Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2: (32 commits)
  [PATCH] ocfs2: zero_user_page conversion
  ocfs2: Support xfs style space reservation ioctls
  ocfs2: support for removing file regions
  ocfs2: update truncate handling of partial clusters
  ocfs2: btree support for removal of arbirtrary extents
  ocfs2: Support creation of unwritten extents
  ocfs2: support writing of unwritten extents
  ocfs2: small cleanup of ocfs2_write_begin_nolock()
  ocfs2: btree changes for unwritten extents
  ocfs2: abstract btree growing calls
  ocfs2: use all extent block suballocators
  ocfs2: plug truncate into cached dealloc routines
  ocfs2: simplify deallocation locking
  ocfs2: harden buffer check during mapping of page blocks
  ocfs2: shared writeable mmap
  ocfs2: factor out write aops into nolock variants
  ocfs2: rework ocfs2_buffered_write_cluster()
  ocfs2: take ip_alloc_sem during entire truncate
  ocfs2: Add "preferred slot" mount option
  [KJ PATCH] Replacing memset(<addr>,0,PAGE_SIZE) with clear_page() in fs/ocfs2/dlm/dlmrecovery.c
  ...

Documentation/filesystems/configfs/configfs.txt

@@ -238,6 +238,8 @@ config_item_type.
 		struct config_group *(*make_group)(struct config_group *group,
 						   const char *name);
 		int (*commit_item)(struct config_item *item);
+		void (*disconnect_notify)(struct config_group *group,
+					  struct config_item *item);
 		void (*drop_item)(struct config_group *group,
 				  struct config_item *item);
 	};
@@ -268,6 +270,16 @@ the item in other threads, the memory is safe.  It may take some time
 for the item to actually disappear from the subsystem's usage.  But it
 is gone from configfs.
 
+When drop_item() is called, the item's linkage has already been torn
+down.  It no longer has a reference on its parent and has no place in
+the item hierarchy.  If a client needs to do some cleanup before this
+teardown happens, the subsystem can implement the
+ct_group_ops->disconnect_notify() method.  The method is called after
+configfs has removed the item from the filesystem view but before the
+item is removed from its parent group.  Like drop_item(),
+disconnect_notify() is void and cannot fail.  Client subsystems should
+not drop any references here, as they still must do it in drop_item().
+
 A config_group cannot be removed while it still has child items.  This
 is implemented in the configfs rmdir(2) code.  ->drop_item() will not be
 called, as the item has not been dropped.  rmdir(2) will fail, as the
@@ -280,18 +292,18 @@ tells configfs to make the subsystem appear in the file tree.
 
 	struct configfs_subsystem {
 		struct config_group	su_group;
-		struct semaphore	su_sem;
+		struct mutex		su_mutex;
 	};
 
 	int configfs_register_subsystem(struct configfs_subsystem *subsys);
 	void configfs_unregister_subsystem(struct configfs_subsystem *subsys);
 
-	A subsystem consists of a toplevel config_group and a semaphore.
+	A subsystem consists of a toplevel config_group and a mutex.
 The group is where child config_items are created.  For a subsystem,
 this group is usually defined statically.  Before calling
 configfs_register_subsystem(), the subsystem must have initialized the
 group via the usual group _init() functions, and it must also have
-initialized the semaphore.
+initialized the mutex.
 	When the register call returns, the subsystem is live, and it
 will be visible via configfs.  At that point, mkdir(2) can be called and
 the subsystem must be ready for it.
@@ -303,7 +315,7 @@ subsystem/group and the simple_child item in configfs_example.c  It
 shows a trivial object displaying and storing an attribute, and a simple
 group creating and destroying these children.
 
-[Hierarchy Navigation and the Subsystem Semaphore]
+[Hierarchy Navigation and the Subsystem Mutex]
 
 There is an extra bonus that configfs provides.  The config_groups and
 config_items are arranged in a hierarchy due to the fact that they
@@ -314,19 +326,19 @@ and config_item->ci_parent structure members.
 
 A subsystem can navigate the cg_children list and the ci_parent pointer
 to see the tree created by the subsystem.  This can race with configfs'
-management of the hierarchy, so configfs uses the subsystem semaphore to
+management of the hierarchy, so configfs uses the subsystem mutex to
 protect modifications.  Whenever a subsystem wants to navigate the
 hierarchy, it must do so under the protection of the subsystem
-semaphore.
+mutex.
 
-A subsystem will be prevented from acquiring the semaphore while a newly
+A subsystem will be prevented from acquiring the mutex while a newly
 allocated item has not been linked into this hierarchy.   Similarly, it
-will not be able to acquire the semaphore while a dropping item has not
+will not be able to acquire the mutex while a dropping item has not
 yet been unlinked.  This means that an item's ci_parent pointer will
 never be NULL while the item is in configfs, and that an item will only
 be in its parent's cg_children list for the same duration.  This allows
 a subsystem to trust ci_parent and cg_children while they hold the
-semaphore.
+mutex.
 
 [Item Aggregation Via symlink(2)]
 
@@ -386,6 +398,33 @@ As a consequence of this, default_groups cannot be removed directly via
 rmdir(2).  They also are not considered when rmdir(2) on the parent
 group is checking for children.
 
+[Dependant Subsystems]
+
+Sometimes other drivers depend on particular configfs items.  For
+example, ocfs2 mounts depend on a heartbeat region item.  If that
+region item is removed with rmdir(2), the ocfs2 mount must BUG or go
+readonly.  Not happy.
+
+configfs provides two additional API calls: configfs_depend_item() and
+configfs_undepend_item().  A client driver can call
+configfs_depend_item() on an existing item to tell configfs that it is
+depended on.  configfs will then return -EBUSY from rmdir(2) for that
+item.  When the item is no longer depended on, the client driver calls
+configfs_undepend_item() on it.
+
+These API cannot be called underneath any configfs callbacks, as
+they will conflict.  They can block and allocate.  A client driver
+probably shouldn't calling them of its own gumption.  Rather it should
+be providing an API that external subsystems call.
+
+How does this work?  Imagine the ocfs2 mount process.  When it mounts,
+it asks for a heartbeat region item.  This is done via a call into the
+heartbeat code.  Inside the heartbeat code, the region item is looked
+up.  Here, the heartbeat code calls configfs_depend_item().  If it
+succeeds, then heartbeat knows the region is safe to give to ocfs2.
+If it fails, it was being torn down anyway, and heartbeat can gracefully
+pass up an error.
+
 [Committable Items]
 
 NOTE: Committable items are currently unimplemented.

Documentation/filesystems/configfs/configfs_example.c

@@ -453,7 +453,7 @@ static int __init configfs_example_init(void)
 		subsys = example_subsys[i];
 
 		config_group_init(&subsys->su_group);
-		init_MUTEX(&subsys->su_sem);
+		mutex_init(&subsys->su_mutex);
 		ret = configfs_register_subsystem(subsys);
 		if (ret) {
 			printk(KERN_ERR "Error %d while registering subsystem %s\n",

fs/configfs/configfs_internal.h

@@ -29,10 +29,11 @@
 
 struct configfs_dirent {
 	atomic_t		s_count;
+	int			s_dependent_count;
 	struct list_head	s_sibling;
 	struct list_head	s_children;
 	struct list_head	s_links;
-	void 			* s_element;
+	void			* s_element;
 	int			s_type;
 	umode_t			s_mode;
 	struct dentry		* s_dentry;
@@ -41,8 +42,8 @@ struct configfs_dirent {
 
 #define CONFIGFS_ROOT		0x0001
 #define CONFIGFS_DIR		0x0002
-#define CONFIGFS_ITEM_ATTR 	0x0004
-#define CONFIGFS_ITEM_LINK 	0x0020
+#define CONFIGFS_ITEM_ATTR	0x0004
+#define CONFIGFS_ITEM_LINK	0x0020
 #define CONFIGFS_USET_DIR	0x0040
 #define CONFIGFS_USET_DEFAULT	0x0080
 #define CONFIGFS_USET_DROPPING	0x0100

fs/configfs/dir.c

@@ -355,6 +355,10 @@ static int configfs_detach_prep(struct dentry *dentry)
 			/* Mark that we've taken i_mutex */
 			sd->s_type |= CONFIGFS_USET_DROPPING;
 
+			/*
+			 * Yup, recursive.  If there's a problem, blame
+			 * deep nesting of default_groups
+			 */
 			ret = configfs_detach_prep(sd->s_dentry);
 			if (!ret)
 				continue;
@@ -562,7 +566,7 @@ static int populate_groups(struct config_group *group)
 
 /*
  * All of link_obj/unlink_obj/link_group/unlink_group require that
- * subsys->su_sem is held.
+ * subsys->su_mutex is held.
  */
 
 static void unlink_obj(struct config_item *item)
@@ -713,6 +717,28 @@ static void configfs_detach_group(struct config_item *item)
 	configfs_detach_item(item);
 }
 
+/*
+ * After the item has been detached from the filesystem view, we are
+ * ready to tear it out of the hierarchy.  Notify the client before
+ * we do that so they can perform any cleanup that requires
+ * navigating the hierarchy.  A client does not need to provide this
+ * callback.  The subsystem semaphore MUST be held by the caller, and
+ * references must be valid for both items.  It also assumes the
+ * caller has validated ci_type.
+ */
+static void client_disconnect_notify(struct config_item *parent_item,
+				     struct config_item *item)
+{
+	struct config_item_type *type;
+
+	type = parent_item->ci_type;
+	BUG_ON(!type);
+
+	if (type->ct_group_ops && type->ct_group_ops->disconnect_notify)
+		type->ct_group_ops->disconnect_notify(to_config_group(parent_item),
+						      item);
+}
+
 /*
  * Drop the initial reference from make_item()/make_group()
  * This function assumes that reference is held on item
@@ -733,11 +759,244 @@ static void client_drop_item(struct config_item *parent_item,
 	 */
 	if (type->ct_group_ops && type->ct_group_ops->drop_item)
 		type->ct_group_ops->drop_item(to_config_group(parent_item),
-						item);
+					      item);
 	else
 		config_item_put(item);
 }
 
+#ifdef DEBUG
+static void configfs_dump_one(struct configfs_dirent *sd, int level)
+{
+	printk(KERN_INFO "%*s\"%s\":\n", level, " ", configfs_get_name(sd));
+
+#define type_print(_type) if (sd->s_type & _type) printk(KERN_INFO "%*s %s\n", level, " ", #_type);
+	type_print(CONFIGFS_ROOT);
+	type_print(CONFIGFS_DIR);
+	type_print(CONFIGFS_ITEM_ATTR);
+	type_print(CONFIGFS_ITEM_LINK);
+	type_print(CONFIGFS_USET_DIR);
+	type_print(CONFIGFS_USET_DEFAULT);
+	type_print(CONFIGFS_USET_DROPPING);
+#undef type_print
+}
+
+static int configfs_dump(struct configfs_dirent *sd, int level)
+{
+	struct configfs_dirent *child_sd;
+	int ret = 0;
+
+	configfs_dump_one(sd, level);
+
+	if (!(sd->s_type & (CONFIGFS_DIR|CONFIGFS_ROOT)))
+		return 0;
+
+	list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
+		ret = configfs_dump(child_sd, level + 2);
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+#endif
+
+
+/*
+ * configfs_depend_item() and configfs_undepend_item()
+ *
+ * WARNING: Do not call these from a configfs callback!
+ *
+ * This describes these functions and their helpers.
+ *
+ * Allow another kernel system to depend on a config_item.  If this
+ * happens, the item cannot go away until the dependant can live without
+ * it.  The idea is to give client modules as simple an interface as
+ * possible.  When a system asks them to depend on an item, they just
+ * call configfs_depend_item().  If the item is live and the client
+ * driver is in good shape, we'll happily do the work for them.
+ *
+ * Why is the locking complex?  Because configfs uses the VFS to handle
+ * all locking, but this function is called outside the normal
+ * VFS->configfs path.  So it must take VFS locks to prevent the
+ * VFS->configfs stuff (configfs_mkdir(), configfs_rmdir(), etc).  This is
+ * why you can't call these functions underneath configfs callbacks.
+ *
+ * Note, btw, that this can be called at *any* time, even when a configfs
+ * subsystem isn't registered, or when configfs is loading or unloading.
+ * Just like configfs_register_subsystem().  So we take the same
+ * precautions.  We pin the filesystem.  We lock each i_mutex _in_order_
+ * on our way down the tree.  If we can find the target item in the
+ * configfs tree, it must be part of the subsystem tree as well, so we
+ * do not need the subsystem semaphore.  Holding the i_mutex chain locks
+ * out mkdir() and rmdir(), who might be racing us.
+ */
+
+/*
+ * configfs_depend_prep()
+ *
+ * Only subdirectories count here.  Files (CONFIGFS_NOT_PINNED) are
+ * attributes.  This is similar but not the same to configfs_detach_prep().
+ * Note that configfs_detach_prep() expects the parent to be locked when it
+ * is called, but we lock the parent *inside* configfs_depend_prep().  We
+ * do that so we can unlock it if we find nothing.
+ *
+ * Here we do a depth-first search of the dentry hierarchy looking for
+ * our object.  We take i_mutex on each step of the way down.  IT IS
+ * ESSENTIAL THAT i_mutex LOCKING IS ORDERED.  If we come back up a branch,
+ * we'll drop the i_mutex.
+ *
+ * If the target is not found, -ENOENT is bubbled up and we have released
+ * all locks.  If the target was found, the locks will be cleared by
+ * configfs_depend_rollback().
+ *
+ * This adds a requirement that all config_items be unique!
+ *
+ * This is recursive because the locking traversal is tricky.  There isn't
+ * much on the stack, though, so folks that need this function - be careful
+ * about your stack!  Patches will be accepted to make it iterative.
+ */
+static int configfs_depend_prep(struct dentry *origin,
+				struct config_item *target)
+{
+	struct configfs_dirent *child_sd, *sd = origin->d_fsdata;
+	int ret = 0;
+
+	BUG_ON(!origin || !sd);
+
+	/* Lock this guy on the way down */
+	mutex_lock(&sd->s_dentry->d_inode->i_mutex);
+	if (sd->s_element == target)  /* Boo-yah */
+		goto out;
+
+	list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
+		if (child_sd->s_type & CONFIGFS_DIR) {
+			ret = configfs_depend_prep(child_sd->s_dentry,
+						   target);
+			if (!ret)
+				goto out;  /* Child path boo-yah */
+		}
+	}
+
+	/* We looped all our children and didn't find target */
+	mutex_unlock(&sd->s_dentry->d_inode->i_mutex);
+	ret = -ENOENT;
+
+out:
+	return ret;
+}
+
+/*
+ * This is ONLY called if configfs_depend_prep() did its job.  So we can
+ * trust the entire path from item back up to origin.
+ *
+ * We walk backwards from item, unlocking each i_mutex.  We finish by
+ * unlocking origin.
+ */
+static void configfs_depend_rollback(struct dentry *origin,
+				     struct config_item *item)
+{
+	struct dentry *dentry = item->ci_dentry;
+
+	while (dentry != origin) {
+		mutex_unlock(&dentry->d_inode->i_mutex);
+		dentry = dentry->d_parent;
+	}
+
+	mutex_unlock(&origin->d_inode->i_mutex);
+}
+
+int configfs_depend_item(struct configfs_subsystem *subsys,
+			 struct config_item *target)
+{
+	int ret;
+	struct configfs_dirent *p, *root_sd, *subsys_sd = NULL;
+	struct config_item *s_item = &subsys->su_group.cg_item;
+
+	/*
+	 * Pin the configfs filesystem.  This means we can safely access
+	 * the root of the configfs filesystem.
+	 */
+	ret = configfs_pin_fs();
+	if (ret)
+		return ret;
+
+	/*
+	 * Next, lock the root directory.  We're going to check that the
+	 * subsystem is really registered, and so we need to lock out
+	 * configfs_[un]register_subsystem().
+	 */
+	mutex_lock(&configfs_sb->s_root->d_inode->i_mutex);
+
+	root_sd = configfs_sb->s_root->d_fsdata;
+
+	list_for_each_entry(p, &root_sd->s_children, s_sibling) {
+		if (p->s_type & CONFIGFS_DIR) {
+			if (p->s_element == s_item) {
+				subsys_sd = p;
+				break;
+			}
+		}
+	}
+
+	if (!subsys_sd) {
+		ret = -ENOENT;
+		goto out_unlock_fs;
+	}
+
+	/* Ok, now we can trust subsys/s_item */
+
+	/* Scan the tree, locking i_mutex recursively, return 0 if found */
+	ret = configfs_depend_prep(subsys_sd->s_dentry, target);
+	if (ret)
+		goto out_unlock_fs;
+
+	/* We hold all i_mutexes from the subsystem down to the target */
+	p = target->ci_dentry->d_fsdata;
+	p->s_dependent_count += 1;
+
+	configfs_depend_rollback(subsys_sd->s_dentry, target);
+
+out_unlock_fs:
+	mutex_unlock(&configfs_sb->s_root->d_inode->i_mutex);
+
+	/*
+	 * If we succeeded, the fs is pinned via other methods.  If not,
+	 * we're done with it anyway.  So release_fs() is always right.
+	 */
+	configfs_release_fs();
+
+	return ret;
+}
+EXPORT_SYMBOL(configfs_depend_item);
+
+/*
+ * Release the dependent linkage.  This is much simpler than
+ * configfs_depend_item() because we know that that the client driver is
+ * pinned, thus the subsystem is pinned, and therefore configfs is pinned.
+ */
+void configfs_undepend_item(struct configfs_subsystem *subsys,
+			    struct config_item *target)
+{
+	struct configfs_dirent *sd;
+
+	/*
+	 * Since we can trust everything is pinned, we just need i_mutex
+	 * on the item.
+	 */
+	mutex_lock(&target->ci_dentry->d_inode->i_mutex);
+
+	sd = target->ci_dentry->d_fsdata;
+	BUG_ON(sd->s_dependent_count < 1);
+
+	sd->s_dependent_count -= 1;
+
+	/*
+	 * After this unlock, we cannot trust the item to stay alive!
+	 * DO NOT REFERENCE item after this unlock.
+	 */
+	mutex_unlock(&target->ci_dentry->d_inode->i_mutex);
+}
+EXPORT_SYMBOL(configfs_undepend_item);
 
 static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 {
@@ -783,7 +1042,7 @@ static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 
 	snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
 
-	down(&subsys->su_sem);
+	mutex_lock(&subsys->su_mutex);
 	group = NULL;
 	item = NULL;
 	if (type->ct_group_ops->make_group) {
@@ -797,7 +1056,7 @@ static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 		if (item)
 			link_obj(parent_item, item);
 	}
-	up(&subsys->su_sem);
+	mutex_unlock(&subsys->su_mutex);
 
 	kfree(name);
 	if (!item) {
@@ -841,13 +1100,16 @@ static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 out_unlink:
 	if (ret) {
 		/* Tear down everything we built up */
-		down(&subsys->su_sem);
+		mutex_lock(&subsys->su_mutex);
+
+		client_disconnect_notify(parent_item, item);
 		if (group)
 			unlink_group(group);
 		else
 			unlink_obj(item);
 		client_drop_item(parent_item, item);
-		up(&subsys->su_sem);
+
+		mutex_unlock(&subsys->su_mutex);
 
 		if (module_got)
 			module_put(owner);
@@ -881,6 +1143,13 @@ static int configfs_rmdir(struct inode *dir, struct dentry *dentry)
 	if (sd->s_type & CONFIGFS_USET_DEFAULT)
 		return -EPERM;
 
+	/*
+	 * Here's where we check for dependents.  We're protected by
+	 * i_mutex.
+	 */
+	if (sd->s_dependent_count)
+		return -EBUSY;
+
 	/* Get a working ref until we have the child */
 	parent_item = configfs_get_config_item(dentry->d_parent);
 	subsys = to_config_group(parent_item)->cg_subsys;
@@ -910,17 +1179,19 @@ static int configfs_rmdir(struct inode *dir, struct dentry *dentry)
 	if (sd->s_type & CONFIGFS_USET_DIR) {
 		configfs_detach_group(item);
 
-		down(&subsys->su_sem);
+		mutex_lock(&subsys->su_mutex);
+		client_disconnect_notify(parent_item, item);
 		unlink_group(to_config_group(item));
 	} else {
 		configfs_detach_item(item);
 
-		down(&subsys->su_sem);
+		mutex_lock(&subsys->su_mutex);
+		client_disconnect_notify(parent_item, item);
 		unlink_obj(item);
 	}
 
 	client_drop_item(parent_item, item);
-	up(&subsys->su_sem);
+	mutex_unlock(&subsys->su_mutex);
 
 	/* Drop our reference from above */
 	config_item_put(item);

fs/configfs/file.c

@@ -27,19 +27,26 @@
 #include <linux/fs.h>
 #include <linux/module.h>
 #include <linux/slab.h>
+#include <linux/mutex.h>
 #include <asm/uaccess.h>
-#include <asm/semaphore.h>
 
 #include <linux/configfs.h>
 #include "configfs_internal.h"
 
+/*
+ * A simple attribute can only be 4096 characters.  Why 4k?  Because the
+ * original code limited it to PAGE_SIZE.  That's a bad idea, though,
+ * because an attribute of 16k on ia64 won't work on x86.  So we limit to
+ * 4k, our minimum common page size.
+ */
+#define SIMPLE_ATTR_SIZE 4096
 
 struct configfs_buffer {
 	size_t			count;
 	loff_t			pos;
 	char			* page;
 	struct configfs_item_operations	* ops;
-	struct semaphore	sem;
+	struct mutex		mutex;
 	int			needs_read_fill;
 };
 
@@ -69,7 +76,7 @@ static int fill_read_buffer(struct dentry * dentry, struct configfs_buffer * buf
 
 	count = ops->show_attribute(item,attr,buffer->page);
 	buffer->needs_read_fill = 0;
-	BUG_ON(count > (ssize_t)PAGE_SIZE);
+	BUG_ON(count > (ssize_t)SIMPLE_ATTR_SIZE);
 	if (count >= 0)
 		buffer->count = count;
 	else
@@ -102,7 +109,7 @@ configfs_read_file(struct file *file, char __user *buf, size_t count, loff_t *pp
 	struct configfs_buffer * buffer = file->private_data;
 	ssize_t retval = 0;
 
-	down(&buffer->sem);
+	mutex_lock(&buffer->mutex);
 	if (buffer->needs_read_fill) {
 		if ((retval = fill_read_buffer(file->f_path.dentry,buffer)))
 			goto out;
@@ -112,7 +119,7 @@ configfs_read_file(struct file *file, char __user *buf, size_t count, loff_t *pp
 	retval = simple_read_from_buffer(buf, count, ppos, buffer->page,
 					 buffer->count);
 out:
-	up(&buffer->sem);
+	mutex_unlock(&buffer->mutex);
 	return retval;
 }
 
@@ -137,8 +144,8 @@ fill_write_buffer(struct configfs_buffer * buffer, const char __user * buf, size
 	if (!buffer->page)
 		return -ENOMEM;
 
-	if (count >= PAGE_SIZE)
-		count = PAGE_SIZE - 1;
+	if (count >= SIMPLE_ATTR_SIZE)
+		count = SIMPLE_ATTR_SIZE - 1;
 	error = copy_from_user(buffer->page,buf,count);
 	buffer->needs_read_fill = 1;
 	/* if buf is assumed to contain a string, terminate it by \0,
@@ -193,13 +200,13 @@ configfs_write_file(struct file *file, const char __user *buf, size_t count, lof
 	struct configfs_buffer * buffer = file->private_data;
 	ssize_t len;
 
-	down(&buffer->sem);
+	mutex_lock(&buffer->mutex);
 	len = fill_write_buffer(buffer, buf, count);
 	if (len > 0)
 		len = flush_write_buffer(file->f_path.dentry, buffer, count);
 	if (len > 0)
 		*ppos += len;
-	up(&buffer->sem);
+	mutex_unlock(&buffer->mutex);
 	return len;
 }
 
@@ -253,7 +260,7 @@ static int check_perm(struct inode * inode, struct file * file)
 		error = -ENOMEM;
 		goto Enomem;
 	}
-	init_MUTEX(&buffer->sem);
+	mutex_init(&buffer->mutex);
 	buffer->needs_read_fill = 1;
 	buffer->ops = ops;
 	file->private_data = buffer;
@@ -292,6 +299,7 @@ static int configfs_release(struct inode * inode, struct file * filp)
 	if (buffer) {
 		if (buffer->page)
 			free_page((unsigned long)buffer->page);
+		mutex_destroy(&buffer->mutex);
 		kfree(buffer);
 	}
 	return 0;

fs/configfs/item.c

@@ -62,7 +62,6 @@ void config_item_init(struct config_item * item)
  *	dynamically allocated string that @item->ci_name points to.
  *	Otherwise, use the static @item->ci_namebuf array.
  */
-
 int config_item_set_name(struct config_item * item, const char * fmt, ...)
 {
 	int error = 0;
@@ -139,12 +138,7 @@ struct config_item * config_item_get(struct config_item * item)
 	return item;
 }
 
-/**
- *	config_item_cleanup - free config_item resources.
- *	@item:	item.
- */
-
-void config_item_cleanup(struct config_item * item)
+static void config_item_cleanup(struct config_item * item)
 {
 	struct config_item_type * t = item->ci_type;
 	struct config_group * s = item->ci_group;
@@ -179,39 +173,35 @@ void config_item_put(struct config_item * item)
 		kref_put(&item->ci_kref, config_item_release);
 }
 
