
 		 * calculated the correct meta data allocation required. The

index ba0fb9e..d17c34b 100644
@@ -1383,9 +1383,9 @@ static int ocfs2_do_extend_dir(struct super_block *sb,
 	if (extend) {
 		u32 offset = OCFS2_I(dir)->ip_clusters;
 
-		status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
-						    1, 0, parent_fe_bh, handle,
-						    data_ac, meta_ac, NULL);
+		status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
+					      1, 0, parent_fe_bh, handle,
+					      data_ac, meta_ac, NULL);
 		BUG_ON(status == -EAGAIN);
 		if (status < 0) {
 			mlog_errno(status);

157 lines not on this page · raw message

index 18e5c80..e92382c 100644
@@ -33,6 +33,7 @@ extern const struct file_operations ocfs2_dops_no_plocks;
 extern const struct inode_operations ocfs2_file_iops;
 extern const struct inode_operations ocfs2_special_file_iops;
 struct ocfs2_alloc_context;
+enum ocfs2_alloc_restarted;
 
 struct ocfs2_file_private {
 	struct file		*fp_file;

@@ -40,21 +41,16 @@ struct ocfs2_file_private {
 	struct ocfs2_lock_res	fp_flock;
 };
 
-enum ocfs2_alloc_restarted {
-	RESTART_NONE = 0,
-	RESTART_TRANS,
-	RESTART_META
-};
-int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
-			       struct inode *inode,
-			       u32 *logical_offset,
-			       u32 clusters_to_add,
-			       int mark_unwritten,
-			       struct buffer_head *fe_bh,
-			       handle_t *handle,
-			       struct ocfs2_alloc_context *data_ac,
-			       struct ocfs2_alloc_context *meta_ac,
-			       enum ocfs2_alloc_restarted *reason_ret);
+int ocfs2_add_inode_data(struct ocfs2_super *osb,
+			 struct inode *inode,
+			 u32 *logical_offset,
+			 u32 clusters_to_add,
+			 int mark_unwritten,
+			 struct buffer_head *fe_bh,
+			 handle_t *handle,
+			 struct ocfs2_alloc_context *data_ac,
+			 struct ocfs2_alloc_context *meta_ac,
+			 enum ocfs2_alloc_restarted *reason_ret);
 int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size,
 			  u64 zero_to);
 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr);

index d5d808f..2cd6f50 100644
@@ -1598,10 +1598,10 @@ static int ocfs2_symlink(struct inode *dir,
 		u32 offset = 0;
 
 		inode->i_op = &ocfs2_symlink_inode_operations;
-		status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
-						    new_fe_bh,
-						    handle, data_ac, NULL,
-						    NULL);
+		status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
+					      new_fe_bh,
+					      handle, data_ac, NULL,
+					      NULL);
 		if (status < 0) {
 			if (status != -ENOSPC && status != -EINTR) {
 				mlog(ML_ERROR,

index 4f61985..1b46505 100644
@@ -64,6 +64,7 @@
 #define OCFS2_INODE_SIGNATURE		"INODE01"
 #define OCFS2_EXTENT_BLOCK_SIGNATURE	"EXBLK01"
 #define OCFS2_GROUP_DESC_SIGNATURE      "GROUP01"
+#define OCFS2_XATTR_BLOCK_SIGNATURE	"XATTR01"
 
 /* Compatibility flags */
 #define OCFS2_HAS_COMPAT_FEATURE(sb,mask)			\

@@ -715,6 +716,123 @@ struct ocfs2_group_desc
 /*40*/	__u8    bg_bitmap[0];
 };
 