-
 /**
  *	config_group_init - initialize a group for use
  *	@k:	group
  */
-
 void config_group_init(struct config_group *group)
 {
 	config_item_init(&group->cg_item);
 	INIT_LIST_HEAD(&group->cg_children);
 }
 
-
 /**
- *	config_group_find_obj - search for item in group.
+ *	config_group_find_item - search for item in group.
  *	@group:	group we're looking in.
  *	@name:	item's name.
  *
- *	Lock group via @group->cg_subsys, and iterate over @group->cg_list,
- *	looking for a matching config_item. If matching item is found
- *	take a reference and return the item.
+ *	Iterate over @group->cg_list, looking for a matching config_item.
+ *	If matching item is found take a reference and return the item.
+ *	Caller must have locked group via @group->cg_subsys->su_mtx.
  */
-
-struct config_item * config_group_find_obj(struct config_group * group, const char * name)
+struct config_item *config_group_find_item(struct config_group *group,
+					   const char *name)
 {
 	struct list_head * entry;
 	struct config_item * ret = NULL;
 
-        /* XXX LOCKING! */
 	list_for_each(entry,&group->cg_children) {
 		struct config_item * item = to_item(entry);
 		if (config_item_name(item) &&
-                    !strcmp(config_item_name(item), name)) {
+		    !strcmp(config_item_name(item), name)) {
 			ret = config_item_get(item);
 			break;
 		}
@@ -219,9 +209,8 @@ struct config_item * config_group_find_obj(struct config_group * group, const ch
 	return ret;
 }
 
-
 EXPORT_SYMBOL(config_item_init);
 EXPORT_SYMBOL(config_group_init);
 EXPORT_SYMBOL(config_item_get);
 EXPORT_SYMBOL(config_item_put);
-EXPORT_SYMBOL(config_group_find_obj);
+EXPORT_SYMBOL(config_group_find_item);

fs/dlm/config.c

@@ -133,14 +133,6 @@ static ssize_t cluster_set(struct cluster *cl, unsigned int *cl_field,
 	return len;
 }
 
-#define __CONFIGFS_ATTR(_name,_mode,_read,_write) {                           \
-	.attr   = { .ca_name = __stringify(_name),                            \
-		    .ca_mode = _mode,                                         \
-		    .ca_owner = THIS_MODULE },                                \
-	.show   = _read,                                                      \
-	.store  = _write,                                                     \
-}
-
 #define CLUSTER_ATTR(name, check_zero)                                        \
 static ssize_t name##_write(struct cluster *cl, const char *buf, size_t len)  \
 {                                                                             \
@@ -615,7 +607,7 @@ static struct clusters clusters_root = {
 int dlm_config_init(void)
 {
 	config_group_init(&clusters_root.subsys.su_group);
-	init_MUTEX(&clusters_root.subsys.su_sem);
+	mutex_init(&clusters_root.subsys.su_mutex);
 	return configfs_register_subsystem(&clusters_root.subsys);
 }
 
@@ -759,9 +751,9 @@ static struct space *get_space(char *name)
 	if (!space_list)
 		return NULL;
 
-	down(&space_list->cg_subsys->su_sem);
-	i = config_group_find_obj(space_list, name);
-	up(&space_list->cg_subsys->su_sem);
+	mutex_lock(&space_list->cg_subsys->su_mutex);
+	i = config_group_find_item(space_list, name);
+	mutex_unlock(&space_list->cg_subsys->su_mutex);
 
 	return to_space(i);
 }
@@ -780,7 +772,7 @@ static struct comm *get_comm(int nodeid, struct sockaddr_storage *addr)
 	if (!comm_list)
 		return NULL;
 
-	down(&clusters_root.subsys.su_sem);
+	mutex_lock(&clusters_root.subsys.su_mutex);
 
 	list_for_each_entry(i, &comm_list->cg_children, ci_entry) {
 		cm = to_comm(i);
@@ -800,7 +792,7 @@ static struct comm *get_comm(int nodeid, struct sockaddr_storage *addr)
 			break;
 		}
 	}
-	up(&clusters_root.subsys.su_sem);
+	mutex_unlock(&clusters_root.subsys.su_mutex);
 
 	if (!found)
 		cm = NULL;

fs/ocfs2/alloc.c

@@ -50,6 +50,8 @@
 #include "buffer_head_io.h"
 
 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);
 
 /*
  * Structures which describe a path through a btree, and functions to
@@ -116,6 +118,31 @@ static void ocfs2_free_path(struct ocfs2_path *path)
 	}
 }
 
+/*
+ * All the elements of src into dest. After this call, src could be freed
+ * without affecting dest.
+ *
+ * Both paths should have the same root. Any non-root elements of dest
+ * will be freed.
+ */
+static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src)
+{
+	int i;
+
+	BUG_ON(path_root_bh(dest) != path_root_bh(src));
+	BUG_ON(path_root_el(dest) != path_root_el(src));
+
+	ocfs2_reinit_path(dest, 1);
+
+	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
+		dest->p_node[i].bh = src->p_node[i].bh;
+		dest->p_node[i].el = src->p_node[i].el;
+
+		if (dest->p_node[i].bh)
+			get_bh(dest->p_node[i].bh);
+	}
+}
+
 /*
  * Make the *dest path the same as src and re-initialize src path to
  * have a root only.
@@ -212,10 +239,41 @@ out:
 	return ret;
 }
 
+/*
+ * Return the index of the extent record which contains cluster #v_cluster.
+ * -1 is returned if it was not found.
+ *
+ * Should work fine on interior and exterior nodes.
+ */
+int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
+{
+	int ret = -1;
+	int i;
+	struct ocfs2_extent_rec *rec;
+	u32 rec_end, rec_start, clusters;
+
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
+
+		rec_start = le32_to_cpu(rec->e_cpos);
+		clusters = ocfs2_rec_clusters(el, rec);
+
+		rec_end = rec_start + clusters;
+
+		if (v_cluster >= rec_start && v_cluster < rec_end) {
+			ret = i;
+			break;
+		}
+	}
+
+	return ret;
+}
+
 enum ocfs2_contig_type {
 	CONTIG_NONE = 0,
 	CONTIG_LEFT,
-	CONTIG_RIGHT
+	CONTIG_RIGHT,
+	CONTIG_LEFTRIGHT,
 };
 
 
@@ -253,6 +311,14 @@ static enum ocfs2_contig_type
 {
 	u64 blkno = le64_to_cpu(insert_rec->e_blkno);
 
+	/*
+	 * Refuse to coalesce extent records with different flag
+	 * fields - we don't want to mix unwritten extents with user
+	 * data.
+	 */
+	if (ext->e_flags != insert_rec->e_flags)
+		return CONTIG_NONE;
+
 	if (ocfs2_extents_adjacent(ext, insert_rec) &&
 	    ocfs2_block_extent_contig(inode->i_sb, ext, blkno))
 			return CONTIG_RIGHT;
@@ -277,7 +343,14 @@ enum ocfs2_append_type {
 	APPEND_TAIL,
 };
 
+enum ocfs2_split_type {
+	SPLIT_NONE = 0,
+	SPLIT_LEFT,
+	SPLIT_RIGHT,
+};
+
 struct ocfs2_insert_type {
+	enum ocfs2_split_type	ins_split;
 	enum ocfs2_append_type	ins_appending;
 	enum ocfs2_contig_type	ins_contig;
 	int			ins_contig_index;
@@ -285,6 +358,13 @@ struct ocfs2_insert_type {
 	int			ins_tree_depth;
 };
 
+struct ocfs2_merge_ctxt {
+	enum ocfs2_contig_type	c_contig_type;
+	int			c_has_empty_extent;
+	int			c_split_covers_rec;
+	int			c_used_tail_recs;
+};
+
 /*
  * How many free extents have we got before we need more meta data?
  */
@@ -384,13 +464,7 @@ static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb,
 			strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
 			eb->h_blkno = cpu_to_le64(first_blkno);
 			eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
-
-#ifndef OCFS2_USE_ALL_METADATA_SUBALLOCATORS
-			/* we always use slot zero's suballocator */
-			eb->h_suballoc_slot = 0;
-#else
 			eb->h_suballoc_slot = cpu_to_le16(osb->slot_num);
-#endif
 			eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
 			eb->h_list.l_count =
 				cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
@@ -461,7 +535,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 			    struct inode *inode,
 			    struct buffer_head *fe_bh,
 			    struct buffer_head *eb_bh,
-			    struct buffer_head *last_eb_bh,
+			    struct buffer_head **last_eb_bh,
 			    struct ocfs2_alloc_context *meta_ac)
 {
 	int status, new_blocks, i;
@@ -476,7 +550,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 
 	mlog_entry_void();
 
-	BUG_ON(!last_eb_bh);
+	BUG_ON(!last_eb_bh || !*last_eb_bh);
 
 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
@@ -507,7 +581,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	eb = (struct ocfs2_extent_block *)last_eb_bh->b_data;
+	eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
 	new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
 
 	/* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
@@ -568,7 +642,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 	 * journal_dirty erroring as it won't unless we've aborted the
 	 * handle (in which case we would never be here) so reserving
 	 * the write with journal_access is all we need to do. */
-	status = ocfs2_journal_access(handle, inode, last_eb_bh,
+	status = ocfs2_journal_access(handle, inode, *last_eb_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -601,10 +675,10 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 	 * next_leaf on the previously last-extent-block. */
 	fe->i_last_eb_blk = cpu_to_le64(new_last_eb_blk);
 
-	eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
 	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
 
-	status = ocfs2_journal_dirty(handle, last_eb_bh);
+	status = ocfs2_journal_dirty(handle, *last_eb_bh);
 	if (status < 0)
 		mlog_errno(status);
 	status = ocfs2_journal_dirty(handle, fe_bh);
@@ -616,6 +690,14 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 			mlog_errno(status);
 	}
 
+	/*
+	 * Some callers want to track the rightmost leaf so pass it
+	 * back here.
+	 */
+	brelse(*last_eb_bh);
+	get_bh(new_eb_bhs[0]);
+	*last_eb_bh = new_eb_bhs[0];
+
 	status = 0;
 bail:
 	if (new_eb_bhs) {
@@ -828,6 +910,87 @@ bail:
 	return status;
 }
 
+/*
+ * Grow a b-tree so that it has more records.
+ *
+ * We might shift the tree depth in which case existing paths should
+ * be considered invalid.
+ *
+ * Tree depth after the grow is returned via *final_depth.
+ *
+ * *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 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_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);
+	if (shift < 0) {
+		ret = shift;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* We traveled all the way to the bottom of the allocation tree
+	 * and didn't find room for any more extents - we need to add
+	 * another tree level */
+	if (shift) {
+		BUG_ON(bh);
+		mlog(0, "need to shift tree depth (current = %d)\n", depth);
+
+		/* 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,
+					     meta_ac, &bh);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+		depth++;
+		if (depth == 1) {
+			/*
+			 * Special case: we have room now if we shifted from
+			 * tree_depth 0, so no more work needs to be done.
+			 *
+			 * We won't be calling add_branch, so pass
+			 * back *last_eb_bh as the new leaf. At depth
+			 * zero, it should always be null so there's
+			 * no reason to brelse.
+			 */
+			BUG_ON(*last_eb_bh);
+			get_bh(bh);
+			*last_eb_bh = bh;
+			goto out;
+		}
+	}
+
+	/* 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,
+			       meta_ac);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out:
+	if (final_depth)
+		*final_depth = depth;
+	brelse(bh);
+	return ret;
+}
+
 /*
  * This is only valid for leaf nodes, which are the only ones that can
  * have empty extents anyway.
@@ -934,6 +1097,22 @@ static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
 
 }
 
+static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
+{
+	int size, num_recs = le16_to_cpu(el->l_next_free_rec);
+
+	BUG_ON(num_recs == 0);
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
+		num_recs--;
+		size = num_recs * sizeof(struct ocfs2_extent_rec);
+		memmove(&el->l_recs[0], &el->l_recs[1], size);
+		memset(&el->l_recs[num_recs], 0,
+		       sizeof(struct ocfs2_extent_rec));
+		el->l_next_free_rec = cpu_to_le16(num_recs);
+	}
+}
+
 /*
  * Create an empty extent record .
  *
@@ -1211,6 +1390,10 @@ static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
 	 * immediately to their right.
 	 */
 	left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
+	if (ocfs2_is_empty_extent(&right_child_el->l_recs[0])) {
+		BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
+		left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
+	}
 	left_clusters -= le32_to_cpu(left_rec->e_cpos);
 	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
 
@@ -1531,10 +1714,16 @@ out:
 	return ret;
 }
 
+/*
+ * Extend the transaction by enough credits to complete the rotation,
+ * and still leave at least the original number of credits allocated
+ * to this transaction.
+ */
 static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
+					   int op_credits,
 					   struct ocfs2_path *path)
 {
-	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1;
+	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
 
 	if (handle->h_buffer_credits < credits)
 		return ocfs2_extend_trans(handle, credits);
@@ -1568,6 +1757,29 @@ static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
 	return 0;
 }
 
+static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
+{
+	int next_free = le16_to_cpu(el->l_next_free_rec);
+	unsigned int range;
+	struct ocfs2_extent_rec *rec;
+
+	if (next_free == 0)
+		return 0;
+
+	rec = &el->l_recs[0];
+	if (ocfs2_is_empty_extent(rec)) {
+		/* Empty list. */
+		if (next_free == 1)
+			return 0;
+		rec = &el->l_recs[1];
+	}
+
+	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+	if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
+		return 1;
+	return 0;
+}
+
 /*
  * Rotate all the records in a btree right one record, starting at insert_cpos.
  *
@@ -1586,11 +1798,12 @@ static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
  */
 static int ocfs2_rotate_tree_right(struct inode *inode,
 				   handle_t *handle,
+				   enum ocfs2_split_type split,
 				   u32 insert_cpos,
 				   struct ocfs2_path *right_path,
 				   struct ocfs2_path **ret_left_path)
 {
-	int ret, start;
+	int ret, start, orig_credits = handle->h_buffer_credits;
 	u32 cpos;
 	struct ocfs2_path *left_path = NULL;
 
@@ -1657,9 +1870,9 @@ static int ocfs2_rotate_tree_right(struct inode *inode,
 				(unsigned long long)
 				path_leaf_bh(left_path)->b_blocknr);
 
-		if (ocfs2_rotate_requires_path_adjustment(left_path,
+		if (split == SPLIT_NONE &&
+		    ocfs2_rotate_requires_path_adjustment(left_path,
 							  insert_cpos)) {
-			mlog(0, "Path adjustment required\n");
 
 			/*
 			 * We've rotated the tree as much as we
@@ -1687,7 +1900,7 @@ static int ocfs2_rotate_tree_right(struct inode *inode,
 		     right_path->p_tree_depth);
 
 		ret = ocfs2_extend_rotate_transaction(handle, start,
-						      right_path);
+						      orig_credits, right_path);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
@@ -1700,6 +1913,24 @@ static int ocfs2_rotate_tree_right(struct inode *inode,
 			goto out;
 		}
 
+		if (split != SPLIT_NONE &&
+		    ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
+						insert_cpos)) {
+			/*
+			 * A rotate moves the rightmost left leaf
+			 * record over to the leftmost right leaf
+			 * slot. If we're doing an extent split
+			 * instead of a real insert, then we have to
+			 * check that the extent to be split wasn't
+			 * just moved over. If it was, then we can
+			 * exit here, passing left_path back -
+			 * ocfs2_split_extent() is smart enough to
+			 * search both leaves.
+			 */
+			*ret_left_path = left_path;
+			goto out_ret_path;
+		}
+
 		/*
 		 * There is no need to re-read the next right path
 		 * as we know that it'll be our current left
@@ -1722,557 +1953,1725 @@ out_ret_path:
 	return ret;
 }
 
-/*
- * Do the final bits of extent record insertion at the target leaf
- * list. If this leaf is part of an allocation tree, it is assumed
- * that the tree above has been prepared.
- */
-static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec,
-				 struct ocfs2_extent_list *el,
-				 struct ocfs2_insert_type *insert,
-				 struct inode *inode)
+static void ocfs2_update_edge_lengths(struct inode *inode, handle_t *handle,
+				      struct ocfs2_path *path)
 {
-	int i = insert->ins_contig_index;
-	unsigned int range;
+	int i, idx;
 	struct ocfs2_extent_rec *rec;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_block *eb;
+	u32 range;
 
-	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+	/* Path should always be rightmost. */
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
+	BUG_ON(eb->h_next_leaf_blk != 0ULL);
 
-	/*
-	 * Contiguous insert - either left or right.
-	 */
-	if (insert->ins_contig != CONTIG_NONE) {
-		rec = &el->l_recs[i];
-		if (insert->ins_contig == CONTIG_LEFT) {
-			rec->e_blkno = insert_rec->e_blkno;
-			rec->e_cpos = insert_rec->e_cpos;
-		}
-		le16_add_cpu(&rec->e_leaf_clusters,
-			     le16_to_cpu(insert_rec->e_leaf_clusters));
-		return;
-	}
+	el = &eb->h_list;
+	BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
+	idx = le16_to_cpu(el->l_next_free_rec) - 1;
+	rec = &el->l_recs[idx];
+	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
 
-	/*
-	 * Handle insert into an empty leaf.
-	 */
-	if (le16_to_cpu(el->l_next_free_rec) == 0 ||
-	    ((le16_to_cpu(el->l_next_free_rec) == 1) &&
-	     ocfs2_is_empty_extent(&el->l_recs[0]))) {
-		el->l_recs[0] = *insert_rec;
-		el->l_next_free_rec = cpu_to_le16(1);
-		return;
-	}
+	for (i = 0; i < path->p_tree_depth; i++) {
+		el = path->p_node[i].el;
+		idx = le16_to_cpu(el->l_next_free_rec) - 1;
+		rec = &el->l_recs[idx];
 
-	/*
-	 * Appending insert.
-	 */
-	if (insert->ins_appending == APPEND_TAIL) {
-		i = le16_to_cpu(el->l_next_free_rec) - 1;
-		rec = &el->l_recs[i];
-		range = le32_to_cpu(rec->e_cpos)
-			+ le16_to_cpu(rec->e_leaf_clusters);
-		BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
+		rec->e_int_clusters = cpu_to_le32(range);
+		le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
 
-		mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
-				le16_to_cpu(el->l_count),
-				"inode %lu, depth %u, count %u, next free %u, "
-				"rec.cpos %u, rec.clusters %u, "
-				"insert.cpos %u, insert.clusters %u\n",
-				inode->i_ino,
-				le16_to_cpu(el->l_tree_depth),
-				le16_to_cpu(el->l_count),
-				le16_to_cpu(el->l_next_free_rec),
-				le32_to_cpu(el->l_recs[i].e_cpos),
-				le16_to_cpu(el->l_recs[i].e_leaf_clusters),
-				le32_to_cpu(insert_rec->e_cpos),
-				le16_to_cpu(insert_rec->e_leaf_clusters));
-		i++;
-		el->l_recs[i] = *insert_rec;
-		le16_add_cpu(&el->l_next_free_rec, 1);
-		return;
+		ocfs2_journal_dirty(handle, path->p_node[i].bh);
 	}
-
-	/*
-	 * Ok, we have to rotate.
-	 *
-	 * At this point, it is safe to assume that inserting into an
-	 * empty leaf and appending to a leaf have both been handled
-	 * above.
-	 *
-	 * This leaf needs to have space, either by the empty 1st
-	 * extent record, or by virtue of an l_next_rec < l_count.
-	 */
-	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 int ocfs2_append_rec_to_path(struct inode *inode, handle_t *handle,
-				    struct ocfs2_extent_rec *insert_rec,
-				    struct ocfs2_path *right_path,
-				    struct ocfs2_path **ret_left_path)
+static void ocfs2_unlink_path(struct inode *inode, handle_t *handle,
+			      struct ocfs2_cached_dealloc_ctxt *dealloc,
+			      struct ocfs2_path *path, int unlink_start)
 {
-	int ret, i, next_free;
-	struct buffer_head *bh;
+	int ret, i;
+	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list *el;
-	struct ocfs2_path *left_path = NULL;
+	struct buffer_head *bh;
 
-	*ret_left_path = NULL;
+	for(i = unlink_start; i < path_num_items(path); i++) {
+		bh = path->p_node[i].bh;
 
-	/*
-	 * This shouldn't happen for non-trees. The extent rec cluster
-	 * count manipulation below only works for interior nodes.
-	 */
-	BUG_ON(right_path->p_tree_depth == 0);
+		eb = (struct ocfs2_extent_block *)bh->b_data;
+		/*
+		 * Not all nodes might have had their final count
+		 * decremented by the caller - handle this here.
+		 */
+		el = &eb->h_list;
+		if (le16_to_cpu(el->l_next_free_rec) > 1) {
+			mlog(ML_ERROR,
+			     "Inode %llu, attempted to remove extent block "
+			     "%llu with %u records\n",
+			     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			     (unsigned long long)le64_to_cpu(eb->h_blkno),
+			     le16_to_cpu(el->l_next_free_rec));
+
+			ocfs2_journal_dirty(handle, bh);
+			ocfs2_remove_from_cache(inode, bh);
+			continue;
+		}
 
-	/*
-	 * If our appending insert is at the leftmost edge of a leaf,
-	 * then we might need to update the rightmost records of the
-	 * neighboring path.
-	 */
-	el = path_leaf_el(right_path);
-	next_free = le16_to_cpu(el->l_next_free_rec);
-	if (next_free == 0 ||
-	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
-		u32 left_cpos;
+		el->l_next_free_rec = 0;
+		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
 
-		ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path,
-						    &left_cpos);
-		if (ret) {
+		ocfs2_journal_dirty(handle, bh);
+
+		ret = ocfs2_cache_extent_block_free(dealloc, eb);
+		if (ret)
 			mlog_errno(ret);
-			goto out;
-		}
 
-		mlog(0, "Append may need a left path update. cpos: %u, "
-		     "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos),
-		     left_cpos);
+		ocfs2_remove_from_cache(inode, bh);
+	}
+}
 
-		/*
-		 * No need to worry if the append is already in the
-		 * leftmost leaf.
+static void ocfs2_unlink_subtree(struct inode *inode, handle_t *handle,
+				 struct ocfs2_path *left_path,
+				 struct ocfs2_path *right_path,
+				 int subtree_index,
+				 struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int i;
+	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
+	struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_block *eb;
+
+	el = path_leaf_el(left_path);
+
+	eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
+
+	for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
+		if (root_el->l_recs[i].e_blkno == eb->h_blkno)
+			break;
+
+	BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
+
+	memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
+	le16_add_cpu(&root_el->l_next_free_rec, -1);
+
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
+	eb->h_next_leaf_blk = 0;
+
+	ocfs2_journal_dirty(handle, root_bh);
+	ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
+
+	ocfs2_unlink_path(inode, handle, dealloc, right_path,
+			  subtree_index + 1);
+}
+
+static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle,
+				     struct ocfs2_path *left_path,
+				     struct ocfs2_path *right_path,
+				     int subtree_index,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc,
+				     int *deleted)
+{
+	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 ocfs2_extent_list *right_leaf_el, *left_leaf_el;
+	struct ocfs2_extent_block *eb;
+
+	*deleted = 0;
+
+	right_leaf_el = path_leaf_el(right_path);
+	left_leaf_el = path_leaf_el(left_path);
+	root_bh = left_path->p_node[subtree_index].bh;
+	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
+
+	if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
+		return 0;
+
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
+	if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
+		/*
+		 * It's legal for us to proceed if the right leaf is
+		 * the rightmost one and it has an empty extent. There
+		 * are two cases to handle - whether the leaf will be
+		 * empty after removal or not. If the leaf isn't empty
+		 * then just remove the empty extent up front. The
+		 * next block will handle empty leaves by flagging
+		 * them for unlink.
+		 *
+		 * Non rightmost leaves will throw -EAGAIN and the
+		 * caller can manually move the subtree and retry.
 		 */
-		if (left_cpos) {
-			left_path = ocfs2_new_path(path_root_bh(right_path),
-						   path_root_el(right_path));
-			if (!left_path) {
-				ret = -ENOMEM;
-				mlog_errno(ret);
-				goto out;
-			}
 
-			ret = ocfs2_find_path(inode, left_path, left_cpos);
+		if (eb->h_next_leaf_blk != 0ULL)
+			return -EAGAIN;
+
+		if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
+			ret = ocfs2_journal_access(handle, inode,
+						   path_leaf_bh(right_path),
+						   OCFS2_JOURNAL_ACCESS_WRITE);
 			if (ret) {
 				mlog_errno(ret);
 				goto out;
 			}
 
-			/*
-			 * ocfs2_insert_path() will pass the left_path to the
-			 * journal for us.
-			 */
+			ocfs2_remove_empty_extent(right_leaf_el);
+		} else
+			right_has_empty = 1;
+	}
+
+	if (eb->h_next_leaf_blk == 0ULL &&
+	    le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
+		/*
+		 * We have to update i_last_eb_blk during the meta
+		 * data delete.
+		 */
+		ret = ocfs2_journal_access(handle, inode, di_bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
 		}
+
+		del_right_subtree = 1;
 	}
 
-	ret = ocfs2_journal_access_path(inode, handle, right_path);
+	/*
+	 * Getting here with an empty extent in the right path implies
+	 * that it's the rightmost path and will be deleted.
+	 */
+	BUG_ON(right_has_empty && !del_right_subtree);
+
+	ret = ocfs2_journal_access(handle, inode, root_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;
 	}
 
-	el = path_root_el(right_path);
-	bh = path_root_bh(right_path);
-	i = 0;
-	while (1) {
-		struct ocfs2_extent_rec *rec;
+	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
+		ret = ocfs2_journal_access(handle, inode,
+					   right_path->p_node[i].bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 
-		next_free = le16_to_cpu(el->l_next_free_rec);
-		if (next_free == 0) {
-			ocfs2_error(inode->i_sb,
-				    "Dinode %llu has a bad extent list",
-				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
-			ret = -EIO;
+		ret = ocfs2_journal_access(handle, inode,
+					   left_path->p_node[i].bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
 			goto out;
 		}
+	}
 
-		rec = &el->l_recs[next_free - 1];
+	if (!right_has_empty) {
+		/*
+		 * Only do this if we're moving a real
+		 * record. Otherwise, the action is delayed until
+		 * after removal of the right path in which case we
+		 * can do a simple shift to remove the empty extent.
+		 */
+		ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
+		memset(&right_leaf_el->l_recs[0], 0,
+		       sizeof(struct ocfs2_extent_rec));
+	}
+	if (eb->h_next_leaf_blk == 0ULL) {
+		/*
+		 * Move recs over to get rid of empty extent, decrease
+		 * next_free. This is allowed to remove the last
+		 * extent in our leaf (setting l_next_free_rec to
+		 * zero) - the delete code below won't care.
+		 */
+		ocfs2_remove_empty_extent(right_leaf_el);
+	}
 
-		rec->e_int_clusters = insert_rec->e_cpos;
-		le32_add_cpu(&rec->e_int_clusters,
-			     le16_to_cpu(insert_rec->e_leaf_clusters));
-		le32_add_cpu(&rec->e_int_clusters,
-			     -le32_to_cpu(rec->e_cpos));
+	ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
+	if (ret)
+		mlog_errno(ret);
+	ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
+	if (ret)
+		mlog_errno(ret);
 
-		ret = ocfs2_journal_dirty(handle, bh);
+	if (del_right_subtree) {
+		ocfs2_unlink_subtree(inode, handle, left_path, right_path,
+				     subtree_index, dealloc);
+		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;
+
+		/*
+		 * Removal of the extent in the left leaf was skipped
+		 * above so we could delete the right path
+		 * 1st.
+		 */
+		if (right_has_empty)
+			ocfs2_remove_empty_extent(left_leaf_el);
+
+		ret = ocfs2_journal_dirty(handle, di_bh);
 		if (ret)
 			mlog_errno(ret);
 
-		/* Don't touch the leaf node */
-		if (++i >= right_path->p_tree_depth)
-			break;
-
-		bh = right_path->p_node[i].bh;
-		el = right_path->p_node[i].el;
-	}
+		*deleted = 1;
+	} else
+		ocfs2_complete_edge_insert(inode, handle, left_path, right_path,
+					   subtree_index);
 
-	*ret_left_path = left_path;
-	ret = 0;
 out:
-	if (ret != 0)
-		ocfs2_free_path(left_path);
-
 	return ret;
 }
 
 /*
- * This function only does inserts on an allocation b-tree. For dinode
- * lists, ocfs2_insert_at_leaf() is called directly.
+ * Given a full path, determine what cpos value would return us a path
+ * containing the leaf immediately to the right of the current one.
  *
- * 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
- * portion of the tree after an edge insert.
+ * Will return zero if the path passed in is already the rightmost path.
+ *
+ * This looks similar, but is subtly different to
+ * ocfs2_find_cpos_for_left_leaf().
  */
-static int ocfs2_insert_path(struct inode *inode,
-			     handle_t *handle,
-			     struct ocfs2_path *left_path,
-			     struct ocfs2_path *right_path,
-			     struct ocfs2_extent_rec *insert_rec,
-			     struct ocfs2_insert_type *insert)
+static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
+					  struct ocfs2_path *path, u32 *cpos)
 {
-	int ret, subtree_index;
-	struct buffer_head *leaf_bh = path_leaf_bh(right_path);
+	int i, j, ret = 0;
+	u64 blkno;
 	struct ocfs2_extent_list *el;
 
-	/*
-	 * Pass both paths to the journal. The majority of inserts
-	 * will be touching all components anyway.
-	 */
-	ret = ocfs2_journal_access_path(inode, handle, right_path);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out;
-	}
+	*cpos = 0;
 
-	if (left_path) {
-		int credits = handle->h_buffer_credits;
+	if (path->p_tree_depth == 0)
+		return 0;
+
+	blkno = path_leaf_bh(path)->b_blocknr;
+
+	/* Start at the tree node just above the leaf and work our way up. */
+	i = path->p_tree_depth - 1;
+	while (i >= 0) {
+		int next_free;
+
+		el = path->p_node[i].el;
 
 		/*
-		 * There's a chance that left_path got passed back to
-		 * us without being accounted for in the
-		 * journal. Extend our transaction here to be sure we
-		 * can change those blocks.
+		 * Find the extent record just after the one in our
+		 * path.
 		 */
-		credits += left_path->p_tree_depth;
+		next_free = le16_to_cpu(el->l_next_free_rec);
+		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
+			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
+				if (j == (next_free - 1)) {
+					if (i == 0) {
+						/*
+						 * We've determined that the
+						 * path specified is already
+						 * the rightmost one - return a
+						 * cpos of zero.
+						 */
+						goto out;
+					}
+					/*
+					 * The rightmost record points to our
+					 * leaf - we need to travel up the
+					 * tree one level.
+					 */
+					goto next_node;
+				}
 
-		ret = ocfs2_extend_trans(handle, credits);
-		if (ret < 0) {
-			mlog_errno(ret);
-			goto out;
+				*cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
+				goto out;
+			}
 		}
 
-		ret = ocfs2_journal_access_path(inode, handle, left_path);
-		if (ret < 0) {
-			mlog_errno(ret);
-			goto out;
-		}
+		/*
+		 * If we got here, we never found a valid node where
+		 * the tree indicated one should be.
+		 */
+		ocfs2_error(sb,
+			    "Invalid extent tree at extent block %llu\n",
+			    (unsigned long long)blkno);
+		ret = -EROFS;
+		goto out;
+
+next_node:
+		blkno = path->p_node[i].bh->b_blocknr;
+		i--;
 	}
 
-	el = path_leaf_el(right_path);
+out:
+	return ret;
+}
 
-	ocfs2_insert_at_leaf(insert_rec, el, insert, inode);
-	ret = ocfs2_journal_dirty(handle, leaf_bh);
-	if (ret)
-		mlog_errno(ret);
+static int ocfs2_rotate_rightmost_leaf_left(struct inode *inode,
+					    handle_t *handle,
+					    struct buffer_head *bh,
+					    struct ocfs2_extent_list *el)
+{
+	int ret;
 
-	if (left_path) {
-		/*
-		 * The rotate code has indicated that we need to fix
-		 * up portions of the tree after the insert.
-		 *
-		 * XXX: Should we extend the transaction here?
-		 */
-		subtree_index = ocfs2_find_subtree_root(inode, left_path,
-							right_path);
-		ocfs2_complete_edge_insert(inode, handle, left_path,
-					   right_path, subtree_index);
+	if (!ocfs2_is_empty_extent(&el->l_recs[0]))
+		return 0;
+
+	ret = ocfs2_journal_access(handle, inode, bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
 	}
 
-	ret = 0;
+	ocfs2_remove_empty_extent(el);
+
+	ret = ocfs2_journal_dirty(handle, bh);
+	if (ret)
+		mlog_errno(ret);
+
 out:
 	return ret;
 }
 
-static int ocfs2_do_insert_extent(struct inode *inode,
-				  handle_t *handle,
-				  struct buffer_head *di_bh,
-				  struct ocfs2_extent_rec *insert_rec,
-				  struct ocfs2_insert_type *type)
+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)
 {
-	int ret, rotate = 0;
-	u32 cpos;
-	struct ocfs2_path *right_path = NULL;
+	int ret, subtree_root, deleted;
+	u32 right_cpos;
 	struct ocfs2_path *left_path = NULL;
-	struct ocfs2_dinode *di;
-	struct ocfs2_extent_list *el;
+	struct ocfs2_path *right_path = NULL;
 
-	di = (struct ocfs2_dinode *) di_bh->b_data;
-	el = &di->id2.i_list;
+	BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])));
 
-	ret = ocfs2_journal_access(handle, inode, di_bh,
-				   OCFS2_JOURNAL_ACCESS_WRITE);
+	*empty_extent_path = NULL;
+
+	ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, path,
+					     &right_cpos);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;
 	}
 
-	if (le16_to_cpu(el->l_tree_depth) == 0) {
-		ocfs2_insert_at_leaf(insert_rec, el, type, inode);
-		goto out_update_clusters;
+	left_path = ocfs2_new_path(path_root_bh(path),
+				   path_root_el(path));
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
 	}
 
-	right_path = ocfs2_new_inode_path(di_bh);
+	ocfs2_cp_path(left_path, path);
+
+	right_path = ocfs2_new_path(path_root_bh(path),
+				    path_root_el(path));
 	if (!right_path) {
 		ret = -ENOMEM;
 		mlog_errno(ret);
 		goto out;
 	}
 
-	/*
-	 * Determine the path to start with. Rotations need the
-	 * rightmost path, everything else can go directly to the
-	 * target leaf.
-	 */
-	cpos = le32_to_cpu(insert_rec->e_cpos);
-	if (type->ins_appending == APPEND_NONE &&
-	    type->ins_contig == CONTIG_NONE) {
-		rotate = 1;
-		cpos = UINT_MAX;
-	}
+	while (right_cpos) {
+		ret = ocfs2_find_path(inode, right_path, right_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 
-	ret = ocfs2_find_path(inode, right_path, cpos);
-	if (ret) {
-		mlog_errno(ret);
-		goto out;
-	}
+		subtree_root = ocfs2_find_subtree_root(inode, left_path,
+						       right_path);
 
-	/*
-	 * Rotations and appends need special treatment - they modify
-	 * parts of the tree's above them.
-	 *
-	 * Both might pass back a path immediate to the left of the
-	 * one being inserted to. This will be cause
-	 * ocfs2_insert_path() to modify the rightmost records of
-	 * left_path to account for an edge insert.
-	 *
-	 * XXX: When modifying this code, keep in mind that an insert
-	 * can wind up skipping both of these two special cases...
-	 */
-	if (rotate) {
-		ret = ocfs2_rotate_tree_right(inode, handle,
-					      le32_to_cpu(insert_rec->e_cpos),
-					      right_path, &left_path);
+		mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n",
+		     subtree_root,
+		     (unsigned long long)
+		     right_path->p_node[subtree_root].bh->b_blocknr,
+		     right_path->p_tree_depth);
+
+		ret = ocfs2_extend_rotate_transaction(handle, subtree_root,
+						      orig_credits, left_path);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
 		}
-	} else if (type->ins_appending == APPEND_TAIL
-		   && type->ins_contig != CONTIG_LEFT) {
-		ret = ocfs2_append_rec_to_path(inode, handle, insert_rec,
-					       right_path, &left_path);
+
+		ret = ocfs2_rotate_subtree_left(inode, handle, left_path,
+						right_path, subtree_root,
+						dealloc, &deleted);
+		if (ret == -EAGAIN) {
+			/*
+			 * The rotation has to temporarily stop due to
+			 * the right subtree having an empty
+			 * extent. Pass it back to the caller for a
+			 * fixup.
+			 */
+			*empty_extent_path = right_path;
+			right_path = NULL;
+			goto out;
+		}
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
 		}
-	}
 
-	ret = ocfs2_insert_path(inode, handle, left_path, right_path,
-				insert_rec, type);
-	if (ret) {
-		mlog_errno(ret);
-		goto out;
-	}
+		/*
+		 * The subtree rotate might have removed records on
+		 * the rightmost edge. If so, then rotation is
+		 * complete.
+		 */
+		if (deleted)
+			break;
 
-out_update_clusters:
-	ocfs2_update_dinode_clusters(inode, di,
-				     le16_to_cpu(insert_rec->e_leaf_clusters));
+		ocfs2_mv_path(left_path, right_path);
 
-	ret = ocfs2_journal_dirty(handle, di_bh);
-	if (ret)
-		mlog_errno(ret);
+		ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path,
+						     &right_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
 
 out:
-	ocfs2_free_path(left_path);
 	ocfs2_free_path(right_path);
+	ocfs2_free_path(left_path);
 
 	return ret;
 }
 
-static void ocfs2_figure_contig_type(struct inode *inode,
-				     struct ocfs2_insert_type *insert,
-				     struct ocfs2_extent_list *el,
-				     struct ocfs2_extent_rec *insert_rec)
+static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle,
+				       struct ocfs2_path *path,
+				       struct ocfs2_cached_dealloc_ctxt *dealloc)
 {
-	int i;
-	enum ocfs2_contig_type contig_type = CONTIG_NONE;
+	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;
 
-	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+	/*
+	 * 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;
+	}
 
-	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
-		contig_type = ocfs2_extent_contig(inode, &el->l_recs[i],
-						  insert_rec);
-		if (contig_type != CONTIG_NONE) {
-			insert->ins_contig_index = i;
-			break;
-		}
+	/*
+	 * There's two ways we handle this depending on
+	 * whether path is the only existing one.
+	 */
+	ret = ocfs2_extend_rotate_transaction(handle, 0,
+					      handle->h_buffer_credits,
+					      path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
 	}
-	insert->ins_contig = contig_type;
-}
 
-/*
- * This should only be called against the righmost leaf extent list.
- *
- * 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
- * then the logic here makes sense.
- */
-static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
-					struct ocfs2_extent_list *el,
-					struct ocfs2_extent_rec *insert_rec)
-{
-	int i;
-	u32 cpos = le32_to_cpu(insert_rec->e_cpos);
-	struct ocfs2_extent_rec *rec;
+	ret = ocfs2_journal_access_path(inode, handle, path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
 
-	insert->ins_appending = APPEND_NONE;
+	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
 
-	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+	if (cpos) {
+		/*
+		 * We have a path to the left of this one - it needs
+		 * an update too.
+		 */
+		left_path = ocfs2_new_path(path_root_bh(path),
+					   path_root_el(path));
+		if (!left_path) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
+		}
 
-	if (!el->l_next_free_rec)
-		goto set_tail_append;
+		ret = ocfs2_find_path(inode, left_path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 
-	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
-		/* Were all records empty? */
-		if (le16_to_cpu(el->l_next_free_rec) == 1)
-			goto set_tail_append;
-	}
+		ret = ocfs2_journal_access_path(inode, handle, left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 
-	i = le16_to_cpu(el->l_next_free_rec) - 1;
-	rec = &el->l_recs[i];
+		subtree_index = ocfs2_find_subtree_root(inode, left_path, path);
 
-	if (cpos >=
-	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
-		goto set_tail_append;
+		ocfs2_unlink_subtree(inode, handle, left_path, path,
+				     subtree_index, dealloc);
+		ocfs2_update_edge_lengths(inode, handle, left_path);
 
-	return;
+		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
+		di->i_last_eb_blk = eb->h_blkno;
+	} else {
+		/*
+		 * 'path' is also the leftmost path which
+		 * means it must be the only one. This gets
+		 * handled differently because we want to
+		 * revert the inode back to having extents
+		 * in-line.
+		 */
+		ocfs2_unlink_path(inode, handle, dealloc, path, 1);
 
-set_tail_append:
-	insert->ins_appending = APPEND_TAIL;
+		el = &di->id2.i_list;
+		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_journal_dirty(handle, path_root_bh(path));
+
+out:
+	ocfs2_free_path(left_path);
+	return ret;
 }
 
 /*
- * Helper function called at the begining of an insert.
+ * Left rotation of btree records.
  *
- * This computes a few things that are commonly used in the process of
- * inserting into the btree:
- *   - Whether the new extent is contiguous with an existing one.
- *   - The current tree depth.
- *   - Whether the insert is an appending one.
- *   - The total # of free records in the tree.
+ * In many ways, this is (unsurprisingly) the opposite of right
+ * rotation. We start at some non-rightmost path containing an empty
+ * extent in the leaf block. The code works its way to the rightmost
+ * path by rotating records to the left in every subtree.
  *
- * All of the information is stored on the ocfs2_insert_type
- * structure.
+ * This is used by any code which reduces the number of extent records
+ * in a leaf. After removal, an empty record should be placed in the
+ * leftmost list position.
+ *
+ * This won't handle a length update of the rightmost path records if
+ * the rightmost tree leaf record is removed so the caller is
+ * responsible for detecting and correcting that.
  */
-static int ocfs2_figure_insert_type(struct inode *inode,
-				    struct buffer_head *di_bh,
-				    struct buffer_head **last_eb_bh,
-				    struct ocfs2_extent_rec *insert_rec,
-				    struct ocfs2_insert_type *insert)
+static int ocfs2_rotate_tree_left(struct inode *inode, handle_t *handle,
+				  struct ocfs2_path *path,
+				  struct ocfs2_cached_dealloc_ctxt *dealloc)
 {
-	int ret;
-	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	int ret, orig_credits = handle->h_buffer_credits;
+	struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list *el;
-	struct ocfs2_path *path = NULL;
-	struct buffer_head *bh = NULL;
 
-	el = &di->id2.i_list;
-	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
+	el = path_leaf_el(path);
+	if (!ocfs2_is_empty_extent(&el->l_recs[0]))
+		return 0;
 
-	if (el->l_tree_depth) {
+	if (path->p_tree_depth == 0) {
+rightmost_no_delete:
 		/*
-		 * If we have tree depth, we read in the
-		 * rightmost extent block ahead of time as
-		 * ocfs2_figure_insert_type() and ocfs2_add_branch()
-		 * may want it later.
+		 * In-inode extents. This is trivially handled, so do
+		 * it up front.
 		 */
-		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
-				       le64_to_cpu(di->i_last_eb_blk), &bh,
-				       OCFS2_BH_CACHED, inode);
-		if (ret) {
-			mlog_exit(ret);
-			goto out;
-		}
-		eb = (struct ocfs2_extent_block *) bh->b_data;
-		el = &eb->h_list;
+		ret = ocfs2_rotate_rightmost_leaf_left(inode, handle,
+						       path_leaf_bh(path),
+						       path_leaf_el(path));
+		if (ret)
+			mlog_errno(ret);
+		goto out;
 	}
 
 	/*
-	 * Unless we have a contiguous insert, we'll need to know if
-	 * there is room left in our allocation tree for another
-	 * extent record.
+	 * Handle rightmost branch now. There's several cases:
+	 *  1) simple rotation leaving records in there. That's trivial.
+	 *  2) rotation requiring a branch delete - there's no more
+	 *     records left. Two cases of this:
+	 *     a) There are branches to the left.
+	 *     b) This is also the leftmost (the only) branch.
 	 *
-	 * XXX: This test is simplistic, we can search for empty
-	 * extent records too.
+	 *  1) is handled via ocfs2_rotate_rightmost_leaf_left()
+	 *  2a) we need the left branch so that we can update it with the unlink
+	 *  2b) we need to bring the inode back to inline extents.
 	 */
-	insert->ins_free_records = le16_to_cpu(el->l_count) -
-		le16_to_cpu(el->l_next_free_rec);
 
-	if (!insert->ins_tree_depth) {
-		ocfs2_figure_contig_type(inode, insert, el, insert_rec);
-		ocfs2_figure_appending_type(insert, el, insert_rec);
-		return 0;
-	}
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
+	el = &eb->h_list;
+	if (eb->h_next_leaf_blk == 0) {
+		/*
+		 * This gets a bit tricky if we're going to delete the
+		 * rightmost path. Get the other cases out of the way
+		 * 1st.
+		 */
+		if (le16_to_cpu(el->l_next_free_rec) > 1)
+			goto rightmost_no_delete;
 
-	path = ocfs2_new_inode_path(di_bh);
-	if (!path) {
-		ret = -ENOMEM;
-		mlog_errno(ret);
+		if (le16_to_cpu(el->l_next_free_rec) == 0) {
+			ret = -EIO;
+			ocfs2_error(inode->i_sb,
+				    "Inode %llu has empty extent block at %llu",
+				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+				    (unsigned long long)le64_to_cpu(eb->h_blkno));
+			goto out;
+		}
+
+		/*
+		 * XXX: The caller can not trust "path" any more after
+		 * this as it will have been deleted. What do we do?
+		 *
+		 * In theory the rotate-for-merge code will never get
+		 * here because it'll always ask for a rotate in a
+		 * nonempty list.
+		 */
+
+		ret = ocfs2_remove_rightmost_path(inode, handle, path,
+						  dealloc);
+		if (ret)
+			mlog_errno(ret);
 		goto out;
 	}
 
 	/*
-	 * In the case that we're inserting past what the tree
-	 * currently accounts for, ocfs2_find_path() will return for
-	 * us the rightmost tree path. This is accounted for below in
-	 * the appending code.
+	 * Now we can loop, remembering the path we get from -EAGAIN
+	 * and restarting from there.
 	 */
-	ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos));
-	if (ret) {
+try_rotate:
+	ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, path,
+				       dealloc, &restart_path);
+	if (ret && ret != -EAGAIN) {
 		mlog_errno(ret);
 		goto out;
 	}
 