+/*
+ * On disk extended attribute structure for OCFS2.
+ */
+
+/*
+ * ocfs2_xattr_entry indicates one extend attribute.
+ *
+ * Note that it can be stored in inode, one block or one xattr bucket.
+ */
+struct ocfs2_xattr_entry {
+	__le32	xe_name_hash;    /* hash value of xattr prefix+suffix. */
+	__le16	xe_name_offset;  /* byte offset from the 1st etnry in the local
+				    local xattr storage(inode, xattr block or
+				    xattr bucket). */
+	__u8	xe_name_len;	 /* xattr name len, does't include prefix. */
+	__u8	xe_type;         /* the low 7 bits indicates the name prefix's
+				  * type and the highest 1 bits indicate whether
+				  * the EA is stored in the local storage. */
+	__le64	xe_value_size;	 /* real xattr value length. */
+};
+
+/*
+ * On disk structure for xattr header.
+ *
+ * One ocfs2_xattr_header describes how many ocfs2_xattr_entry records in
+ * the local xattr storage.
+ */
+struct ocfs2_xattr_header {
+	__le16	xh_count;                       /* contains the count of how
+						   many records are in the
+						   local xattr storage. */
+	__le16	xh_reserved1;
+	__le32	xh_reserved2;
+	__le64  xh_csum;
+	struct ocfs2_xattr_entry xh_entries[0]; /* xattr entry list. */
+};
+
+/*
+ * On disk structure for xattr value root.
+ *
+ * It is used when one extended attribute's size is larger, and we will save it
+ * in an outside cluster. It will stored in a b-tree like file content.
+ */
+struct ocfs2_xattr_value_root {
+/*00*/	__le32	xr_clusters;              /* clusters covered by xattr value. */
+	__le32	xr_reserved0;
+	__le64	xr_last_eb_blk;           /* Pointer to last extent block */
+/*10*/	struct ocfs2_extent_list xr_list; /* Extent record list */
+};
+
+/*
+ * On disk structure for xattr tree root.
+ *
+ * It is used when there are too many extended attributes for one file. These
+ * attributes will be organized and stored in an indexed-btree.
+ */
+struct ocfs2_xattr_tree_root {
+/*00*/	__le32	xt_clusters;              /* clusters covered by xattr. */
+	__le32	xt_reserved0;
+	__le64	xt_last_eb_blk;           /* Pointer to last extent block */
+/*10*/	struct ocfs2_extent_list xt_list; /* Extent record list */
+};
+
+#define OCFS2_XATTR_INDEXED 0x1
+
+/*
+ * On disk structure for xattr block.
+ */
+struct ocfs2_xattr_block {
+/*00*/	__u8	xb_signature[8];     /* Signature for verification */
+	__le16	xb_suballoc_slot;    /* Slot suballocator this
+					block belongs to. */
+	__le16	xb_suballoc_bit;     /* Bit offset in suballocator
+					block group */
+	__le32	xb_fs_generation;    /* Must match super block */
+/*10*/	__le64	xb_blkno;            /* Offset on disk, in blocks */
+	__le64	xb_csum;
+/*20*/	__le16	xb_flags;            /* Indicates whether this block contains
+					real xattr or a xattr tree. */
+	__le16	xb_reserved0;
+	__le32  xb_reserved1;
+	__le64	xb_reserved2;
+/*30*/	union {
+		struct ocfs2_xattr_header xb_header; /* xattr header if this
+							block contains xattr */
+		struct ocfs2_xattr_tree_root xb_root;/* xattr tree root if this
+							block cotains xattr
+							tree. */
+	} xb_attrs;
+};
+
+#define OCFS2_XATTR_ENTRY_LOCAL		0x80
+#define OCFS2_XATTR_TYPE_MASK		0x7F
+static inline void ocfs2_xattr_set_local(struct ocfs2_xattr_entry *xe,
+					 int local)
+{
+	if (local)
+		xe->xe_type |= OCFS2_XATTR_ENTRY_LOCAL;
+	else
+		xe->xe_type &= ~OCFS2_XATTR_ENTRY_LOCAL;
+}
+
+static inline int ocfs2_xattr_is_local(struct ocfs2_xattr_entry *xe)
+{
+	return xe->xe_type & OCFS2_XATTR_ENTRY_LOCAL;
+}
+
+static inline void ocfs2_xattr_set_type(struct ocfs2_xattr_entry *xe, int type)
+{
+	xe->xe_type |= type & OCFS2_XATTR_TYPE_MASK;
+}
+
+static inline int ocfs2_xattr_get_type(struct ocfs2_xattr_entry *xe)
+{
+	return xe->xe_type & OCFS2_XATTR_TYPE_MASK;
+}
+
 #ifdef __KERNEL__
 static inline int ocfs2_fast_symlink_chars(struct super_block *sb)
 {