-	el = path_leaf_el(path);
-
-	/*
-	 * Now that we have the path, there's two things we want to determine:
-	 * 1) Contiguousness (also set contig_index if this is so)
-	 *
-	 * 2) Are we doing an append? We can trivially break this up
-         *     into two types of appends: simple record append, or a
-         *     rotate inside the tail leaf.
-	 */
-	ocfs2_figure_contig_type(inode, insert, el, insert_rec);
+	while (ret == -EAGAIN) {
+		tmp_path = restart_path;
+		restart_path = NULL;
+
+		ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits,
+					       tmp_path, dealloc,
+					       &restart_path);
+		if (ret && ret != -EAGAIN) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ocfs2_free_path(tmp_path);
+		tmp_path = NULL;
+
+		if (ret == 0)
+			goto try_rotate;
+	}
+
+out:
+	ocfs2_free_path(tmp_path);
+	ocfs2_free_path(restart_path);
+	return ret;
+}
+
+static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
+				int index)
+{
+	struct ocfs2_extent_rec *rec = &el->l_recs[index];
+	unsigned int size;
+
+	if (rec->e_leaf_clusters == 0) {
+		/*
+		 * We consumed all of the merged-from record. An empty
+		 * extent cannot exist anywhere but the 1st array
+		 * position, so move things over if the merged-from
+		 * record doesn't occupy that position.
+		 *
+		 * This creates a new empty extent so the caller
+		 * should be smart enough to have removed any existing
+		 * ones.
+		 */
+		if (index > 0) {
+			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
+			size = index * sizeof(struct ocfs2_extent_rec);
+			memmove(&el->l_recs[1], &el->l_recs[0], size);
+		}
+
+		/*
+		 * Always memset - the caller doesn't check whether it
+		 * created an empty extent, so there could be junk in
+		 * the other fields.
+		 */
+		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+	}
+}
+
+/*
+ * Remove split_rec clusters from the record at index and merge them
+ * onto the beginning of the record at index + 1.
+ */
+static int ocfs2_merge_rec_right(struct inode *inode, struct buffer_head *bh,
+				handle_t *handle,
+				struct ocfs2_extent_rec *split_rec,
+				struct ocfs2_extent_list *el, int index)
+{
+	int ret;
+	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
+	struct ocfs2_extent_rec *left_rec;
+	struct ocfs2_extent_rec *right_rec;
+
+	BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
+
+	left_rec = &el->l_recs[index];
+	right_rec = &el->l_recs[index + 1];
+
+	ret = ocfs2_journal_access(handle, inode, bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
+
+	le32_add_cpu(&right_rec->e_cpos, -split_clusters);
+	le64_add_cpu(&right_rec->e_blkno,
+		     -ocfs2_clusters_to_blocks(inode->i_sb, split_clusters));
+	le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
+
+	ocfs2_cleanup_merge(el, index);
+
+	ret = ocfs2_journal_dirty(handle, bh);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+/*
+ * Remove split_rec clusters from the record at index and merge them
+ * onto the tail of the record at index - 1.
+ */
+static int ocfs2_merge_rec_left(struct inode *inode, struct buffer_head *bh,
+				handle_t *handle,
+				struct ocfs2_extent_rec *split_rec,
+				struct ocfs2_extent_list *el, int index)
+{
+	int ret, has_empty_extent = 0;
+	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
+	struct ocfs2_extent_rec *left_rec;
+	struct ocfs2_extent_rec *right_rec;
+
+	BUG_ON(index <= 0);
+
+	left_rec = &el->l_recs[index - 1];
+	right_rec = &el->l_recs[index];
+	if (ocfs2_is_empty_extent(&el->l_recs[0]))
+		has_empty_extent = 1;
+
+	ret = ocfs2_journal_access(handle, inode, bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (has_empty_extent && index == 1) {
+		/*
+		 * The easy case - we can just plop the record right in.
+		 */
+		*left_rec = *split_rec;
+
+		has_empty_extent = 0;
+	} else {
+		le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
+	}
+
+	le32_add_cpu(&right_rec->e_cpos, split_clusters);
+	le64_add_cpu(&right_rec->e_blkno,
+		     ocfs2_clusters_to_blocks(inode->i_sb, split_clusters));
+	le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
+
+	ocfs2_cleanup_merge(el, index);
+
+	ret = ocfs2_journal_dirty(handle, bh);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+static int ocfs2_try_to_merge_extent(struct inode *inode,
+				     handle_t *handle,
+				     struct ocfs2_path *left_path,
+				     int split_index,
+				     struct ocfs2_extent_rec *split_rec,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc,
+				     struct ocfs2_merge_ctxt *ctxt)
+
+{
+	int ret = 0, delete_tail_recs = 0;
+	struct ocfs2_extent_list *el = path_leaf_el(left_path);
+	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
+
+	BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
+
+	if (ctxt->c_split_covers_rec) {
+		delete_tail_recs++;
+
+		if (ctxt->c_contig_type == CONTIG_LEFTRIGHT ||
+		    ctxt->c_has_empty_extent)
+			delete_tail_recs++;
+
+		if (ctxt->c_has_empty_extent) {
+			/*
+			 * The merge code will need to create an empty
+			 * extent to take the place of the newly
+			 * emptied slot. Remove any pre-existing empty
+			 * extents - having more than one in a leaf is
+			 * illegal.
+			 */
+			ret = ocfs2_rotate_tree_left(inode, handle, left_path,
+						     dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+			split_index--;
+			rec = &el->l_recs[split_index];
+		}
+	}
+
+	if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
+		/*
+		 * Left-right contig implies this.
+		 */
+		BUG_ON(!ctxt->c_split_covers_rec);
+		BUG_ON(split_index == 0);
+
+		/*
+		 * Since the leftright insert always covers the entire
+		 * extent, this call will delete the insert record
+		 * entirely, resulting in an empty extent record added to
+		 * the extent block.
+		 *
+		 * Since the adding of an empty extent shifts
+		 * everything back to the right, there's no need to
+		 * update split_index here.
+		 */
+		ret = ocfs2_merge_rec_left(inode, path_leaf_bh(left_path),
+					   handle, split_rec, el, split_index);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * We can only get this from logic error above.
+		 */
+		BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
+
+		/*
+		 * The left merge left us with an empty extent, remove
+		 * it.
+		 */
+		ret = ocfs2_rotate_tree_left(inode, handle, left_path, dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		split_index--;
+		rec = &el->l_recs[split_index];
+
+		/*
+		 * Note that we don't pass split_rec here on purpose -
+		 * we've merged it into the left side.
+		 */
+		ret = ocfs2_merge_rec_right(inode, path_leaf_bh(left_path),
+					    handle, rec, el, split_index);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
+
+		ret = ocfs2_rotate_tree_left(inode, handle, left_path,
+					     dealloc);
+		/*
+		 * Error from this last rotate is not critical, so
+		 * print but don't bubble it up.
+		 */
+		if (ret)
+			mlog_errno(ret);
+		ret = 0;
+	} else {
+		/*
+		 * Merge a record to the left or right.
+		 *
+		 * 'contig_type' is relative to the existing record,
+		 * so for example, if we're "right contig", it's to
+		 * the record on the left (hence the left merge).
+		 */
+		if (ctxt->c_contig_type == CONTIG_RIGHT) {
+			ret = ocfs2_merge_rec_left(inode,
+						   path_leaf_bh(left_path),
+						   handle, split_rec, el,
+						   split_index);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		} else {
+			ret = ocfs2_merge_rec_right(inode,
+						    path_leaf_bh(left_path),
+						    handle, split_rec, el,
+						    split_index);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		if (ctxt->c_split_covers_rec) {
+			/*
+			 * The merge may have left an empty extent in
+			 * our leaf. Try to rotate it away.
+			 */
+			ret = ocfs2_rotate_tree_left(inode, handle, left_path,
+						     dealloc);
+			if (ret)
+				mlog_errno(ret);
+			ret = 0;
+		}
+	}
+
+out:
+	return ret;
+}
+
+static void ocfs2_subtract_from_rec(struct super_block *sb,
+				    enum ocfs2_split_type split,
+				    struct ocfs2_extent_rec *rec,
+				    struct ocfs2_extent_rec *split_rec)
+{
+	u64 len_blocks;
+
+	len_blocks = ocfs2_clusters_to_blocks(sb,
+				le16_to_cpu(split_rec->e_leaf_clusters));
+
+	if (split == SPLIT_LEFT) {
+		/*
+		 * Region is on the left edge of the existing
+		 * record.
+		 */
+		le32_add_cpu(&rec->e_cpos,
+			     le16_to_cpu(split_rec->e_leaf_clusters));
+		le64_add_cpu(&rec->e_blkno, len_blocks);
+		le16_add_cpu(&rec->e_leaf_clusters,
+			     -le16_to_cpu(split_rec->e_leaf_clusters));
+	} else {
+		/*
+		 * Region is on the right edge of the existing
+		 * record.
+		 */
+		le16_add_cpu(&rec->e_leaf_clusters,
+			     -le16_to_cpu(split_rec->e_leaf_clusters));
+	}
+}
+
+/*
+ * Do the final bits of extent record insertion at the target leaf
+ * list. If this leaf is part of an allocation tree, it is assumed
+ * that the tree above has been prepared.
+ */
+static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec,
+				 struct ocfs2_extent_list *el,
+				 struct ocfs2_insert_type *insert,
+				 struct inode *inode)
+{
+	int i = insert->ins_contig_index;
+	unsigned int range;
+	struct ocfs2_extent_rec *rec;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	if (insert->ins_split != SPLIT_NONE) {
+		i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
+		BUG_ON(i == -1);
+		rec = &el->l_recs[i];
+		ocfs2_subtract_from_rec(inode->i_sb, insert->ins_split, rec,
+					insert_rec);
+		goto rotate;
+	}
+
+	/*
+	 * Contiguous insert - either left or right.
+	 */
+	if (insert->ins_contig != CONTIG_NONE) {
+		rec = &el->l_recs[i];
+		if (insert->ins_contig == CONTIG_LEFT) {
+			rec->e_blkno = insert_rec->e_blkno;
+			rec->e_cpos = insert_rec->e_cpos;
+		}
+		le16_add_cpu(&rec->e_leaf_clusters,
+			     le16_to_cpu(insert_rec->e_leaf_clusters));
+		return;
+	}
+
+	/*
+	 * Handle insert into an empty leaf.
+	 */
+	if (le16_to_cpu(el->l_next_free_rec) == 0 ||
+	    ((le16_to_cpu(el->l_next_free_rec) == 1) &&
+	     ocfs2_is_empty_extent(&el->l_recs[0]))) {
+		el->l_recs[0] = *insert_rec;
+		el->l_next_free_rec = cpu_to_le16(1);
+		return;
+	}
+
+	/*
+	 * Appending insert.
+	 */
+	if (insert->ins_appending == APPEND_TAIL) {
+		i = le16_to_cpu(el->l_next_free_rec) - 1;
+		rec = &el->l_recs[i];
+		range = le32_to_cpu(rec->e_cpos)
+			+ le16_to_cpu(rec->e_leaf_clusters);
+		BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
+
+		mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
+				le16_to_cpu(el->l_count),
+				"inode %lu, depth %u, count %u, next free %u, "
+				"rec.cpos %u, rec.clusters %u, "
+				"insert.cpos %u, insert.clusters %u\n",
+				inode->i_ino,
+				le16_to_cpu(el->l_tree_depth),
+				le16_to_cpu(el->l_count),
+				le16_to_cpu(el->l_next_free_rec),
+				le32_to_cpu(el->l_recs[i].e_cpos),
+				le16_to_cpu(el->l_recs[i].e_leaf_clusters),
+				le32_to_cpu(insert_rec->e_cpos),
+				le16_to_cpu(insert_rec->e_leaf_clusters));
+		i++;
+		el->l_recs[i] = *insert_rec;
+		le16_add_cpu(&el->l_next_free_rec, 1);
+		return;
+	}
+
+rotate:
+	/*
+	 * Ok, we have to rotate.
+	 *
+	 * At this point, it is safe to assume that inserting into an
+	 * empty leaf and appending to a leaf have both been handled
+	 * above.
+	 *
+	 * This leaf needs to have space, either by the empty 1st
+	 * extent record, or by virtue of an l_next_rec < l_count.
+	 */
+	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,
+					   struct ocfs2_extent_rec *insert_rec)
+{
+	int ret, i, next_free;
+	struct buffer_head *bh;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+
+	/*
+	 * Update everything except the leaf block.
+	 */
+	for (i = 0; i < path->p_tree_depth; i++) {
+		bh = path->p_node[i].bh;
+		el = path->p_node[i].el;
+
+		next_free = le16_to_cpu(el->l_next_free_rec);
+		if (next_free == 0) {
+			ocfs2_error(inode->i_sb,
+				    "Dinode %llu has a bad extent list",
+				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
+			ret = -EIO;
+			return;
+		}
+
+		rec = &el->l_recs[next_free - 1];
+
+		rec->e_int_clusters = insert_rec->e_cpos;
+		le32_add_cpu(&rec->e_int_clusters,
+			     le16_to_cpu(insert_rec->e_leaf_clusters));
+		le32_add_cpu(&rec->e_int_clusters,
+			     -le32_to_cpu(rec->e_cpos));
+
+		ret = ocfs2_journal_dirty(handle, bh);
+		if (ret)
+			mlog_errno(ret);
+
+	}
+}
+
+static int ocfs2_append_rec_to_path(struct inode *inode, handle_t *handle,
+				    struct ocfs2_extent_rec *insert_rec,
+				    struct ocfs2_path *right_path,
+				    struct ocfs2_path **ret_left_path)
+{
+	int ret, next_free;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_path *left_path = NULL;
+
+	*ret_left_path = NULL;
+
+	/*
+	 * This shouldn't happen for non-trees. The extent rec cluster
+	 * count manipulation below only works for interior nodes.
+	 */
+	BUG_ON(right_path->p_tree_depth == 0);
+
+	/*
+	 * If our appending insert is at the leftmost edge of a leaf,
+	 * then we might need to update the rightmost records of the
+	 * neighboring path.
+	 */
+	el = path_leaf_el(right_path);
+	next_free = le16_to_cpu(el->l_next_free_rec);
+	if (next_free == 0 ||
+	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
+		u32 left_cpos;
+
+		ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path,
+						    &left_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		mlog(0, "Append may need a left path update. cpos: %u, "
+		     "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos),
+		     left_cpos);
+
+		/*
+		 * No need to worry if the append is already in the
+		 * leftmost leaf.
+		 */
+		if (left_cpos) {
+			left_path = ocfs2_new_path(path_root_bh(right_path),
+						   path_root_el(right_path));
+			if (!left_path) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_find_path(inode, left_path, left_cpos);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			/*
+			 * ocfs2_insert_path() will pass the left_path to the
+			 * journal for us.
+			 */
+		}
+	}
+
+	ret = ocfs2_journal_access_path(inode, handle, right_path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_adjust_rightmost_records(inode, handle, right_path, insert_rec);
+
+	*ret_left_path = left_path;
+	ret = 0;
+out:
+	if (ret != 0)
+		ocfs2_free_path(left_path);
+
+	return ret;
+}
+
+static void ocfs2_split_record(struct inode *inode,
+			       struct ocfs2_path *left_path,
+			       struct ocfs2_path *right_path,
+			       struct ocfs2_extent_rec *split_rec,
+			       enum ocfs2_split_type split)
+{
+	int index;
+	u32 cpos = le32_to_cpu(split_rec->e_cpos);
+	struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
+	struct ocfs2_extent_rec *rec, *tmprec;
+
+	right_el = path_leaf_el(right_path);;
+	if (left_path)
+		left_el = path_leaf_el(left_path);
+
+	el = right_el;
+	insert_el = right_el;
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index != -1) {
+		if (index == 0 && left_path) {
+			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
+
+			/*
+			 * This typically means that the record
+			 * started in the left path but moved to the
+			 * right as a result of rotation. We either
+			 * move the existing record to the left, or we
+			 * do the later insert there.
+			 *
+			 * In this case, the left path should always
+			 * exist as the rotate code will have passed
+			 * it back for a post-insert update.
+			 */
+
+			if (split == SPLIT_LEFT) {
+				/*
+				 * It's a left split. Since we know
+				 * that the rotate code gave us an
+				 * empty extent in the left path, we
+				 * can just do the insert there.
+				 */
+				insert_el = left_el;
+			} else {
+				/*
+				 * Right split - we have to move the
+				 * existing record over to the left
+				 * leaf. The insert will be into the
+				 * newly created empty extent in the
+				 * right leaf.
+				 */
+				tmprec = &right_el->l_recs[index];
+				ocfs2_rotate_leaf(left_el, tmprec);
+				el = left_el;
+
+				memset(tmprec, 0, sizeof(*tmprec));
+				index = ocfs2_search_extent_list(left_el, cpos);
+				BUG_ON(index == -1);
+			}
+		}
+	} else {
+		BUG_ON(!left_path);
+		BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
+		/*
+		 * Left path is easy - we can just allow the insert to
+		 * happen.
+		 */
+		el = left_el;
+		insert_el = left_el;
+		index = ocfs2_search_extent_list(el, cpos);
+		BUG_ON(index == -1);
+	}
+
+	rec = &el->l_recs[index];
+	ocfs2_subtract_from_rec(inode->i_sb, split, rec, split_rec);
+	ocfs2_rotate_leaf(insert_el, split_rec);
+}
+
+/*
+ * This function only does inserts on an allocation b-tree. For dinode
+ * lists, 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
+ * portion of the tree after an edge insert.
+ */
+static int ocfs2_insert_path(struct inode *inode,
+			     handle_t *handle,
+			     struct ocfs2_path *left_path,
+			     struct ocfs2_path *right_path,
+			     struct ocfs2_extent_rec *insert_rec,
+			     struct ocfs2_insert_type *insert)
+{
+	int ret, subtree_index;
+	struct buffer_head *leaf_bh = path_leaf_bh(right_path);
+
+	/*
+	 * Pass both paths to the journal. The majority of inserts
+	 * will be touching all components anyway.
+	 */
+	ret = ocfs2_journal_access_path(inode, handle, right_path);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (left_path) {
+		int credits = handle->h_buffer_credits;
+
+		/*
+		 * There's a chance that left_path got passed back to
+		 * us without being accounted for in the
+		 * journal. Extend our transaction here to be sure we
+		 * can change those blocks.
+		 */
+		credits += left_path->p_tree_depth;
+
+		ret = ocfs2_extend_trans(handle, credits);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_journal_access_path(inode, handle, left_path);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (insert->ins_split != SPLIT_NONE) {
+		/*
+		 * We could call ocfs2_insert_at_leaf() for some types
+		 * of splits, but it's easier to just let one seperate
+		 * function sort it all out.
+		 */
+		ocfs2_split_record(inode, left_path, right_path,
+				   insert_rec, insert->ins_split);
+	} else
+		ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path),
+				     insert, inode);
+
+	ret = ocfs2_journal_dirty(handle, leaf_bh);
+	if (ret)
+		mlog_errno(ret);
+
+	if (left_path) {
+		/*
+		 * The rotate code has indicated that we need to fix
+		 * up portions of the tree after the insert.
+		 *
+		 * XXX: Should we extend the transaction here?
+		 */
+		subtree_index = ocfs2_find_subtree_root(inode, left_path,
+							right_path);
+		ocfs2_complete_edge_insert(inode, handle, left_path,
+					   right_path, subtree_index);
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int ocfs2_do_insert_extent(struct inode *inode,
+				  handle_t *handle,
+				  struct buffer_head *di_bh,
+				  struct ocfs2_extent_rec *insert_rec,
+				  struct ocfs2_insert_type *type)
+{
+	int ret, rotate = 0;
+	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;
+
+	ret = ocfs2_journal_access(handle, inode, di_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (le16_to_cpu(el->l_tree_depth) == 0) {
+		ocfs2_insert_at_leaf(insert_rec, el, type, inode);
+		goto out_update_clusters;
+	}
+
+	right_path = ocfs2_new_inode_path(di_bh);
+	if (!right_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Determine the path to start with. Rotations need the
+	 * rightmost path, everything else can go directly to the
+	 * target leaf.
+	 */
+	cpos = le32_to_cpu(insert_rec->e_cpos);
+	if (type->ins_appending == APPEND_NONE &&
+	    type->ins_contig == CONTIG_NONE) {
+		rotate = 1;
+		cpos = UINT_MAX;
+	}
+
+	ret = ocfs2_find_path(inode, right_path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Rotations and appends need special treatment - they modify
+	 * parts of the tree's above them.
+	 *
+	 * Both might pass back a path immediate to the left of the
+	 * one being inserted to. This will be cause
+	 * ocfs2_insert_path() to modify the rightmost records of
+	 * left_path to account for an edge insert.
+	 *
+	 * XXX: When modifying this code, keep in mind that an insert
+	 * can wind up skipping both of these two special cases...
+	 */
+	if (rotate) {
+		ret = ocfs2_rotate_tree_right(inode, handle, type->ins_split,
+					      le32_to_cpu(insert_rec->e_cpos),
+					      right_path, &left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else if (type->ins_appending == APPEND_TAIL
+		   && type->ins_contig != CONTIG_LEFT) {
+		ret = ocfs2_append_rec_to_path(inode, handle, insert_rec,
+					       right_path, &left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_insert_path(inode, handle, left_path, right_path,
+				insert_rec, type);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out_update_clusters:
+	if (type->ins_split == SPLIT_NONE)
+		ocfs2_update_dinode_clusters(inode, di,
+					     le16_to_cpu(insert_rec->e_leaf_clusters));
+
+	ret = ocfs2_journal_dirty(handle, di_bh);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	ocfs2_free_path(left_path);
+	ocfs2_free_path(right_path);
+
+	return ret;
+}
+
+static enum ocfs2_contig_type
+ocfs2_figure_merge_contig_type(struct inode *inode,
+			       struct ocfs2_extent_list *el, int index,
+			       struct ocfs2_extent_rec *split_rec)
+{
+	struct ocfs2_extent_rec *rec;
+	enum ocfs2_contig_type ret = CONTIG_NONE;
+
+	/*
+	 * We're careful to check for an empty extent record here -
+	 * the merge code will know what to do if it sees one.
+	 */
+
+	if (index > 0) {
+		rec = &el->l_recs[index - 1];
+		if (index == 1 && ocfs2_is_empty_extent(rec)) {
+			if (split_rec->e_cpos == el->l_recs[index].e_cpos)
+				ret = CONTIG_RIGHT;
+		} else {
+			ret = ocfs2_extent_contig(inode, rec, split_rec);
+		}
+	}
+
+	if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) {
+		enum ocfs2_contig_type contig_type;
+
+		rec = &el->l_recs[index + 1];
+		contig_type = ocfs2_extent_contig(inode, rec, split_rec);
+
+		if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
+			ret = CONTIG_LEFTRIGHT;
+		else if (ret == CONTIG_NONE)
+			ret = contig_type;
+	}
+
+	return ret;
+}
+
+static void ocfs2_figure_contig_type(struct inode *inode,
+				     struct ocfs2_insert_type *insert,
+				     struct ocfs2_extent_list *el,
+				     struct ocfs2_extent_rec *insert_rec)
+{
+	int i;
+	enum ocfs2_contig_type contig_type = CONTIG_NONE;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		contig_type = ocfs2_extent_contig(inode, &el->l_recs[i],
+						  insert_rec);
+		if (contig_type != CONTIG_NONE) {
+			insert->ins_contig_index = i;
+			break;
+		}
+	}
+	insert->ins_contig = contig_type;
+}
+
+/*
+ * This should only be called against the righmost leaf extent list.
+ *
+ * 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
+ * then the logic here makes sense.
+ */
+static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
+					struct ocfs2_extent_list *el,
+					struct ocfs2_extent_rec *insert_rec)
+{
+	int i;
+	u32 cpos = le32_to_cpu(insert_rec->e_cpos);
+	struct ocfs2_extent_rec *rec;
+
+	insert->ins_appending = APPEND_NONE;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	if (!el->l_next_free_rec)
+		goto set_tail_append;
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
+		/* Were all records empty? */
+		if (le16_to_cpu(el->l_next_free_rec) == 1)
+			goto set_tail_append;
+	}
+
+	i = le16_to_cpu(el->l_next_free_rec) - 1;
+	rec = &el->l_recs[i];
+
+	if (cpos >=
+	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
+		goto set_tail_append;
+
+	return;
+
+set_tail_append:
+	insert->ins_appending = APPEND_TAIL;
+}
+
+/*
+ * Helper function called at the begining of an insert.
+ *
+ * This computes a few things that are commonly used in the process of
+ * inserting into the btree:
+ *   - Whether the new extent is contiguous with an existing one.
+ *   - The current tree depth.
+ *   - Whether the insert is an appending one.
+ *   - The total # of free records in the tree.
+ *
+ * All of the information is stored on the ocfs2_insert_type
+ * structure.
+ */
+static int ocfs2_figure_insert_type(struct inode *inode,
+				    struct buffer_head *di_bh,
+				    struct buffer_head **last_eb_bh,
+				    struct ocfs2_extent_rec *insert_rec,
+				    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;
+	struct buffer_head *bh = NULL;
+
+	insert->ins_split = SPLIT_NONE;
+
+	el = &di->id2.i_list;
+	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
+
+	if (el->l_tree_depth) {
+		/*
+		 * If we have tree depth, we read in the
+		 * rightmost extent block ahead of time as
+		 * ocfs2_figure_insert_type() and ocfs2_add_branch()
+		 * may want it later.
+		 */
+		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
+				       le64_to_cpu(di->i_last_eb_blk), &bh,
+				       OCFS2_BH_CACHED, inode);
+		if (ret) {
+			mlog_exit(ret);
+			goto out;
+		}
+		eb = (struct ocfs2_extent_block *) bh->b_data;
+		el = &eb->h_list;
+	}
+
+	/*
+	 * Unless we have a contiguous insert, we'll need to know if
+	 * there is room left in our allocation tree for another
+	 * extent record.
+	 *
+	 * XXX: This test is simplistic, we can search for empty
+	 * extent records too.
+	 */
+	insert->ins_free_records = le16_to_cpu(el->l_count) -
+		le16_to_cpu(el->l_next_free_rec);
+
+	if (!insert->ins_tree_depth) {
+		ocfs2_figure_contig_type(inode, insert, el, insert_rec);
+		ocfs2_figure_appending_type(insert, el, insert_rec);
+		return 0;
+	}
+
+	path = ocfs2_new_inode_path(di_bh);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * In the case that we're inserting past what the tree
+	 * currently accounts for, ocfs2_find_path() will return for
+	 * us the rightmost tree path. This is accounted for below in
+	 * the appending code.
+	 */
+	ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos));
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+
+	/*
+	 * Now that we have the path, there's two things we want to determine:
+	 * 1) Contiguousness (also set contig_index if this is so)
+	 *
+	 * 2) Are we doing an append? We can trivially break this up
+         *     into two types of appends: simple record append, or a
+         *     rotate inside the tail leaf.
+	 */
+	ocfs2_figure_contig_type(inode, insert, el, insert_rec);
 
 	/*
 	 * The insert code isn't quite ready to deal with all cases of
@@ -2295,143 +3694,819 @@ static int ocfs2_figure_insert_type(struct inode *inode,
 	 */
 	if (le64_to_cpu(di->i_last_eb_blk) == path_leaf_bh(path)->b_blocknr) {
 		/*
-		 * Ok, ocfs2_find_path() returned us the rightmost
-		 * tree path. This might be an appending insert. There are
-		 * two cases:
-		 *    1) We're doing a true append at the tail:
-		 *	-This might even be off the end of the leaf
-		 *    2) We're "appending" by rotating in the tail
+		 * Ok, ocfs2_find_path() returned us the rightmost
+		 * tree path. This might be an appending insert. There are
+		 * two cases:
+		 *    1) We're doing a true append at the tail:
+		 *	-This might even be off the end of the leaf
+		 *    2) We're "appending" by rotating in the tail
+		 */
+		ocfs2_figure_appending_type(insert, el, insert_rec);
+	}
+
+out:
+	ocfs2_free_path(path);
+
+	if (ret == 0)
+		*last_eb_bh = bh;
+	else
+		brelse(bh);
+	return ret;
+}
+
+/*
+ * Insert an extent into an inode btree.
+ *
+ * The caller needs to update fe->i_clusters
+ */
+int ocfs2_insert_extent(struct ocfs2_super *osb,
+			handle_t *handle,
+			struct inode *inode,
+			struct buffer_head *fe_bh,
+			u32 cpos,
+			u64 start_blk,
+			u32 new_clusters,
+			u8 flags,
+			struct ocfs2_alloc_context *meta_ac)
+{
+	int status;
+	struct buffer_head *last_eb_bh = NULL;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_insert_type insert = {0, };
+	struct ocfs2_extent_rec rec;
+
+	mlog(0, "add %u clusters at position %u to inode %llu\n",
+	     new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
+			(OCFS2_I(inode)->ip_clusters != cpos),
+			"Device %s, asking for sparse allocation: inode %llu, "
+			"cpos %u, clusters %u\n",
+			osb->dev_str,
+			(unsigned long long)OCFS2_I(inode)->ip_blkno, cpos,
+			OCFS2_I(inode)->ip_clusters);
+
+	memset(&rec, 0, sizeof(rec));
+	rec.e_cpos = cpu_to_le32(cpos);
+	rec.e_blkno = cpu_to_le64(start_blk);
+	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,
+					  &insert);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	mlog(0, "Insert.appending: %u, Insert.Contig: %u, "
+	     "Insert.contig_index: %d, Insert.free_records: %d, "
+	     "Insert.tree_depth: %d\n",
+	     insert.ins_appending, insert.ins_contig, insert.ins_contig_index,
+	     insert.ins_free_records, insert.ins_tree_depth);
+
+	if (insert.ins_contig == CONTIG_NONE && insert.ins_free_records == 0) {
+		status = ocfs2_grow_tree(inode, handle, fe_bh,
+					 &insert.ins_tree_depth, &last_eb_bh,
+					 meta_ac);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* Finally, we can add clusters. This might rotate the tree for us. */
+	status = ocfs2_do_insert_extent(inode, handle, fe_bh, &rec, &insert);
+	if (status < 0)
+		mlog_errno(status);
+	else
+		ocfs2_extent_map_insert_rec(inode, &rec);
+
+bail:
+	if (bh)
+		brelse(bh);
+
+	if (last_eb_bh)
+		brelse(last_eb_bh);
+
+	mlog_exit(status);
+	return status;
+}
+
+static void ocfs2_make_right_split_rec(struct super_block *sb,
+				       struct ocfs2_extent_rec *split_rec,
+				       u32 cpos,
+				       struct ocfs2_extent_rec *rec)
+{
+	u32 rec_cpos = le32_to_cpu(rec->e_cpos);
+	u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
+
+	memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
+
+	split_rec->e_cpos = cpu_to_le32(cpos);
+	split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
+
+	split_rec->e_blkno = rec->e_blkno;
+	le64_add_cpu(&split_rec->e_blkno,
+		     ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
+
+	split_rec->e_flags = rec->e_flags;
+}
+
+static int ocfs2_split_and_insert(struct inode *inode,
+				  handle_t *handle,
+				  struct ocfs2_path *path,
+				  struct buffer_head *di_bh,
+				  struct buffer_head **last_eb_bh,
+				  int split_index,
+				  struct ocfs2_extent_rec *orig_split_rec,
+				  struct ocfs2_alloc_context *meta_ac)
+{
+	int ret = 0, depth;
+	unsigned int insert_range, rec_range, do_leftright = 0;
+	struct ocfs2_extent_rec tmprec;
+	struct ocfs2_extent_list *rightmost_el;
+	struct ocfs2_extent_rec rec;
+	struct ocfs2_extent_rec split_rec = *orig_split_rec;
+	struct ocfs2_insert_type insert;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_dinode *di;
+
+leftright:
+	/*
+	 * Store a copy of the record on the stack - it might move
+	 * around as the tree is manipulated below.
+	 */
+	rec = path_leaf_el(path)->l_recs[split_index];
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	rightmost_el = &di->id2.i_list;
+
+	depth = le16_to_cpu(rightmost_el->l_tree_depth);
+	if (depth) {
+		BUG_ON(!(*last_eb_bh));
+		eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
+		rightmost_el = &eb->h_list;
+	}
+
+	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
+	    le16_to_cpu(rightmost_el->l_count)) {
+		int old_depth = depth;
+
+		ret = ocfs2_grow_tree(inode, handle, di_bh, &depth, last_eb_bh,
+				      meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (old_depth != depth) {
+			eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
+			rightmost_el = &eb->h_list;
+		}
+	}
+
+	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
+	insert.ins_appending = APPEND_NONE;
+	insert.ins_contig = CONTIG_NONE;
+	insert.ins_free_records = le16_to_cpu(rightmost_el->l_count)
+		- le16_to_cpu(rightmost_el->l_next_free_rec);
+	insert.ins_tree_depth = depth;
+
+	insert_range = le32_to_cpu(split_rec.e_cpos) +
+		le16_to_cpu(split_rec.e_leaf_clusters);
+	rec_range = le32_to_cpu(rec.e_cpos) +
+		le16_to_cpu(rec.e_leaf_clusters);
+
+	if (split_rec.e_cpos == rec.e_cpos) {
+		insert.ins_split = SPLIT_LEFT;
+	} else if (insert_range == rec_range) {
+		insert.ins_split = SPLIT_RIGHT;
+	} else {
+		/*
+		 * Left/right split. We fake this as a right split
+		 * first and then make a second pass as a left split.
 		 */
-		ocfs2_figure_appending_type(insert, el, insert_rec);
+		insert.ins_split = SPLIT_RIGHT;
+
+		ocfs2_make_right_split_rec(inode->i_sb, &tmprec, insert_range,
+					   &rec);
+
+		split_rec = tmprec;
+
+		BUG_ON(do_leftright);
+		do_leftright = 1;
+	}
+
+	ret = ocfs2_do_insert_extent(inode, handle, di_bh, &split_rec,
+				     &insert);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
 	}
 
+	if (do_leftright == 1) {
+		u32 cpos;
+		struct ocfs2_extent_list *el;
+
+		do_leftright++;
+		split_rec = *orig_split_rec;
+
+		ocfs2_reinit_path(path, 1);
+
+		cpos = le32_to_cpu(split_rec.e_cpos);
+		ret = ocfs2_find_path(inode, path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		el = path_leaf_el(path);
+		split_index = ocfs2_search_extent_list(el, cpos);
+		goto leftright;
+	}
 out:
-	ocfs2_free_path(path);
 
-	if (ret == 0)
-		*last_eb_bh = bh;
+	return ret;
+}
+
+/*
+ * Mark part or all of the extent record at split_index in the leaf
+ * pointed to by path as written. This removes the unwritten
+ * extent flag.
+ *
+ * Care is taken to handle contiguousness so as to not grow the tree.
+ *
+ * meta_ac is not strictly necessary - we only truly need it if growth
+ * 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.
+ *
+ * This code is optimized for readability - several passes might be
+ * made over certain portions of the tree. All of those blocks will
+ * have been brought into cache (and pinned via the journal), so the
+ * extra overhead is not expressed in terms of disk reads.
+ */
+static int __ocfs2_mark_extent_written(struct inode *inode,
+				       struct buffer_head *di_bh,
+				       handle_t *handle,
+				       struct ocfs2_path *path,
+				       int split_index,
+				       struct ocfs2_extent_rec *split_rec,
+				       struct ocfs2_alloc_context *meta_ac,
+				       struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0;
+	struct ocfs2_extent_list *el = path_leaf_el(path);
+	struct buffer_head *eb_bh, *last_eb_bh = NULL;
+	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
+	struct ocfs2_merge_ctxt ctxt;
+	struct ocfs2_extent_list *rightmost_el;
+
+	if (!rec->e_flags & OCFS2_EXT_UNWRITTEN) {
+		ret = -EIO;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
+	    ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
+	     (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
+		ret = -EIO;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	eb_bh = path_leaf_bh(path);
+	ret = ocfs2_journal_access(handle, inode, eb_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, el,
+							    split_index,
+							    split_rec);
+
+	/*
+	 * The core merge / split code wants to know how much room is
+	 * left in this inodes allocation tree, so we pass the
+	 * rightmost extent list.
+	 */
+	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),
+				       &last_eb_bh, OCFS2_BH_CACHED, inode);
+		if (ret) {
+			mlog_exit(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
+			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
+			ret = -EROFS;
+			goto out;
+		}
+
+		rightmost_el = &eb->h_list;
+	} else
+		rightmost_el = path_root_el(path);
+
+	ctxt.c_used_tail_recs = le16_to_cpu(rightmost_el->l_next_free_rec);
+	if (ctxt.c_used_tail_recs > 0 &&
+	    ocfs2_is_empty_extent(&rightmost_el->l_recs[0]))
+		ctxt.c_used_tail_recs--;
+
+	if (rec->e_cpos == split_rec->e_cpos &&
+	    rec->e_leaf_clusters == split_rec->e_leaf_clusters)
+		ctxt.c_split_covers_rec = 1;
 	else
-		brelse(bh);
+		ctxt.c_split_covers_rec = 0;
+
+	ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
+
+	mlog(0, "index: %d, contig: %u, used_tail_recs: %u, "
+	     "has_empty: %u, split_covers: %u\n", split_index,
+	     ctxt.c_contig_type, ctxt.c_used_tail_recs,
+	     ctxt.c_has_empty_extent, ctxt.c_split_covers_rec);
+
+	if (ctxt.c_contig_type == CONTIG_NONE) {
+		if (ctxt.c_split_covers_rec)
+			el->l_recs[split_index] = *split_rec;
+		else
+			ret = ocfs2_split_and_insert(inode, handle, path, di_bh,
+						     &last_eb_bh, split_index,
+						     split_rec, meta_ac);
+		if (ret)
+			mlog_errno(ret);
+	} else {
+		ret = ocfs2_try_to_merge_extent(inode, handle, path,
+						split_index, split_rec,
+						dealloc, &ctxt);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+	ocfs2_journal_dirty(handle, eb_bh);
+
+out:
+	brelse(last_eb_bh);
 	return ret;
 }
 
 /*
- * Insert an extent into an inode btree.
+ * Mark the already-existing extent at cpos as written for len clusters.
  *
- * The caller needs to update fe->i_clusters
+ * If the existing extent is larger than the request, initiate a
+ * split. An attempt will be made at merging with adjacent extents.
+ *
+ * The caller is responsible for passing down meta_ac if we'll need it.
  */
-int ocfs2_insert_extent(struct ocfs2_super *osb,
-			handle_t *handle,
-			struct inode *inode,
-			struct buffer_head *fe_bh,
-			u32 cpos,
-			u64 start_blk,
-			u32 new_clusters,
-			struct ocfs2_alloc_context *meta_ac)
+int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *di_bh,
+			      handle_t *handle, u32 cpos, u32 len, u32 phys,
+			      struct ocfs2_alloc_context *meta_ac,
+			      struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	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;
+
+	mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n",
+	     inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno);
+
+	if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents "
+			    "that are being written to, but the feature bit "
+			    "is not set in the super block.",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
+		ret = -EROFS;
+		goto out;
+	}
+
+	/*
+	 * XXX: This should be fixed up so that we just re-insert the
+	 * next extent records.
+	 */
+	ocfs2_extent_map_trunc(inode, 0);
+
+	left_path = ocfs2_new_inode_path(di_bh);
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(inode, left_path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	el = path_leaf_el(left_path);
+
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has an extent at cpos %u which can no "
+			    "longer be found.\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
+		ret = -EROFS;
+		goto out;
+	}
+
+	memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
+	split_rec.e_cpos = cpu_to_le32(cpos);
+	split_rec.e_leaf_clusters = cpu_to_le16(len);
+	split_rec.e_blkno = cpu_to_le64(start_blkno);
+	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);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	ocfs2_free_path(left_path);
+	return ret;
+}
+
+static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
+			    handle_t *handle, struct ocfs2_path *path,
+			    int index, u32 new_range,
+			    struct ocfs2_alloc_context *meta_ac)
 {
-	int status, shift;
+	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 buffer_head *bh = NULL;
-	struct ocfs2_insert_type insert = {0, };
-	struct ocfs2_extent_rec rec;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *rightmost_el, *el;
+	struct ocfs2_extent_rec split_rec;
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_insert_type insert;
+
+	/*
+	 * Setup the record to split before we grow the tree.
+	 */
+	el = path_leaf_el(path);
+	rec = &el->l_recs[index];
+	ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec);
+
+	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),
+				       &last_eb_bh, OCFS2_BH_CACHED, inode);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+		rightmost_el = &eb->h_list;
+	} else
+		rightmost_el = path_leaf_el(path);
+
+	credits += path->p_tree_depth + ocfs2_extend_meta_needed(di);
+	ret = ocfs2_extend_trans(handle, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
+	    le16_to_cpu(rightmost_el->l_count)) {
+		int old_depth = depth;
+
+		ret = ocfs2_grow_tree(inode, handle, di_bh, &depth, &last_eb_bh,
+				      meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (old_depth != depth) {
+			eb = (struct ocfs2_extent_block *)last_eb_bh->b_data;
+			rightmost_el = &eb->h_list;
+		}
+	}
+
+	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
+	insert.ins_appending = APPEND_NONE;
+	insert.ins_contig = CONTIG_NONE;
+	insert.ins_split = SPLIT_RIGHT;
+	insert.ins_free_records = le16_to_cpu(rightmost_el->l_count)
+		- le16_to_cpu(rightmost_el->l_next_free_rec);
+	insert.ins_tree_depth = depth;
+
+	ret = ocfs2_do_insert_extent(inode, handle, di_bh, &split_rec, &insert);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	brelse(last_eb_bh);
+	return ret;
+}
+
+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)
+{
+	int ret;
+	u32 left_cpos, rec_range, trunc_range;
+	int wants_rotate = 0, is_rightmost_tree_rec = 0;
+	struct super_block *sb = inode->i_sb;
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_extent_list *el = path_leaf_el(path);
+	struct ocfs2_extent_rec *rec;
+	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);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		index--;
+	}
+
+	if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
+	    path->p_tree_depth) {
+		/*
+		 * Check whether this is the rightmost tree record. If
+		 * we remove all of this record or part of its right
+		 * edge then an update of the record lengths above it
+		 * will be required.
+		 */
+		eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
+		if (eb->h_next_leaf_blk == 0)
+			is_rightmost_tree_rec = 1;
+	}
+
+	rec = &el->l_recs[index];
+	if (index == 0 && path->p_tree_depth &&
+	    le32_to_cpu(rec->e_cpos) == cpos) {
+		/*
+		 * Changing the leftmost offset (via partial or whole
+		 * record truncate) of an interior (or rightmost) path
+		 * means we have to update the subtree that is formed
+		 * by this leaf and the one to it's left.
+		 *
+		 * There are two cases we can skip:
+		 *   1) Path is the leftmost one in our inode tree.
+		 *   2) The leaf is rightmost and will be empty after
+		 *      we remove the extent record - the rotate code
+		 *      knows how to update the newly formed edge.
+		 */
+
+		ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path,
+						    &left_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
+			left_path = ocfs2_new_path(path_root_bh(path),
+						   path_root_el(path));
+			if (!left_path) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_find_path(inode, left_path, left_cpos);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+	}
+
+	ret = ocfs2_extend_rotate_transaction(handle, 0,
+					      handle->h_buffer_credits,
+					      path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_path(inode, handle, path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_path(inode, handle, left_path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+	trunc_range = cpos + len;
+
+	if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
+		int next_free;
+
+		memset(rec, 0, sizeof(*rec));
+		ocfs2_cleanup_merge(el, index);
+		wants_rotate = 1;
+
+		next_free = le16_to_cpu(el->l_next_free_rec);
+		if (is_rightmost_tree_rec && next_free > 1) {
+			/*
+			 * We skip the edge update if this path will
+			 * be deleted by the rotate code.
+			 */
+			rec = &el->l_recs[next_free - 1];
+			ocfs2_adjust_rightmost_records(inode, handle, path,
+						       rec);
+		}
+	} else if (le32_to_cpu(rec->e_cpos) == cpos) {
+		/* Remove leftmost portion of the record. */
+		le32_add_cpu(&rec->e_cpos, len);
+		le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
+		le16_add_cpu(&rec->e_leaf_clusters, -len);
+	} else if (rec_range == trunc_range) {
+		/* Remove rightmost portion of the record */
+		le16_add_cpu(&rec->e_leaf_clusters, -len);
+		if (is_rightmost_tree_rec)
+			ocfs2_adjust_rightmost_records(inode, handle, path, rec);
+	} else {
+		/* Caller should have trapped this. */
+		mlog(ML_ERROR, "Inode %llu: Invalid record truncate: (%u, %u) "
+		     "(%u, %u)\n", (unsigned long long)OCFS2_I(inode)->ip_blkno,
+		     le32_to_cpu(rec->e_cpos),
+		     le16_to_cpu(rec->e_leaf_clusters), cpos, len);
+		BUG();
+	}
+
+	if (left_path) {
+		int subtree_index;
+
+		subtree_index = ocfs2_find_subtree_root(inode, left_path, path);
+		ocfs2_complete_edge_insert(inode, handle, left_path, path,
+					   subtree_index);
+	}
+
+	ocfs2_journal_dirty(handle, path_leaf_bh(path));
+
+	ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out:
+	ocfs2_free_path(left_path);
+	return ret;
+}
 
-	mlog(0, "add %u clusters at position %u to inode %llu\n",
-	     new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno);
+int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
+			u32 cpos, u32 len, handle_t *handle,
+			struct ocfs2_alloc_context *meta_ac,
+			struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, index;
+	u32 rec_range, trunc_range;
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_path *path;
 
-	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
-			(OCFS2_I(inode)->ip_clusters != cpos),
-			"Device %s, asking for sparse allocation: inode %llu, "
-			"cpos %u, clusters %u\n",
-			osb->dev_str,
-			(unsigned long long)OCFS2_I(inode)->ip_blkno, cpos,
-			OCFS2_I(inode)->ip_clusters);
+	ocfs2_extent_map_trunc(inode, 0);
 
-	memset(&rec, 0, sizeof(rec));
-	rec.e_cpos = cpu_to_le32(cpos);
-	rec.e_blkno = cpu_to_le64(start_blk);
-	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
+	path = ocfs2_new_inode_path(di_bh);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
 
-	status = ocfs2_figure_insert_type(inode, fe_bh, &last_eb_bh, &rec,
-					  &insert);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
+	ret = ocfs2_find_path(inode, path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
 	}
 
-	mlog(0, "Insert.appending: %u, Insert.Contig: %u, "
-	     "Insert.contig_index: %d, Insert.free_records: %d, "
-	     "Insert.tree_depth: %d\n",
-	     insert.ins_appending, insert.ins_contig, insert.ins_contig_index,
-	     insert.ins_free_records, insert.ins_tree_depth);
+	el = path_leaf_el(path);
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has an extent at cpos %u which can no "
+			    "longer be found.\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
+		ret = -EROFS;
+		goto out;
+	}
 
 	/*
-	 * Avoid growing the tree unless we're out of records and the
-	 * insert type requres one.
+	 * We have 3 cases of extent removal:
+	 *   1) Range covers the entire extent rec
+	 *   2) Range begins or ends on one edge of the extent rec
+	 *   3) Range is in the middle of the extent rec (no shared edges)
+	 *
+	 * For case 1 we remove the extent rec and left rotate to
+	 * fill the hole.
+	 *
+	 * For case 2 we just shrink the existing extent rec, with a
+	 * tree update if the shrinking edge is also the edge of an
+	 * extent block.
+	 *
+	 * For case 3 we do a right split to turn the extent rec into
+	 * something case 2 can handle.
 	 */
-	if (insert.ins_contig != CONTIG_NONE || insert.ins_free_records)
-		goto out_add;
+	rec = &el->l_recs[index];
+	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+	trunc_range = cpos + len;
 
-	shift = ocfs2_find_branch_target(osb, inode, fe_bh, &bh);
-	if (shift < 0) {
-		status = shift;
-		mlog_errno(status);
-		goto bail;
-	}
+	BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
 
-	/* We traveled all the way to the bottom of the allocation tree
-	 * and didn't find room for any more extents - we need to add
-	 * another tree level */
-	if (shift) {
-		BUG_ON(bh);
-		mlog(0, "need to shift tree depth "
-		     "(current = %d)\n", insert.ins_tree_depth);
+	mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d "
+	     "(cpos %u, len %u)\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index,
+	     le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec));
 
-		/* ocfs2_shift_tree_depth will return us a buffer with
-		 * the new extent block (so we can pass that to
-		 * ocfs2_add_branch). */
-		status = ocfs2_shift_tree_depth(osb, handle, inode, fe_bh,
-						meta_ac, &bh);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
+	if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
+		ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
+					 cpos, len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else {
+		ret = ocfs2_split_tree(inode, di_bh, handle, path, index,
+				       trunc_range, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
 		}
-		insert.ins_tree_depth++;
-		/* Special case: we have room now if we shifted from
-		 * tree_depth 0 */
-		if (insert.ins_tree_depth == 1)
-			goto out_add;
-	}
 
-	/* call ocfs2_add_branch to add the final part of the tree with
-	 * the new data. */
-	mlog(0, "add branch. bh = %p\n", bh);
-	status = ocfs2_add_branch(osb, handle, inode, fe_bh, bh, last_eb_bh,
-				  meta_ac);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
-	}
+		/*
+		 * The split could have manipulated the tree enough to
+		 * move the record location, so we have to look for it again.
+		 */
+		ocfs2_reinit_path(path, 1);
 
-out_add:
-	/* Finally, we can add clusters. This might rotate the tree for us. */
-	status = ocfs2_do_insert_extent(inode, handle, fe_bh, &rec, &insert);
-	if (status < 0)
-		mlog_errno(status);
-	else
-		ocfs2_extent_map_insert_rec(inode, &rec);
+		ret = ocfs2_find_path(inode, path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 
-bail:
-	if (bh)
-		brelse(bh);
+		el = path_leaf_el(path);
+		index = ocfs2_search_extent_list(el, cpos);
+		if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %llu: split at cpos %u lost record.",
+				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+				    cpos);
+			ret = -EROFS;
+			goto out;
+		}
 
-	if (last_eb_bh)
-		brelse(last_eb_bh);
+		/*
+		 * Double check our values here. If anything is fishy,
+		 * it's easier to catch it at the top level.
+		 */
+		rec = &el->l_recs[index];
+		rec_range = le32_to_cpu(rec->e_cpos) +
+			ocfs2_rec_clusters(el, rec);
+		if (rec_range != trunc_range) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %llu: error after split at cpos %u"
+				    "trunc len %u, existing record is (%u,%u)",
+				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+				    cpos, len, le32_to_cpu(rec->e_cpos),
+				    ocfs2_rec_clusters(el, rec));
+			ret = -EROFS;
+			goto out;
+		}
 
-	mlog_exit(status);
-	return status;
+		ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
+					 cpos, len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	ocfs2_free_path(path);
+	return ret;
 }
 
-static inline int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
+int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
 {
 	struct buffer_head *tl_bh = osb->osb_tl_bh;
 	struct ocfs2_dinode *di;
@@ -2464,10 +4539,10 @@ static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
 	return current_tail == new_start;
 }
 
-static int ocfs2_truncate_log_append(struct ocfs2_super *osb,
-				     handle_t *handle,
-				     u64 start_blk,
-				     unsigned int num_clusters)
+int ocfs2_truncate_log_append(struct ocfs2_super *osb,
+			      handle_t *handle,
+			      u64 start_blk,
+			      unsigned int num_clusters)
 {
 	int status, index;
 	unsigned int start_cluster, tl_count;
@@ -2623,7 +4698,7 @@ bail:
 }
 
 /* Expects you to already be holding tl_inode->i_mutex */
-static int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
+int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
 {
 	int status;
 	unsigned int num_to_flush;
@@ -2957,6 +5032,219 @@ int ocfs2_truncate_log_init(struct ocfs2_super *osb)
 	return status;
 }
 
+/*
+ * Delayed de-allocation of suballocator blocks.
+ *
+ * Some sets of block de-allocations might involve multiple suballocator inodes.
+ *
+ * The locking for this can get extremely complicated, especially when
+ * the suballocator inodes to delete from aren't known until deep
+ * within an unrelated codepath.
+ *
+ * ocfs2_extent_block structures are a good example of this - an inode
+ * btree could have been grown by any number of nodes each allocating
+ * out of their own suballoc inode.
+ *
+ * These structures allow the delay of block de-allocation until a
+ * later time, when locking of multiple cluster inodes won't cause
+ * deadlock.
+ */
+
+/*
+ * Describes a single block free from a suballocator
+ */
+struct ocfs2_cached_block_free {
+	struct ocfs2_cached_block_free		*free_next;
+	u64					free_blk;
+	unsigned int				free_bit;
+};
+
+struct ocfs2_per_slot_free_list {
+	struct ocfs2_per_slot_free_list		*f_next_suballocator;
+	int					f_inode_type;
+	int					f_slot;
+	struct ocfs2_cached_block_free		*f_first;
+};
+
+static int ocfs2_free_cached_items(struct ocfs2_super *osb,
+				   int sysfile_type,
+				   int slot,
+				   struct ocfs2_cached_block_free *head)
+{
+	int ret;
+	u64 bg_blkno;
+	handle_t *handle;
+	struct inode *inode;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_cached_block_free *tmp;
+
+	inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
+	if (!inode) {
+		ret = -EINVAL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mutex_lock(&inode->i_mutex);
+
+	ret = ocfs2_meta_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_mutex;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	while (head) {
+		bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
+						      head->free_bit);
+		mlog(0, "Free bit: (bit %u, blkno %llu)\n",
+		     head->free_bit, (unsigned long long)head->free_blk);
+
+		ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
+					       head->free_bit, bg_blkno, 1);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_journal;
+		}
+
+		ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_journal;
+		}
+
+		tmp = head;
+		head = head->free_next;
+		kfree(tmp);
+	}
+
+out_journal:
+	ocfs2_commit_trans(osb, handle);
+
+out_unlock:
+	ocfs2_meta_unlock(inode, 1);
+	brelse(di_bh);
+out_mutex:
+	mutex_unlock(&inode->i_mutex);
+	iput(inode);
+out:
+	while(head) {
+		/* Premature exit may have left some dangling items. */
+		tmp = head;
+		head = head->free_next;
+		kfree(tmp);
+	}
+
+	return ret;
+}
+
+int ocfs2_run_deallocs(struct ocfs2_super *osb,
+		       struct ocfs2_cached_dealloc_ctxt *ctxt)
+{
+	int ret = 0, ret2;
+	struct ocfs2_per_slot_free_list *fl;
+
+	if (!ctxt)
+		return 0;
+
+	while (ctxt->c_first_suballocator) {
+		fl = ctxt->c_first_suballocator;
+
+		if (fl->f_first) {
+			mlog(0, "Free items: (type %u, slot %d)\n",
+			     fl->f_inode_type, fl->f_slot);
+			ret2 = ocfs2_free_cached_items(osb, fl->f_inode_type,
+						       fl->f_slot, fl->f_first);
+			if (ret2)
+				mlog_errno(ret2);
+			if (!ret)
+				ret = ret2;
+		}
+
+		ctxt->c_first_suballocator = fl->f_next_suballocator;
+		kfree(fl);
+	}
+
+	return ret;
+}
+
+static struct ocfs2_per_slot_free_list *
+ocfs2_find_per_slot_free_list(int type,
+			      int slot,
+			      struct ocfs2_cached_dealloc_ctxt *ctxt)
+{
+	struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
+
+	while (fl) {
+		if (fl->f_inode_type == type && fl->f_slot == slot)
+			return fl;
+
+		fl = fl->f_next_suballocator;
+	}
+
+	fl = kmalloc(sizeof(*fl), GFP_NOFS);
+	if (fl) {
+		fl->f_inode_type = type;
+		fl->f_slot = slot;
+		fl->f_first = NULL;
+		fl->f_next_suballocator = ctxt->c_first_suballocator;
+
+		ctxt->c_first_suballocator = fl;
+	}
+	return fl;
+}
+
+static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
+				     int type, int slot, u64 blkno,
+				     unsigned int bit)
+{
+	int ret;
+	struct ocfs2_per_slot_free_list *fl;
+	struct ocfs2_cached_block_free *item;
+
+	fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
+	if (fl == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	item = kmalloc(sizeof(*item), GFP_NOFS);
+	if (item == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n",
+	     type, slot, bit, (unsigned long long)blkno);
+
+	item->free_blk = blkno;
+	item->free_bit = bit;
+	item->free_next = fl->f_first;
+
+	fl->f_first = item;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
+					 struct ocfs2_extent_block *eb)
+{
+	return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
+					 le16_to_cpu(eb->h_suballoc_slot),
+					 le64_to_cpu(eb->h_blkno),
+					 le16_to_cpu(eb->h_suballoc_bit));
+}
+
 /* This function will figure out whether the currently last extent
  * block will be deleted, and if it will, what the new last extent
  * block will be so we can update his h_next_leaf_blk field, as well
@@ -3238,27 +5526,10 @@ delete:
 			BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos));
 			BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno));
 
-			if (le16_to_cpu(eb->h_suballoc_slot) == 0) {
-				/*
-				 * This code only understands how to
-				 * lock the suballocator in slot 0,
-				 * which is fine because allocation is
-				 * only ever done out of that
-				 * suballocator too. A future version
-				 * might change that however, so avoid
-				 * a free if we don't know how to
-				 * handle it. This way an fs incompat
-				 * bit will not be necessary.
-				 */
-				ret = ocfs2_free_extent_block(handle,
-							      tc->tc_ext_alloc_inode,
-							      tc->tc_ext_alloc_bh,
-							      eb);
-
-				/* An error here is not fatal. */
-				if (ret < 0)
-					mlog_errno(ret);
-			}
+			ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb);
+			/* An error here is not fatal. */
+			if (ret < 0)
+				mlog_errno(ret);
 		} else {
 			deleted_eb = 0;
 		}
@@ -3397,9 +5668,9 @@ static int ocfs2_ordered_zero_func(handle_t *handle, struct buffer_head *bh)
 	return ocfs2_journal_dirty_data(handle, bh);
 }
 
-static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t isize,
-				     struct page **pages, int numpages,
-				     u64 phys, handle_t *handle)
+static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
+				     loff_t end, struct page **pages,
+				     int numpages, u64 phys, handle_t *handle)
 {
 	int i, ret, partial = 0;
 	void *kaddr;
@@ -3412,26 +5683,14 @@ static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t isize,
 	if (numpages == 0)
 		goto out;
 
-	from = isize & (PAGE_CACHE_SIZE - 1); /* 1st page offset */
-	if (PAGE_CACHE_SHIFT > OCFS2_SB(sb)->s_clustersize_bits) {
-		/*
-		 * Since 'from' has been capped to a value below page
-		 * size, this calculation won't be able to overflow
-		 * 'to'
-		 */
-		to = ocfs2_align_bytes_to_clusters(sb, from);
-
-		/*
-		 * The truncate tail in this case should never contain
-		 * more than one page at maximum. The loop below also
-		 * assumes this.
-		 */
-		BUG_ON(numpages != 1);
-	}
-
+	to = PAGE_CACHE_SIZE;
 	for(i = 0; i < numpages; i++) {
 		page = pages[i];
 
+		from = start & (PAGE_CACHE_SIZE - 1);
+		if ((end >> PAGE_CACHE_SHIFT) == page->index)
+			to = end & (PAGE_CACHE_SIZE - 1);
+
 		BUG_ON(from > PAGE_CACHE_SIZE);
 		BUG_ON(to > PAGE_CACHE_SIZE);
 
@@ -3468,10 +5727,7 @@ static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t isize,
 
 		flush_dcache_page(page);
 
-		/*
-		 * Every page after the 1st one should be completely zero'd.
-		 */
-		from = 0;
+		start = (page->index + 1) << PAGE_CACHE_SHIFT;
 	}
 out:
 	if (pages) {
@@ -3484,24 +5740,26 @@ out:
 	}
 }
 
-static int ocfs2_grab_eof_pages(struct inode *inode, loff_t isize, struct page **pages,
-				int *num, u64 *phys)
+static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
+				struct page **pages, int *num, u64 *phys)
 {
 	int i, numpages = 0, ret = 0;
-	unsigned int csize = OCFS2_SB(inode->i_sb)->s_clustersize;
 	unsigned int ext_flags;
 	struct super_block *sb = inode->i_sb;
 	struct address_space *mapping = inode->i_mapping;
 	unsigned long index;
-	u64 next_cluster_bytes;
+	loff_t last_page_bytes;
 
 	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
+	BUG_ON(start > end);
 
-	/* Cluster boundary, so we don't need to grab any pages. */
-	if ((isize & (csize - 1)) == 0)
+	if (start == end)
 		goto out;
 
-	ret = ocfs2_extent_map_get_blocks(inode, isize >> sb->s_blocksize_bits,
+	BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
+	       (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
+
+	ret = ocfs2_extent_map_get_blocks(inode, start >> sb->s_blocksize_bits,
 					  phys, NULL, &ext_flags);
 	if (ret) {
 		mlog_errno(ret);
@@ -3517,8 +5775,8 @@ static int ocfs2_grab_eof_pages(struct inode *inode, loff_t isize, struct page *
 	if (ext_flags & OCFS2_EXT_UNWRITTEN)
 		goto out;
 
-	next_cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, isize);
-	index = isize >> PAGE_CACHE_SHIFT;
+	last_page_bytes = PAGE_ALIGN(end);
+	index = start >> PAGE_CACHE_SHIFT;
 	do {
 		pages[numpages] = grab_cache_page(mapping, index);
 		if (!pages[numpages]) {
@@ -3529,7 +5787,7 @@ static int ocfs2_grab_eof_pages(struct inode *inode, loff_t isize, struct page *
 
 		numpages++;
 		index++;
-	} while (index < (next_cluster_bytes >> PAGE_CACHE_SHIFT));
+	} while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
 
 out:
 	if (ret != 0) {
@@ -3558,11 +5816,10 @@ out:
  * otherwise block_write_full_page() will skip writeout of pages past
  * i_size. The new_i_size parameter is passed for this reason.
  */
-int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
-				 u64 new_i_size)
+int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
+				  u64 range_start, u64 range_end)
 {
 	int ret, numpages;
-	loff_t endbyte;
 	struct page **pages = NULL;
 	u64 phys;
 
@@ -3581,7 +5838,8 @@ int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
 		goto out;
 	}
 
-	ret = ocfs2_grab_eof_pages(inode, new_i_size, pages, &numpages, &phys);
+	ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
+				   &numpages, &phys);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;
@@ -3590,17 +5848,16 @@ int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
 	if (numpages == 0)
 		goto out;
 
-	ocfs2_zero_cluster_pages(inode, new_i_size, pages, numpages, phys,
-				 handle);
+	ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
+				 numpages, phys, handle);
 
 	/*
 	 * Initiate writeout of the pages we zero'd here. We don't
 	 * wait on them - the truncate_inode_pages() call later will
 	 * do that for us.
 	 */
-	endbyte = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
-	ret = do_sync_mapping_range(inode->i_mapping, new_i_size,
-				    endbyte - 1, SYNC_FILE_RANGE_WRITE);
+	ret = do_sync_mapping_range(inode->i_mapping, range_start,
+				    range_end - 1, SYNC_FILE_RANGE_WRITE);
 	if (ret)
 		mlog_errno(ret);
 
@@ -3631,8 +5888,6 @@ int ocfs2_commit_truncate(struct ocfs2_super *osb,
 
 	mlog_entry_void();
 
-	down_write(&OCFS2_I(inode)->ip_alloc_sem);
-
 	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
 						     i_size_read(inode));
 
@@ -3754,7 +6009,6 @@ start:
 	goto start;
 
 bail:
-	up_write(&OCFS2_I(inode)->ip_alloc_sem);
 
 	ocfs2_schedule_truncate_log_flush(osb, 1);
 
@@ -3764,6 +6018,8 @@ bail:
 	if (handle)
 		ocfs2_commit_trans(osb, handle);
 
+	ocfs2_run_deallocs(osb, &tc->tc_dealloc);
+
 	ocfs2_free_path(path);
 
 	/* This will drop the ext_alloc cluster lock for us */
@@ -3774,23 +6030,18 @@ bail:
 }
 
 /*
- * Expects the inode to already be locked. This will figure out which
- * inodes need to be locked and will put them on the returned truncate
- * context.
+ * Expects the inode to already be locked.
  */
 int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 			   struct inode *inode,
 			   struct buffer_head *fe_bh,
 			   struct ocfs2_truncate_context **tc)
 {
-	int status, metadata_delete, i;
+	int status;
 	unsigned int new_i_clusters;
 	struct ocfs2_dinode *fe;
 	struct ocfs2_extent_block *eb;
-	struct ocfs2_extent_list *el;
 	struct buffer_head *last_eb_bh = NULL;
-	struct inode *ext_alloc_inode = NULL;
-	struct buffer_head *ext_alloc_bh = NULL;
 
 	mlog_entry_void();
 
@@ -3810,12 +6061,9 @@ int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 		mlog_errno(status);
 		goto bail;
 	}
+	ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc);
 
-	metadata_delete = 0;
 	if (fe->id2.i_list.l_tree_depth) {
-		/* If we have a tree, then the truncate may result in
-		 * metadata deletes. Figure this out from the
-		 * rightmost leaf block.*/
 		status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
 					  &last_eb_bh, OCFS2_BH_CACHED, inode);
 		if (status < 0) {
@@ -3830,43 +6078,10 @@ int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 			status = -EIO;
 			goto bail;
 		}
-		el = &(eb->h_list);
-
-		i = 0;
-		if (ocfs2_is_empty_extent(&el->l_recs[0]))
-			i = 1;
-		/*
-		 * XXX: Should we check that next_free_rec contains
-		 * the extent?
-		 */
-		if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_i_clusters)
-			metadata_delete = 1;
 	}
 
 	(*tc)->tc_last_eb_bh = last_eb_bh;
 
-	if (metadata_delete) {
-		mlog(0, "Will have to delete metadata for this trunc. "
-		     "locking allocator.\n");
-		ext_alloc_inode = ocfs2_get_system_file_inode(osb, EXTENT_ALLOC_SYSTEM_INODE, 0);
-		if (!ext_alloc_inode) {
-			status = -ENOMEM;
-			mlog_errno(status);
-			goto bail;
-		}
-
-		mutex_lock(&ext_alloc_inode->i_mutex);
-		(*tc)->tc_ext_alloc_inode = ext_alloc_inode;
-
-		status = ocfs2_meta_lock(ext_alloc_inode, &ext_alloc_bh, 1);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
-		}
-		(*tc)->tc_ext_alloc_bh = ext_alloc_bh;
-		(*tc)->tc_ext_alloc_locked = 1;
-	}
-
 	status = 0;
 bail:
 	if (status < 0) {
@@ -3880,16 +6095,13 @@ bail:
 
 static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc)
 {
-	if (tc->tc_ext_alloc_inode) {
-		if (tc->tc_ext_alloc_locked)
-			ocfs2_meta_unlock(tc->tc_ext_alloc_inode, 1);
-
-		mutex_unlock(&tc->tc_ext_alloc_inode->i_mutex);
-		iput(tc->tc_ext_alloc_inode);
-	}
-
-	if (tc->tc_ext_alloc_bh)
-		brelse(tc->tc_ext_alloc_bh);
+	/*
+	 * The caller is responsible for completing deallocation
+	 * before freeing the context.
+	 */
+	if (tc->tc_dealloc.c_first_suballocator != NULL)
+		mlog(ML_NOTICE,
+		     "Truncate completion has non-empty dealloc context\n");
 
 	if (tc->tc_last_eb_bh)
 		brelse(tc->tc_last_eb_bh);

fs/ocfs2/alloc.h

@@ -34,7 +34,17 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 			u32 cpos,
 			u64 start_blk,
 			u32 new_clusters,
+			u8 flags,
 			struct ocfs2_alloc_context *meta_ac);
+struct ocfs2_cached_dealloc_ctxt;
+int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *di_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,
+			u32 cpos, u32 len, handle_t *handle,
+			struct ocfs2_alloc_context *meta_ac,
+			struct ocfs2_cached_dealloc_ctxt *dealloc);
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
 			   struct inode *inode,
 			   struct ocfs2_dinode *fe);
@@ -62,17 +72,41 @@ int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
 				      struct ocfs2_dinode **tl_copy);
 int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
 					 struct ocfs2_dinode *tl_copy);
+int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb);
+int ocfs2_truncate_log_append(struct ocfs2_super *osb,
+			      handle_t *handle,
+			      u64 start_blk,
+			      unsigned int num_clusters);
+int __ocfs2_flush_truncate_log(struct ocfs2_super *osb);
+
+/*
+ * Process local structure which describes the block unlinks done
+ * during an operation. This is populated via
+ * ocfs2_cache_block_dealloc().
+ *
+ * ocfs2_run_deallocs() should be called after the potentially
+ * de-allocating routines. No journal handles should be open, and most
+ * locks should have been dropped.
+ */
+struct ocfs2_cached_dealloc_ctxt {
+	struct ocfs2_per_slot_free_list		*c_first_suballocator;
+};
+static inline void ocfs2_init_dealloc_ctxt(struct ocfs2_cached_dealloc_ctxt *c)
+{
+	c->c_first_suballocator = NULL;
+}
+int ocfs2_run_deallocs(struct ocfs2_super *osb,
+		       struct ocfs2_cached_dealloc_ctxt *ctxt);
 
 struct ocfs2_truncate_context {
-	struct inode *tc_ext_alloc_inode;
-	struct buffer_head *tc_ext_alloc_bh;
+	struct ocfs2_cached_dealloc_ctxt tc_dealloc;
 	int tc_ext_alloc_locked; /* is it cluster locked? */
 	/* these get destroyed once it's passed to ocfs2_commit_truncate. */
 	struct buffer_head *tc_last_eb_bh;
 };
 
-int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
-				 u64 new_i_size);
+int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
+				  u64 range_start, u64 range_end);
 int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 			   struct inode *inode,
 			   struct buffer_head *fe_bh,
@@ -84,6 +118,7 @@ int ocfs2_commit_truncate(struct ocfs2_super *osb,
 
 int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el,
 		    u32 cpos, struct buffer_head **leaf_bh);
+int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster);
 
 /*
  * Helper function to look at the # of clusters in an extent record.

fs/ocfs2/aops.c

@@ -684,6 +684,8 @@ int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 	     bh = bh->b_this_page, block_start += bsize) {
 		block_end = block_start + bsize;
 
+		clear_buffer_new(bh);
+
 		/*
 		 * Ignore blocks outside of our i/o range -
 		 * they may belong to unallocated clusters.
@@ -698,9 +700,8 @@ int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 		 * For an allocating write with cluster size >= page
 		 * size, we always write the entire page.
 		 */
-
-		if (buffer_new(bh))
-			clear_buffer_new(bh);
+		if (new)
+			set_buffer_new(bh);
 
 		if (!buffer_mapped(bh)) {
 			map_bh(bh, inode->i_sb, *p_blkno);
@@ -711,7 +712,8 @@ int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 			if (!buffer_uptodate(bh))
 				set_buffer_uptodate(bh);
 		} else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
-		     (block_start < from || block_end > to)) {
+			   !buffer_new(bh) &&
+			   (block_start < from || block_end > to)) {
 			ll_rw_block(READ, 1, &bh);
 			*wait_bh++=bh;
 		}
@@ -738,18 +740,13 @@ int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 	bh = head;
 	block_start = 0;
 	do {
-		void *kaddr;
-
 		block_end = block_start + bsize;
 		if (block_end <= from)
 			goto next_bh;
 		if (block_start >= to)
 			break;
 
-		kaddr = kmap_atomic(page, KM_USER0);
-		memset(kaddr+block_start, 0, bh->b_size);
-		flush_dcache_page(page);
-		kunmap_atomic(kaddr, KM_USER0);
+		zero_user_page(page, block_start, bh->b_size, KM_USER0);
 		set_buffer_uptodate(bh);
 		mark_buffer_dirty(bh);
 
@@ -761,217 +758,240 @@ next_bh:
 	return ret;
 }
 
+#if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
+#define OCFS2_MAX_CTXT_PAGES	1
+#else
+#define OCFS2_MAX_CTXT_PAGES	(OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
+#endif
+
+#define OCFS2_MAX_CLUSTERS_PER_PAGE	(PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
+
 /*
- * This will copy user data from the buffer page in the splice
- * context.
- *
- * For now, we ignore SPLICE_F_MOVE as that would require some extra
- * communication out all the way to ocfs2_write().
+ * Describe the state of a single cluster to be written to.
  */
-int ocfs2_map_and_write_splice_data(struct inode *inode,
-				  struct ocfs2_write_ctxt *wc, u64 *p_blkno,
-				  unsigned int *ret_from, unsigned int *ret_to)
+struct ocfs2_write_cluster_desc {
+	u32		c_cpos;
+	u32		c_phys;
+	/*
+	 * Give this a unique field because c_phys eventually gets
+	 * filled.
+	 */
+	unsigned	c_new;
+	unsigned	c_unwritten;
+};
+
+static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc *d)
 {
-	int ret;
-	unsigned int to, from, cluster_start, cluster_end;
-	char *src, *dst;
-	struct ocfs2_splice_write_priv *sp = wc->w_private;
-	struct pipe_buffer *buf = sp->s_buf;
-	unsigned long bytes, src_from;
-	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	return d->c_new || d->c_unwritten;
+}
 
-	ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
-					&cluster_end);
+struct ocfs2_write_ctxt {
+	/* Logical cluster position / len of write */
+	u32				w_cpos;
+	u32				w_clen;
 
-	from = sp->s_offset;
-	src_from = sp->s_buf_offset;
-	bytes = wc->w_count;
+	struct ocfs2_write_cluster_desc	w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
 
-	if (wc->w_large_pages) {
-		/*
-		 * For cluster size < page size, we have to
-		 * calculate pos within the cluster and obey
-		 * the rightmost boundary.
-		 */
-		bytes = min(bytes, (unsigned long)(osb->s_clustersize
-				   - (wc->w_pos & (osb->s_clustersize - 1))));
-	}
-	to = from + bytes;
+	/*
+	 * This is true if page_size > cluster_size.
+	 *
+	 * It triggers a set of special cases during write which might
+	 * have to deal with allocating writes to partial pages.
+	 */
+	unsigned int			w_large_pages;
+
+	/*
+	 * Pages involved in this write.
+	 *
+	 * w_target_page is the page being written to by the user.
+	 *
+	 * w_pages is an array of pages which always contains
+	 * w_target_page, and in the case of an allocating write with
+	 * page_size < cluster size, it will contain zero'd and mapped
+	 * pages adjacent to w_target_page which need to be written
+	 * out in so that future reads from that region will get
+	 * zero's.
+	 */
+	struct page			*w_pages[OCFS2_MAX_CTXT_PAGES];
+	unsigned int			w_num_pages;
+	struct page			*w_target_page;
 
-	BUG_ON(from > PAGE_CACHE_SIZE);
-	BUG_ON(to > PAGE_CACHE_SIZE);
-	BUG_ON(from < cluster_start);
-	BUG_ON(to > cluster_end);
+	/*
+	 * ocfs2_write_end() uses this to know what the real range to
+	 * write in the target should be.
+	 */
+	unsigned int			w_target_from;
+	unsigned int			w_target_to;
 
-	if (wc->w_this_page_new)
-		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
-					    cluster_start, cluster_end, 1);
-	else
-		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
-					    from, to, 0);
-	if (ret) {
-		mlog_errno(ret);
-		goto out;
+	/*
+	 * We could use journal_current_handle() but this is cleaner,
+	 * IMHO -Mark
+	 */
+	handle_t			*w_handle;
+
+	struct buffer_head		*w_di_bh;
+
+	struct ocfs2_cached_dealloc_ctxt w_dealloc;
+};
+
+static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
+{
+	int i;
+
+	for(i = 0; i < wc->w_num_pages; i++) {
+		if (wc->w_pages[i] == NULL)
+			continue;
+
+		unlock_page(wc->w_pages[i]);
+		mark_page_accessed(wc->w_pages[i]);
+		page_cache_release(wc->w_pages[i]);
 	}
 
-	src = buf->ops->map(sp->s_pipe, buf, 1);
-	dst = kmap_atomic(wc->w_this_page, KM_USER1);
-	memcpy(dst + from, src + src_from, bytes);
-	kunmap_atomic(wc->w_this_page, KM_USER1);
-	buf->ops->unmap(sp->s_pipe, buf, src);
+	brelse(wc->w_di_bh);
+	kfree(wc);
+}
+
+static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
+				  struct ocfs2_super *osb, loff_t pos,
+				  unsigned len, struct buffer_head *di_bh)
+{
+	struct ocfs2_write_ctxt *wc;
+
+	wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
+	if (!wc)
+		return -ENOMEM;
 
-	wc->w_finished_copy = 1;
+	wc->w_cpos = pos >> osb->s_clustersize_bits;
+	wc->w_clen = ocfs2_clusters_for_bytes(osb->sb, len);
+	get_bh(di_bh);
+	wc->w_di_bh = di_bh;
 
-	*ret_from = from;
-	*ret_to = to;
-out:
+	if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
+		wc->w_large_pages = 1;
+	else
+		wc->w_large_pages = 0;
+
+	ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
+
+	*wcp = wc;
 
-	return bytes ? (unsigned int)bytes : ret;
+	return 0;
 }
 
 /*
- * This will copy user data from the iovec in the buffered write
- * context.
+ * If a page has any new buffers, zero them out here, and mark them uptodate
+ * and dirty so they'll be written out (in order to prevent uninitialised
+ * block data from leaking). And clear the new bit.
  */
-int ocfs2_map_and_write_user_data(struct inode *inode,
-				  struct ocfs2_write_ctxt *wc, u64 *p_blkno,
-				  unsigned int *ret_from, unsigned int *ret_to)
+static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
 {
-	int ret;
-	unsigned int to, from, cluster_start, cluster_end;
-	unsigned long bytes, src_from;
-	char *dst;
-	struct ocfs2_buffered_write_priv *bp = wc->w_private;
-	const struct iovec *cur_iov = bp->b_cur_iov;
-	char __user *buf;
-	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	unsigned int block_start, block_end;
+	struct buffer_head *head, *bh;
 
-	ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
-					&cluster_end);
+	BUG_ON(!PageLocked(page));
+	if (!page_has_buffers(page))
+		return;
 
-	buf = cur_iov->iov_base + bp->b_cur_off;
-	src_from = (unsigned long)buf & ~PAGE_CACHE_MASK;
+	bh = head = page_buffers(page);
+	block_start = 0;
+	do {
+		block_end = block_start + bh->b_size;
 
-	from = wc->w_pos & (PAGE_CACHE_SIZE - 1);
+		if (buffer_new(bh)) {
+			if (block_end > from && block_start < to) {
+				if (!PageUptodate(page)) {
+					unsigned start, end;
 
-	/*
-	 * This is a lot of comparisons, but it reads quite
-	 * easily, which is important here.
-	 */
-	/* Stay within the src page */
-	bytes = PAGE_SIZE - src_from;
-	/* Stay within the vector */
-	bytes = min(bytes,
-		    (unsigned long)(cur_iov->iov_len - bp->b_cur_off));
-	/* Stay within count */
-	bytes = min(bytes, (unsigned long)wc->w_count);
-	/*
-	 * For clustersize > page size, just stay within
-	 * target page, otherwise we have to calculate pos
-	 * within the cluster and obey the rightmost
-	 * boundary.
-	 */
-	if (wc->w_large_pages) {
-		/*
-		 * For cluster size < page size, we have to
-		 * calculate pos within the cluster and obey
-		 * the rightmost boundary.
-		 */
-		bytes = min(bytes, (unsigned long)(osb->s_clustersize
-				   - (wc->w_pos & (osb->s_clustersize - 1))));
-	} else {
-		/*
-		 * cluster size > page size is the most common
-		 * case - we just stay within the target page
-		 * boundary.
-		 */
-		bytes = min(bytes, PAGE_CACHE_SIZE - from);
-	}
+					start = max(from, block_start);
+					end = min(to, block_end);
 
-	to = from + bytes;
+					zero_user_page(page, start, end - start, KM_USER0);
+					set_buffer_uptodate(bh);
+				}
 
-	BUG_ON(from > PAGE_CACHE_SIZE);
-	BUG_ON(to > PAGE_CACHE_SIZE);
-	BUG_ON(from < cluster_start);
-	BUG_ON(to > cluster_end);
+				clear_buffer_new(bh);
+				mark_buffer_dirty(bh);
+			}
+		}
 
-	if (wc->w_this_page_new)
-		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
-					    cluster_start, cluster_end, 1);
-	else
-		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
-					    from, to, 0);
-	if (ret) {
-		mlog_errno(ret);
-		goto out;
-	}
+		block_start = block_end;
+		bh = bh->b_this_page;
+	} while (bh != head);
+}
 
-	dst = kmap(wc->w_this_page);
-	memcpy(dst + from, bp->b_src_buf + src_from, bytes);
-	kunmap(wc->w_this_page);
+/*
+ * Only called when we have a failure during allocating write to write
+ * zero's to the newly allocated region.
+ */
+static void ocfs2_write_failure(struct inode *inode,
+				struct ocfs2_write_ctxt *wc,
+				loff_t user_pos, unsigned user_len)
+{
+	int i;
+	unsigned from, to;
+	struct page *tmppage;
 
-	/*
-	 * XXX: This is slow, but simple. The caller of
-	 * ocfs2_buffered_write_cluster() is responsible for
-	 * passing through the iovecs, so it's difficult to
-	 * predict what our next step is in here after our
-	 * initial write. A future version should be pushing
-	 * that iovec manipulation further down.
-	 *
-	 * By setting this, we indicate that a copy from user
-	 * data was done, and subsequent calls for this
-	 * cluster will skip copying more data.
-	 */
-	wc->w_finished_copy = 1;
+	ocfs2_zero_new_buffers(wc->w_target_page, user_pos, user_len);
 
-	*ret_from = from;
-	*ret_to = to;
-out:
+	if (wc->w_large_pages) {
+		from = wc->w_target_from;
+		to = wc->w_target_to;
+	} else {
+		from = 0;
+		to = PAGE_CACHE_SIZE;
+	}
+
+	for(i = 0; i < wc->w_num_pages; i++) {
+		tmppage = wc->w_pages[i];
 
-	return bytes ? (unsigned int)bytes : ret;
+		if (ocfs2_should_order_data(inode))
+			walk_page_buffers(wc->w_handle, page_buffers(tmppage),
+					  from, to, NULL,
+					  ocfs2_journal_dirty_data);
+
+		block_commit_write(tmppage, from, to);
+	}
 }
 
-/*
- * Map, fill and write a page to disk.
- *
- * The work of copying data is done via callback.  Newly allocated
- * pages which don't take user data will be zero'd (set 'new' to
- * indicate an allocating write)
- *
- * Returns a negative error code or the number of bytes copied into
- * the page.
- */
-static int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
-				 u64 *p_blkno, struct page *page,
-				 struct ocfs2_write_ctxt *wc, int new)
+static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
+					struct ocfs2_write_ctxt *wc,
+					struct page *page, u32 cpos,
+					loff_t user_pos, unsigned user_len,
+					int new)
 {
-	int ret, copied = 0;
-	unsigned int from = 0, to = 0;
+	int ret;
+	unsigned int map_from = 0, map_to = 0;
 	unsigned int cluster_start, cluster_end;
-	unsigned int zero_from = 0, zero_to = 0;
+	unsigned int user_data_from = 0, user_data_to = 0;
 
-	ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), wc->w_cpos,
+	ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
 					&cluster_start, &cluster_end);
 
-	if ((wc->w_pos >> PAGE_CACHE_SHIFT) == page->index
-	    && !wc->w_finished_copy) {
-
-		wc->w_this_page = page;
-		wc->w_this_page_new = new;
-		ret = wc->w_write_data_page(inode, wc, p_blkno, &from, &to);
-		if (ret < 0) {
+	if (page == wc->w_target_page) {
+		map_from = user_pos & (PAGE_CACHE_SIZE - 1);
+		map_to = map_from + user_len;
+
+		if (new)
+			ret = ocfs2_map_page_blocks(page, p_blkno, inode,
+						    cluster_start, cluster_end,
+						    new);
+		else
+			ret = ocfs2_map_page_blocks(page, p_blkno, inode,
+						    map_from, map_to, new);
+		if (ret) {
 			mlog_errno(ret);
 			goto out;
 		}
 
-		copied = ret;
-
-		zero_from = from;
-		zero_to = to;
+		user_data_from = map_from;
+		user_data_to = map_to;
 		if (new) {
-			from = cluster_start;
-			to = cluster_end;
+			map_from = cluster_start;
+			map_to = cluster_end;
 		}
+
+		wc->w_target_from = map_from;
+		wc->w_target_to = map_to;
 	} else {
 		/*
 		 * If we haven't allocated the new page yet, we
@@ -980,11 +1000,11 @@ static int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
 		 */
 		BUG_ON(!new);
 
-		from = cluster_start;
-		to = cluster_end;
+		map_from = cluster_start;
+		map_to = cluster_end;
 
 		ret = ocfs2_map_page_blocks(page, p_blkno, inode,
-					    cluster_start, cluster_end, 1);
+					    cluster_start, cluster_end, new);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
@@ -1003,108 +1023,113 @@ static int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
 	 */
 	if (new && !PageUptodate(page))
 		ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
-					 wc->w_cpos, zero_from, zero_to);
+					 cpos, user_data_from, user_data_to);
 
 	flush_dcache_page(page);
 
-	if (ocfs2_should_order_data(inode)) {
-		ret = walk_page_buffers(handle,
-					page_buffers(page),
-					from, to, NULL,
-					ocfs2_journal_dirty_data);
-		if (ret < 0)
-			mlog_errno(ret);
-	}
-
-	/*
-	 * We don't use generic_commit_write() because we need to
-	 * handle our own i_size update.
-	 */
-	ret = block_commit_write(page, from, to);
-	if (ret)
-		mlog_errno(ret);
 out:
-
-	return copied ? copied : ret;
+	return ret;
 }
 
 /*
- * Do the actual write of some data into an inode. Optionally allocate
- * in order to fulfill the write.
- *
- * cpos is the logical cluster offset within the file to write at
- *
- * 'phys' is the physical mapping of that offset. a 'phys' value of
- * zero indicates that allocation is required. In this case, data_ac
- * and meta_ac should be valid (meta_ac can be null if metadata
- * allocation isn't required).
+ * This function will only grab one clusters worth of pages.
  */
-static ssize_t ocfs2_write(struct file *file, u32 phys, handle_t *handle,
-			   struct buffer_head *di_bh,
-			   struct ocfs2_alloc_context *data_ac,
-			   struct ocfs2_alloc_context *meta_ac,
-			   struct ocfs2_write_ctxt *wc)
+static int ocfs2_grab_pages_for_write(struct address_space *mapping,
+				      struct ocfs2_write_ctxt *wc,
+				      u32 cpos, loff_t user_pos, int new,
+				      struct page *mmap_page)
 {
-	int ret, i, numpages = 1, new;
-	unsigned int copied = 0;
-	u32 tmp_pos;
-	u64 v_blkno, p_blkno;
-	struct address_space *mapping = file->f_mapping;
+	int ret = 0, i;
+	unsigned long start, target_index, index;
 	struct inode *inode = mapping->host;
-	unsigned long index, start;
-	struct page **cpages;
 
-	new = phys == 0 ? 1 : 0;
+	target_index = user_pos >> PAGE_CACHE_SHIFT;
 
 	/*
 	 * Figure out how many pages we'll be manipulating here. For
 	 * non allocating write, we just change the one
 	 * page. Otherwise, we'll need a whole clusters worth.
 	 */
-	if (new)
-		numpages = ocfs2_pages_per_cluster(inode->i_sb);
-
-	cpages = kzalloc(sizeof(*cpages) * numpages, GFP_NOFS);
-	if (!cpages) {
-		ret = -ENOMEM;
-		mlog_errno(ret);
-		return ret;
-	}
-
-	/*
-	 * Fill our page array first. That way we've grabbed enough so
-	 * that we can zero and flush if we error after adding the
-	 * extent.
-	 */
 	if (new) {
-		start = ocfs2_align_clusters_to_page_index(inode->i_sb,
-							   wc->w_cpos);
-		v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, wc->w_cpos);
+		wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
+		start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
 	} else {
-		start = wc->w_pos >> PAGE_CACHE_SHIFT;
-		v_blkno = wc->w_pos >> inode->i_sb->s_blocksize_bits;
+		wc->w_num_pages = 1;
+		start = target_index;
 	}
 
-	for(i = 0; i < numpages; i++) {
+	for(i = 0; i < wc->w_num_pages; i++) {
 		index = start + i;
 
-		cpages[i] = find_or_create_page(mapping, index, GFP_NOFS);
-		if (!cpages[i]) {
-			ret = -ENOMEM;
-			mlog_errno(ret);
-			goto out;
+		if (index == target_index && mmap_page) {
+			/*
+			 * ocfs2_pagemkwrite() is a little different
+			 * and wants us to directly use the page
+			 * passed in.
+			 */
+			lock_page(mmap_page);
+
+			if (mmap_page->mapping != mapping) {
+				unlock_page(mmap_page);
+				/*
+				 * Sanity check - the locking in
+				 * ocfs2_pagemkwrite() should ensure
+				 * that this code doesn't trigger.
+				 */
+				ret = -EINVAL;
+				mlog_errno(ret);
+				goto out;
+			}
+
+			page_cache_get(mmap_page);
+			wc->w_pages[i] = mmap_page;
+		} else {
+			wc->w_pages[i] = find_or_create_page(mapping, index,
+							     GFP_NOFS);
+			if (!wc->w_pages[i]) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
 		}
+
+		if (index == target_index)
+			wc->w_target_page = wc->w_pages[i];
 	}
+out:
+	return ret;
+}
+
+/*
+ * Prepare a single cluster for write one cluster into the file.
+ */
+static int ocfs2_write_cluster(struct address_space *mapping,
+			       u32 phys, unsigned int unwritten,
+			       struct ocfs2_alloc_context *data_ac,
+			       struct ocfs2_alloc_context *meta_ac,
+			       struct ocfs2_write_ctxt *wc, u32 cpos,
+			       loff_t user_pos, unsigned user_len)
+{
+	int ret, i, new, should_zero = 0;
+	u64 v_blkno, p_blkno;
+	struct inode *inode = mapping->host;
+
+	new = phys == 0 ? 1 : 0;
+	if (new || unwritten)
+		should_zero = 1;
 
 	if (new) {
+		u32 tmp_pos;
+
 		/*
 		 * This is safe to call with the page locks - it won't take
 		 * any additional semaphores or cluster locks.
 		 */
-		tmp_pos = wc->w_cpos;
+		tmp_pos = cpos;
 		ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode,
-						 &tmp_pos, 1, di_bh, handle,
-						 data_ac, meta_ac, NULL);
+						 &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
@@ -1121,159 +1146,433 @@ static ssize_t ocfs2_write(struct file *file, u32 phys, handle_t *handle,
 			mlog_errno(ret);
 			goto out;
 		}
+	} else if (unwritten) {
+		ret = ocfs2_mark_extent_written(inode, wc->w_di_bh,
+						wc->w_handle, cpos, 1, phys,
+						meta_ac, &wc->w_dealloc);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
 	}
 
+	if (should_zero)
+		v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
+	else
+		v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
+
+	/*
+	 * The only reason this should fail is due to an inability to
+	 * find the extent added.
+	 */
 	ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
 					  NULL);
 	if (ret < 0) {
-
-		/*
-		 * XXX: Should we go readonly here?
-		 */
-
-		mlog_errno(ret);
+		ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
+			    "at logical block %llu",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)v_blkno);
 		goto out;
 	}
 
 	BUG_ON(p_blkno == 0);
 
-	for(i = 0; i < numpages; i++) {
-		ret = ocfs2_write_data_page(inode, handle, &p_blkno, cpages[i],
-					    wc, new);
-		if (ret < 0) {
-			mlog_errno(ret);
-			goto out;
+	for(i = 0; i < wc->w_num_pages; i++) {
+		int tmpret;
+
+		tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
+						      wc->w_pages[i], cpos,
+						      user_pos, user_len,
+						      should_zero);
+		if (tmpret) {
+			mlog_errno(tmpret);
+			if (ret == 0)
+				tmpret = ret;
 		}
-
-		copied += ret;
 	}
 
+	/*
+	 * We only have cleanup to do in case of allocating write.
+	 */
+	if (ret && new)
+		ocfs2_write_failure(inode, wc, user_pos, user_len);
+
 out:
-	for(i = 0; i < numpages; i++) {
-		unlock_page(cpages[i]);
-		mark_page_accessed(cpages[i]);
-		page_cache_release(cpages[i]);
+
+	return ret;
+}
+
+static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
+				       struct ocfs2_alloc_context *data_ac,
+				       struct ocfs2_alloc_context *meta_ac,
+				       struct ocfs2_write_ctxt *wc,
+				       loff_t pos, unsigned len)
+{
+	int ret, i;
+	struct ocfs2_write_cluster_desc *desc;
+
+	for (i = 0; i < wc->w_clen; i++) {
+		desc = &wc->w_desc[i];
+
+		ret = ocfs2_write_cluster(mapping, desc->c_phys,
+					  desc->c_unwritten, data_ac, meta_ac,
+					  wc, desc->c_cpos, pos, len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 	}
-	kfree(cpages);
 
-	return copied ? copied : ret;
+	ret = 0;
+out:
+	return ret;
 }
 
-static void ocfs2_write_ctxt_init(struct ocfs2_write_ctxt *wc,
-				  struct ocfs2_super *osb, loff_t pos,
-				  size_t count, ocfs2_page_writer *cb,
-				  void *cb_priv)
+/*
+ * ocfs2_write_end() wants to know which parts of the target page it
+ * should complete the write on. It's easiest to compute them ahead of
+ * time when a more complete view of the write is available.
+ */
+static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
+					struct ocfs2_write_ctxt *wc,
+					loff_t pos, unsigned len, int alloc)
 {
-	wc->w_count = count;
-	wc->w_pos = pos;
-	wc->w_cpos = wc->w_pos >> osb->s_clustersize_bits;
-	wc->w_finished_copy = 0;
+	struct ocfs2_write_cluster_desc *desc;
 
-	if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
-		wc->w_large_pages = 1;
-	else
-		wc->w_large_pages = 0;
+	wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
+	wc->w_target_to = wc->w_target_from + len;
 
-	wc->w_write_data_page = cb;
-	wc->w_private = cb_priv;
+	if (alloc == 0)
+		return;
+
+	/*
+	 * Allocating write - we may have different boundaries based
+	 * on page size and cluster size.
+	 *
+	 * NOTE: We can no longer compute one value from the other as
+	 * the actual write length and user provided length may be
+	 * different.
+	 */
+
+	if (wc->w_large_pages) {
+		/*
+		 * We only care about the 1st and last cluster within
+		 * our range and whether they should be zero'd or not. Either
+		 * value may be extended out to the start/end of a
+		 * newly allocated cluster.
+		 */
+		desc = &wc->w_desc[0];
+		if (ocfs2_should_zero_cluster(desc))
+			ocfs2_figure_cluster_boundaries(osb,
+							desc->c_cpos,
+							&wc->w_target_from,
+							NULL);
+
+		desc = &wc->w_desc[wc->w_clen - 1];
+		if (ocfs2_should_zero_cluster(desc))
+			ocfs2_figure_cluster_boundaries(osb,
+							desc->c_cpos,
+							NULL,
+							&wc->w_target_to);
+	} else {
+		wc->w_target_from = 0;
+		wc->w_target_to = PAGE_CACHE_SIZE;
+	}
 }
 
 /*
- * Write a cluster to an inode. The cluster may not be allocated yet,
- * in which case it will be. This only exists for buffered writes -
- * O_DIRECT takes a more "traditional" path through the kernel.
- *
- * The caller is responsible for incrementing pos, written counts, etc
+ * Populate each single-cluster write descriptor in the write context
+ * with information about the i/o to be done.
  *
- * For file systems that don't support sparse files, pre-allocation
- * and page zeroing up until cpos should be done prior to this
- * function call.
- *
- * Callers should be holding i_sem, and the rw cluster lock.
- *
- * Returns the number of user bytes written, or less than zero for
- * error.
+ * Returns the number of clusters that will have to be allocated, as
+ * well as a worst case estimate of the number of extent records that
+ * would have to be created during a write to an unwritten region.
  */
-ssize_t ocfs2_buffered_write_cluster(struct file *file, loff_t pos,
-				     size_t count, ocfs2_page_writer *actor,
-				     void *priv)
+static int ocfs2_populate_write_desc(struct inode *inode,
+				     struct ocfs2_write_ctxt *wc,
+				     unsigned int *clusters_to_alloc,
+				     unsigned int *extents_to_split)
+{
+	int ret;
+	struct ocfs2_write_cluster_desc *desc;
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+	u32 phys = 0;
+	int i;
+
+	*clusters_to_alloc = 0;
+	*extents_to_split = 0;
+
+	for (i = 0; i < wc->w_clen; i++) {
+		desc = &wc->w_desc[i];
+		desc->c_cpos = wc->w_cpos + i;
+
+		if (num_clusters == 0) {
+			/*
+			 * Need to look up the next extent record.
+			 */
+			ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
+						 &num_clusters, &ext_flags);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			/*
+			 * Assume worst case - that we're writing in
+			 * the middle of the extent.
+			 *
+			 * We can assume that the write proceeds from
+			 * left to right, in which case the extent
+			 * insert code is smart enough to coalesce the
+			 * next splits into the previous records created.
+			 */
+			if (ext_flags & OCFS2_EXT_UNWRITTEN)
+				*extents_to_split = *extents_to_split + 2;
+		} else if (phys) {
+			/*
+			 * Only increment phys if it doesn't describe
+			 * a hole.
+			 */
+			phys++;
+		}
+
+		desc->c_phys = phys;
+		if (phys == 0) {
+			desc->c_new = 1;
+			*clusters_to_alloc = *clusters_to_alloc + 1;
+		}
+		if (ext_flags & OCFS2_EXT_UNWRITTEN)
+			desc->c_unwritten = 1;
+
+		num_clusters--;
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+int ocfs2_write_begin_nolock(struct address_space *mapping,
+			     loff_t pos, unsigned len, unsigned flags,
+			     struct page **pagep, void **fsdata,
+			     struct buffer_head *di_bh, struct page *mmap_page)
 {
 	int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
-	ssize_t written = 0;
-	u32 phys;
-	struct inode *inode = file->f_mapping->host;
+	unsigned int clusters_to_alloc, extents_to_split;
+	struct ocfs2_write_ctxt *wc;
+	struct inode *inode = mapping->host;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-	struct buffer_head *di_bh = NULL;
 	struct ocfs2_dinode *di;
 	struct ocfs2_alloc_context *data_ac = NULL;
 	struct ocfs2_alloc_context *meta_ac = NULL;
 	handle_t *handle;
-	struct ocfs2_write_ctxt wc;
-
-	ocfs2_write_ctxt_init(&wc, osb, pos, count, actor, priv);
 
-	ret = ocfs2_meta_lock(inode, &di_bh, 1);
+	ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
 	if (ret) {
 		mlog_errno(ret);
-		goto out;
+		return ret;
 	}
-	di = (struct ocfs2_dinode *)di_bh->b_data;
-
-	/*
-	 * Take alloc sem here to prevent concurrent lookups. That way
-	 * the mapping, zeroing and tree manipulation within
-	 * ocfs2_write() will be safe against ->readpage(). This
-	 * should also serve to lock out allocation from a shared
-	 * writeable region.
-	 */
-	down_write(&OCFS2_I(inode)->ip_alloc_sem);
 
-	ret = ocfs2_get_clusters(inode, wc.w_cpos, &phys, NULL, NULL);
+	ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
+					&extents_to_split);
 	if (ret) {
 		mlog_errno(ret);
-		goto out_meta;
+		goto out;
 	}
 
-	/* phys == 0 means that allocation is required. */
-	if (phys == 0) {
-		ret = ocfs2_lock_allocators(inode, di, 1, &data_ac, &meta_ac);
+	di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
+
+	/*
+	 * We set w_target_from, w_target_to here so that
+	 * ocfs2_write_end() knows which range in the target page to
+	 * write out. An allocation requires that we write the entire
+	 * cluster range.
+	 */
+	if (clusters_to_alloc || extents_to_split) {
+		/*
+		 * XXX: We are stretching the limits of
+		 * 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);
 		if (ret) {
 			mlog_errno(ret);
-			goto out_meta;
+			goto out;
 		}
 
-		credits = ocfs2_calc_extend_credits(inode->i_sb, di, 1);
-	}
+		credits = ocfs2_calc_extend_credits(inode->i_sb, di,
+						    clusters_to_alloc);
 
-	ret = ocfs2_data_lock(inode, 1);
-	if (ret) {
-		mlog_errno(ret);
-		goto out_meta;
 	}
 
+	ocfs2_set_target_boundaries(osb, wc, pos, len,
+				    clusters_to_alloc + extents_to_split);
+
 	handle = ocfs2_start_trans(osb, credits);
 	if (IS_ERR(handle)) {
 		ret = PTR_ERR(handle);
 		mlog_errno(ret);
-		goto out_data;
+		goto out;
 	}
 
-	written = ocfs2_write(file, phys, handle, di_bh, data_ac,
-			      meta_ac, &wc);
-	if (written < 0) {
-		ret = written;
+	wc->w_handle = handle;
+
+	/*
+	 * We don't want this to fail in ocfs2_write_end(), so do it
+	 * here.
+	 */
+	ret = ocfs2_journal_access(handle, inode, wc->w_di_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
 		mlog_errno(ret);
 		goto out_commit;
 	}
 
-	ret = ocfs2_journal_access(handle, inode, di_bh,
-				   OCFS2_JOURNAL_ACCESS_WRITE);
+	/*
+	 * Fill our page array first. That way we've grabbed enough so
+	 * that we can zero and flush if we error after adding the
+	 * extent.
+	 */
+	ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos,
+					 clusters_to_alloc + extents_to_split,
+					 mmap_page);
 	if (ret) {
 		mlog_errno(ret);
 		goto out_commit;
 	}
 
-	pos += written;
+	ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
+					  len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	*pagep = wc->w_target_page;
+	*fsdata = wc;
+	return 0;
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	ocfs2_free_write_ctxt(wc);
+
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	return ret;
+}
+
+int ocfs2_write_begin(struct file *file, struct address_space *mapping,
+		      loff_t pos, unsigned len, unsigned flags,
+		      struct page **pagep, void **fsdata)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct inode *inode = mapping->host;
+
+	ret = ocfs2_meta_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/*
+	 * Take alloc sem here to prevent concurrent lookups. That way
+	 * the mapping, zeroing and tree manipulation within
+	 * ocfs2_write() will be safe against ->readpage(). This
+	 * should also serve to lock out allocation from a shared
+	 * writeable region.
+	 */
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ret = ocfs2_data_lock(inode, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_fail;
+	}
+
+	ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep,
+				       fsdata, di_bh, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_fail_data;
+	}
+
+	brelse(di_bh);
+
+	return 0;
+
+out_fail_data:
+	ocfs2_data_unlock(inode, 1);
+out_fail:
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	brelse(di_bh);
+	ocfs2_meta_unlock(inode, 1);
+
+	return ret;
+}
+
+int ocfs2_write_end_nolock(struct address_space *mapping,
+			   loff_t pos, unsigned len, unsigned copied,
+			   struct page *page, void *fsdata)
+{
+	int i;
+	unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
+	struct inode *inode = mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_write_ctxt *wc = fsdata;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
+	handle_t *handle = wc->w_handle;
+	struct page *tmppage;
+
+	if (unlikely(copied < len)) {
+		if (!PageUptodate(wc->w_target_page))
+			copied = 0;
+
+		ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
+				       start+len);
+	}
+	flush_dcache_page(wc->w_target_page);
+
+	for(i = 0; i < wc->w_num_pages; i++) {
+		tmppage = wc->w_pages[i];
+
+		if (tmppage == wc->w_target_page) {
+			from = wc->w_target_from;
+			to = wc->w_target_to;
+
+			BUG_ON(from > PAGE_CACHE_SIZE ||
+			       to > PAGE_CACHE_SIZE ||
+			       to < from);
+		} else {
+			/*
+			 * Pages adjacent to the target (if any) imply
+			 * a hole-filling write in which case we want
+			 * to flush their entire range.
+			 */
+			from = 0;
+			to = PAGE_CACHE_SIZE;
+		}
+
+		if (ocfs2_should_order_data(inode))
+			walk_page_buffers(wc->w_handle, page_buffers(tmppage),
+					  from, to, NULL,
+					  ocfs2_journal_dirty_data);
+
+		block_commit_write(tmppage, from, to);
+	}
+
+	pos += copied;
 	if (pos > inode->i_size) {
 		i_size_write(inode, pos);
 		mark_inode_dirty(inode);
@@ -1283,29 +1582,31 @@ ssize_t ocfs2_buffered_write_cluster(struct file *file, loff_t pos,
 	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 	di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
+	ocfs2_journal_dirty(handle, wc->w_di_bh);
 
-	ret = ocfs2_journal_dirty(handle, di_bh);
-	if (ret)
-		mlog_errno(ret);
-
-out_commit:
 	ocfs2_commit_trans(osb, handle);
 
-out_data:
-	ocfs2_data_unlock(inode, 1);
+	ocfs2_run_deallocs(osb, &wc->w_dealloc);
+
+	ocfs2_free_write_ctxt(wc);
+
+	return copied;
+}
+
+int ocfs2_write_end(struct file *file, struct address_space *mapping,
+		    loff_t pos, unsigned len, unsigned copied,
+		    struct page *page, void *fsdata)
+{
+	int ret;
+	struct inode *inode = mapping->host;
 
-out_meta:
+	ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
+
+	ocfs2_data_unlock(inode, 1);
 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
 	ocfs2_meta_unlock(inode, 1);
 
-out:
-	brelse(di_bh);
-	if (data_ac)
-		ocfs2_free_alloc_context(data_ac);
-	if (meta_ac)
-		ocfs2_free_alloc_context(meta_ac);
-
-	return written ? written : ret;
+	return ret;
 }
 
 const struct address_space_operations ocfs2_aops = {

fs/ocfs2/aops.h

@@ -42,57 +42,22 @@ int walk_page_buffers(	handle_t *handle,
 			int (*fn)(	handle_t *handle,
 					struct buffer_head *bh));
 
-struct ocfs2_write_ctxt;
-typedef int (ocfs2_page_writer)(struct inode *, struct ocfs2_write_ctxt *,
-				u64 *, unsigned int *, unsigned int *);
+int ocfs2_write_begin(struct file *file, struct address_space *mapping,
+		      loff_t pos, unsigned len, unsigned flags,
+		      struct page **pagep, void **fsdata);
 
-ssize_t ocfs2_buffered_write_cluster(struct file *file, loff_t pos,
-				     size_t count, ocfs2_page_writer *actor,
-				     void *priv);
+int ocfs2_write_end(struct file *file, struct address_space *mapping,
+		    loff_t pos, unsigned len, unsigned copied,
+		    struct page *page, void *fsdata);
 
-struct ocfs2_write_ctxt {
-	size_t				w_count;
-	loff_t				w_pos;
-	u32				w_cpos;
-	unsigned int			w_finished_copy;
+int ocfs2_write_end_nolock(struct address_space *mapping,
+			   loff_t pos, unsigned len, unsigned copied,
+			   struct page *page, void *fsdata);
 
-	/* This is true if page_size > cluster_size */
-	unsigned int			w_large_pages;
-
-	/* Filler callback and private data */
-	ocfs2_page_writer		*w_write_data_page;
-	void				*w_private;
-
-	/* Only valid for the filler callback */
-	struct page			*w_this_page;
-	unsigned int			w_this_page_new;
-};
-
-struct ocfs2_buffered_write_priv {
-	char				*b_src_buf;
-	const struct iovec		*b_cur_iov; /* Current iovec */
-	size_t				b_cur_off; /* Offset in the
-						    * current iovec */
-};
-int ocfs2_map_and_write_user_data(struct inode *inode,
-				  struct ocfs2_write_ctxt *wc,
-				  u64 *p_blkno,
-				  unsigned int *ret_from,
-				  unsigned int *ret_to);
-
-struct ocfs2_splice_write_priv {
-	struct splice_desc		*s_sd;
-	struct pipe_buffer		*s_buf;
-	struct pipe_inode_info		*s_pipe;
-	/* Neither offset value is ever larger than one page */
-	unsigned int			s_offset;
-	unsigned int			s_buf_offset;
-};
-int ocfs2_map_and_write_splice_data(struct inode *inode,
-				    struct ocfs2_write_ctxt *wc,
-				    u64 *p_blkno,
-				    unsigned int *ret_from,
-				    unsigned int *ret_to);
+int ocfs2_write_begin_nolock(struct address_space *mapping,
+			     loff_t pos, unsigned len, unsigned flags,
+			     struct page **pagep, void **fsdata,
+			     struct buffer_head *di_bh, struct page *mmap_page);
 
 /* all ocfs2_dio_end_io()'s fault */
 #define ocfs2_iocb_is_rw_locked(iocb) \

fs/ocfs2/cluster/heartbeat.c

@@ -1335,6 +1335,7 @@ static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
 	ret = wait_event_interruptible(o2hb_steady_queue,
 				atomic_read(&reg->hr_steady_iterations) == 0);
 	if (ret) {
+		/* We got interrupted (hello ptrace!).  Clean up */
 		spin_lock(&o2hb_live_lock);
 		hb_task = reg->hr_task;
 		reg->hr_task = NULL;
@@ -1345,7 +1346,16 @@ static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
 		goto out;
 	}
 
-	ret = count;
+	/* Ok, we were woken.  Make sure it wasn't by drop_item() */
+	spin_lock(&o2hb_live_lock);
+	hb_task = reg->hr_task;
+	spin_unlock(&o2hb_live_lock);
+
+	if (hb_task)
+		ret = count;
+	else
+		ret = -EIO;
+
 out:
 	if (filp)
 		fput(filp);
@@ -1523,6 +1533,15 @@ static void o2hb_heartbeat_group_drop_item(struct config_group *group,
 	if (hb_task)
 		kthread_stop(hb_task);
 
+	/*
+	 * If we're racing a dev_write(), we need to wake them.  They will
+	 * check reg->hr_task
+	 */
+	if (atomic_read(&reg->hr_steady_iterations) != 0) {
+		atomic_set(&reg->hr_steady_iterations, 0);
+		wake_up(&o2hb_steady_queue);
+	}
+
 	config_item_put(item);
 }
 
@@ -1665,7 +1684,67 @@ void o2hb_setup_callback(struct o2hb_callback_func *hc,
 }
 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
 
-int o2hb_register_callback(struct o2hb_callback_func *hc)
+static struct o2hb_region *o2hb_find_region(const char *region_uuid)
+{
+	struct o2hb_region *p, *reg = NULL;
+
+	assert_spin_locked(&o2hb_live_lock);
+
+	list_for_each_entry(p, &o2hb_all_regions, hr_all_item) {
+		if (!strcmp(region_uuid, config_item_name(&p->hr_item))) {
+			reg = p;
+			break;
+		}
+	}
+
+	return reg;
+}
+
+static int o2hb_region_get(const char *region_uuid)
+{
+	int ret = 0;
+	struct o2hb_region *reg;
+
+	spin_lock(&o2hb_live_lock);
+
+	reg = o2hb_find_region(region_uuid);
+	if (!reg)
+		ret = -ENOENT;
+	spin_unlock(&o2hb_live_lock);
+
+	if (ret)
+		goto out;
+
+	ret = o2nm_depend_this_node();
+	if (ret)
+		goto out;
+
+	ret = o2nm_depend_item(&reg->hr_item);
+	if (ret)
+		o2nm_undepend_this_node();
+
+out:
+	return ret;
+}
+
+static void o2hb_region_put(const char *region_uuid)
+{
+	struct o2hb_region *reg;
+
+	spin_lock(&o2hb_live_lock);
+
+	reg = o2hb_find_region(region_uuid);
+
+	spin_unlock(&o2hb_live_lock);
+
+	if (reg) {
+		o2nm_undepend_item(&reg->hr_item);
+		o2nm_undepend_this_node();
+	}
+}
+
+int o2hb_register_callback(const char *region_uuid,
+			   struct o2hb_callback_func *hc)
 {
 	struct o2hb_callback_func *tmp;
 	struct list_head *iter;
@@ -1681,6 +1760,12 @@ int o2hb_register_callback(struct o2hb_callback_func *hc)
 		goto out;
 	}
 
+	if (region_uuid) {
+		ret = o2hb_region_get(region_uuid);
+		if (ret)
+			goto out;
+	}
+
 	down_write(&o2hb_callback_sem);
 
 	list_for_each(iter, &hbcall->list) {
@@ -1702,16 +1787,21 @@ out:
 }
 EXPORT_SYMBOL_GPL(o2hb_register_callback);
 
-void o2hb_unregister_callback(struct o2hb_callback_func *hc)
+void o2hb_unregister_callback(const char *region_uuid,
+			      struct o2hb_callback_func *hc)
 {
 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
 
 	mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n",
 	     __builtin_return_address(0), hc);
 
+	/* XXX Can this happen _with_ a region reference? */
 	if (list_empty(&hc->hc_item))
 		return;
 
+	if (region_uuid)
+		o2hb_region_put(region_uuid);
+
 	down_write(&o2hb_callback_sem);
 
 	list_del_init(&hc->hc_item);

fs/ocfs2/cluster/heartbeat.h

@@ -69,8 +69,10 @@ void o2hb_setup_callback(struct o2hb_callback_func *hc,
 			 o2hb_cb_func *func,
 			 void *data,
 			 int priority);
-int o2hb_register_callback(struct o2hb_callback_func *hc);
-void o2hb_unregister_callback(struct o2hb_callback_func *hc);
+int o2hb_register_callback(const char *region_uuid,
+			   struct o2hb_callback_func *hc);
+void o2hb_unregister_callback(const char *region_uuid,
+			      struct o2hb_callback_func *hc);
 void o2hb_fill_node_map(unsigned long *map,
 			unsigned bytes);
 void o2hb_init(void);

fs/ocfs2/cluster/nodemanager.c

@@ -900,6 +900,46 @@ static struct o2nm_cluster_group o2nm_cluster_group = {
 	},
 };
 
+int o2nm_depend_item(struct config_item *item)
+{
+	return configfs_depend_item(&o2nm_cluster_group.cs_subsys, item);
+}
+
+void o2nm_undepend_item(struct config_item *item)
+{
+	configfs_undepend_item(&o2nm_cluster_group.cs_subsys, item);
+}
+
+int o2nm_depend_this_node(void)
+{
+	int ret = 0;
+	struct o2nm_node *local_node;
+
+	local_node = o2nm_get_node_by_num(o2nm_this_node());
+	if (!local_node) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = o2nm_depend_item(&local_node->nd_item);
+	o2nm_node_put(local_node);
+
+out:
+	return ret;
+}
+
+void o2nm_undepend_this_node(void)
+{
+	struct o2nm_node *local_node;
+
+	local_node = o2nm_get_node_by_num(o2nm_this_node());
+	BUG_ON(!local_node);
+
+	o2nm_undepend_item(&local_node->nd_item);
+	o2nm_node_put(local_node);
+}
+
+
 static void __exit exit_o2nm(void)
 {
 	if (ocfs2_table_header)
@@ -934,7 +974,7 @@ static int __init init_o2nm(void)
 		goto out_sysctl;
 
 	config_group_init(&o2nm_cluster_group.cs_subsys.su_group);
-	init_MUTEX(&o2nm_cluster_group.cs_subsys.su_sem);
+	mutex_init(&o2nm_cluster_group.cs_subsys.su_mutex);
 	ret = configfs_register_subsystem(&o2nm_cluster_group.cs_subsys);
 	if (ret) {
 		printk(KERN_ERR "nodemanager: Registration returned %d\n", ret);

fs/ocfs2/cluster/nodemanager.h

@@ -77,4 +77,9 @@ struct o2nm_node *o2nm_get_node_by_ip(__be32 addr);
 void o2nm_node_get(struct o2nm_node *node);
 void o2nm_node_put(struct o2nm_node *node);
 
+int o2nm_depend_item(struct config_item *item);
+void o2nm_undepend_item(struct config_item *item);
+int o2nm_depend_this_node(void);
+void o2nm_undepend_this_node(void);
+
 #endif /* O2CLUSTER_NODEMANAGER_H */

fs/ocfs2/cluster/tcp.c

@@ -261,14 +261,12 @@ out:
 
 static void o2net_complete_nodes_nsw(struct o2net_node *nn)
 {
-	struct list_head *iter, *tmp;
+	struct o2net_status_wait *nsw, *tmp;
 	unsigned int num_kills = 0;
-	struct o2net_status_wait *nsw;
 
 	assert_spin_locked(&nn->nn_lock);
 
-	list_for_each_safe(iter, tmp, &nn->nn_status_list) {
-		nsw = list_entry(iter, struct o2net_status_wait, ns_node_item);
+	list_for_each_entry_safe(nsw, tmp, &nn->nn_status_list, ns_node_item) {
 		o2net_complete_nsw_locked(nn, nsw, O2NET_ERR_DIED, 0);
 		num_kills++;
 	}
@@ -764,13 +762,10 @@ EXPORT_SYMBOL_GPL(o2net_register_handler);
 
 void o2net_unregister_handler_list(struct list_head *list)
 {
-	struct list_head *pos, *n;
-	struct o2net_msg_handler *nmh;
+	struct o2net_msg_handler *nmh, *n;
 
 	write_lock(&o2net_handler_lock);
-	list_for_each_safe(pos, n, list) {
-		nmh = list_entry(pos, struct o2net_msg_handler,
-				 nh_unregister_item);
+	list_for_each_entry_safe(nmh, n, list, nh_unregister_item) {
 		mlog(ML_TCP, "unregistering handler func %p type %u key %08x\n",
 		     nmh->nh_func, nmh->nh_msg_type, nmh->nh_key);
 		rb_erase(&nmh->nh_node, &o2net_handler_tree);
@@ -1638,8 +1633,8 @@ static void o2net_hb_node_up_cb(struct o2nm_node *node, int node_num,
 
 void o2net_unregister_hb_callbacks(void)
 {
-	o2hb_unregister_callback(&o2net_hb_up);
-	o2hb_unregister_callback(&o2net_hb_down);
+	o2hb_unregister_callback(NULL, &o2net_hb_up);
+	o2hb_unregister_callback(NULL, &o2net_hb_down);
 }
 
 int o2net_register_hb_callbacks(void)
@@ -1651,9 +1646,9 @@ int o2net_register_hb_callbacks(void)
 	o2hb_setup_callback(&o2net_hb_up, O2HB_NODE_UP_CB,
 			    o2net_hb_node_up_cb, NULL, O2NET_HB_PRI);
 
-	ret = o2hb_register_callback(&o2net_hb_up);
+	ret = o2hb_register_callback(NULL, &o2net_hb_up);
 	if (ret == 0)
-		ret = o2hb_register_callback(&o2net_hb_down);
+		ret = o2hb_register_callback(NULL, &o2net_hb_down);
 
 	if (ret)
 		o2net_unregister_hb_callbacks();

fs/ocfs2/dir.c

@@ -368,7 +368,7 @@ int ocfs2_do_extend_dir(struct super_block *sb,
 		u32 offset = OCFS2_I(dir)->ip_clusters;
 
 		status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
-						    1, parent_fe_bh, handle,
+						    1, 0, parent_fe_bh, handle,
 						    data_ac, meta_ac, NULL);
 		BUG_ON(status == -EAGAIN);
 		if (status < 0) {

fs/ocfs2/dlm/dlmdomain.c

@@ -1128,8 +1128,8 @@ bail:
 
 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
 {
-	o2hb_unregister_callback(&dlm->dlm_hb_up);
-	o2hb_unregister_callback(&dlm->dlm_hb_down);
+	o2hb_unregister_callback(NULL, &dlm->dlm_hb_up);
+	o2hb_unregister_callback(NULL, &dlm->dlm_hb_down);
 	o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
 }
 
@@ -1141,13 +1141,13 @@ static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
 
 	o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
 			    dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
-	status = o2hb_register_callback(&dlm->dlm_hb_down);
+	status = o2hb_register_callback(NULL, &dlm->dlm_hb_down);
 	if (status)
 		goto bail;
 
 	o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
 			    dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
-	status = o2hb_register_callback(&dlm->dlm_hb_up);
+	status = o2hb_register_callback(NULL, &dlm->dlm_hb_up);
 	if (status)
 		goto bail;
 

fs/ocfs2/dlm/dlmmaster.c

@@ -192,25 +192,20 @@ static void dlm_print_one_mle(struct dlm_master_list_entry *mle)
 static void dlm_dump_mles(struct dlm_ctxt *dlm)
 {
 	struct dlm_master_list_entry *mle;
-	struct list_head *iter;
 	
 	mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name);
 	spin_lock(&dlm->master_lock);
-	list_for_each(iter, &dlm->master_list) {
-		mle = list_entry(iter, struct dlm_master_list_entry, list);
+	list_for_each_entry(mle, &dlm->master_list, list)
 		dlm_print_one_mle(mle);
-	}
 	spin_unlock(&dlm->master_lock);
 }
 
 int dlm_dump_all_mles(const char __user *data, unsigned int len)
 {
-	struct list_head *iter;
 	struct dlm_ctxt *dlm;
 
 	spin_lock(&dlm_domain_lock);
-	list_for_each(iter, &dlm_domains) {
-		dlm = list_entry (iter, struct dlm_ctxt, list);
+	list_for_each_entry(dlm, &dlm_domains, list) {
 		mlog(ML_NOTICE, "found dlm: %p, name=%s\n", dlm, dlm->name);
 		dlm_dump_mles(dlm);
 	}
@@ -454,12 +449,10 @@ static int dlm_find_mle(struct dlm_ctxt *dlm,
 			char *name, unsigned int namelen)
 {
 	struct dlm_master_list_entry *tmpmle;
-	struct list_head *iter;
 
 	assert_spin_locked(&dlm->master_lock);
 
-	list_for_each(iter, &dlm->master_list) {
-		tmpmle = list_entry(iter, struct dlm_master_list_entry, list);
+	list_for_each_entry(tmpmle, &dlm->master_list, list) {
 		if (!dlm_mle_equal(dlm, tmpmle, name, namelen))
 			continue;
 		dlm_get_mle(tmpmle);
@@ -472,13 +465,10 @@ static int dlm_find_mle(struct dlm_ctxt *dlm,
 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up)
 {
 	struct dlm_master_list_entry *mle;
-	struct list_head *iter;
 
 	assert_spin_locked(&dlm->spinlock);
 	
-	list_for_each(iter, &dlm->mle_hb_events) {
-		mle = list_entry(iter, struct dlm_master_list_entry, 
-				 hb_events);
+	list_for_each_entry(mle, &dlm->mle_hb_events, hb_events) {
 		if (node_up)
 			dlm_mle_node_up(dlm, mle, NULL, idx);
 		else
@@ -2434,7 +2424,7 @@ static int dlm_is_lockres_migrateable(struct dlm_ctxt *dlm,
 	int ret;
 	int i;
 	int count = 0;
-	struct list_head *queue, *iter;
+	struct list_head *queue;
 	struct dlm_lock *lock;
 
 	assert_spin_locked(&res->spinlock);
@@ -2453,8 +2443,7 @@ static int dlm_is_lockres_migrateable(struct dlm_ctxt *dlm,
 	ret = 0;
 	queue = &res->granted;
 	for (i = 0; i < 3; i++) {
-		list_for_each(iter, queue) {
-			lock = list_entry(iter, struct dlm_lock, list);
+		list_for_each_entry(lock, queue, list) {
 			++count;
 			if (lock->ml.node == dlm->node_num) {
 				mlog(0, "found a lock owned by this node still "
@@ -2923,18 +2912,16 @@ again:
 static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
 				      struct dlm_lock_resource *res)
 {
-	struct list_head *iter, *iter2;
 	struct list_head *queue = &res->granted;
 	int i, bit;
-	struct dlm_lock *lock;
+	struct dlm_lock *lock, *next;
 
 	assert_spin_locked(&res->spinlock);
 
 	BUG_ON(res->owner == dlm->node_num);
 
 	for (i=0; i<3; i++) {
-		list_for_each_safe(iter, iter2, queue) {
-			lock = list_entry (iter, struct dlm_lock, list);
+		list_for_each_entry_safe(lock, next, queue, list) {
 			if (lock->ml.node != dlm->node_num) {
 				mlog(0, "putting lock for node %u\n",
 				     lock->ml.node);
@@ -2976,7 +2963,6 @@ static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
 {
 	int i;
 	struct list_head *queue = &res->granted;
-	struct list_head *iter;
 	struct dlm_lock *lock;
 	int nodenum;
 
@@ -2984,10 +2970,9 @@ static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
 
 	spin_lock(&res->spinlock);
 	for (i=0; i<3; i++) {
-		list_for_each(iter, queue) {
+		list_for_each_entry(lock, queue, list) {
 			/* up to the caller to make sure this node
 			 * is alive */
-			lock = list_entry (iter, struct dlm_lock, list);
 			if (lock->ml.node != dlm->node_num) {
 				spin_unlock(&res->spinlock);
 				return lock->ml.node;
@@ -3234,8 +3219,7 @@ static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
 
 void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node)
 {
-	struct list_head *iter, *iter2;
-	struct dlm_master_list_entry *mle;
+	struct dlm_master_list_entry *mle, *next;
 	struct dlm_lock_resource *res;
 	unsigned int hash;
 
@@ -3245,9 +3229,7 @@ top:
 
 	/* clean the master list */
 	spin_lock(&dlm->master_lock);
-	list_for_each_safe(iter, iter2, &dlm->master_list) {
-		mle = list_entry(iter, struct dlm_master_list_entry, list);
-
+	list_for_each_entry_safe(mle, next, &dlm->master_list, list) {
 		BUG_ON(mle->type != DLM_MLE_BLOCK &&
 		       mle->type != DLM_MLE_MASTER &&
 		       mle->type != DLM_MLE_MIGRATION);

fs/ocfs2/dlm/dlmrecovery.c

@@ -158,8 +158,7 @@ void dlm_dispatch_work(struct work_struct *work)
 	struct dlm_ctxt *dlm =
 		container_of(work, struct dlm_ctxt, dispatched_work);
 	LIST_HEAD(tmp_list);
-	struct list_head *iter, *iter2;
-	struct dlm_work_item *item;
+	struct dlm_work_item *item, *next;
 	dlm_workfunc_t *workfunc;
 	int tot=0;
 
@@ -167,13 +166,12 @@ void dlm_dispatch_work(struct work_struct *work)
 	list_splice_init(&dlm->work_list, &tmp_list);
 	spin_unlock(&dlm->work_lock);
 
-	list_for_each_safe(iter, iter2, &tmp_list) {
+	list_for_each_entry(item, &tmp_list, list) {
 		tot++;
 	}
 	mlog(0, "%s: work thread has %d work items\n", dlm->name, tot);
 
-	list_for_each_safe(iter, iter2, &tmp_list) {
-		item = list_entry(iter, struct dlm_work_item, list);
+	list_for_each_entry_safe(item, next, &tmp_list, list) {
 		workfunc = item->func;
 		list_del_init(&item->list);
 
@@ -549,7 +547,6 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node)
 {
 	int status = 0;
 	struct dlm_reco_node_data *ndata;
-	struct list_head *iter;
 	int all_nodes_done;
 	int destroy = 0;
 	int pass = 0;
@@ -567,8 +564,7 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node)
 
 	/* safe to access the node data list without a lock, since this
 	 * process is the only one to change the list */
-	list_for_each(iter, &dlm->reco.node_data) {
-		ndata = list_entry (iter, struct dlm_reco_node_data, list);
+	list_for_each_entry(ndata, &dlm->reco.node_data, list) {
 		BUG_ON(ndata->state != DLM_RECO_NODE_DATA_INIT);
 		ndata->state = DLM_RECO_NODE_DATA_REQUESTING;
 
@@ -655,9 +651,7 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node)
 		 * done, or if anyone died */
 		all_nodes_done = 1;
 		spin_lock(&dlm_reco_state_lock);
-		list_for_each(iter, &dlm->reco.node_data) {
-			ndata = list_entry (iter, struct dlm_reco_node_data, list);
-
+		list_for_each_entry(ndata, &dlm->reco.node_data, list) {
 			mlog(0, "checking recovery state of node %u\n",
 			     ndata->node_num);
 			switch (ndata->state) {
@@ -774,16 +768,14 @@ static int dlm_init_recovery_area(struct dlm_ctxt *dlm, u8 dead_node)
 
 static void dlm_destroy_recovery_area(struct dlm_ctxt *dlm, u8 dead_node)
 {
-	struct list_head *iter, *iter2;
-	struct dlm_reco_node_data *ndata;
+	struct dlm_reco_node_data *ndata, *next;
 	LIST_HEAD(tmplist);
 
 	spin_lock(&dlm_reco_state_lock);
 	list_splice_init(&dlm->reco.node_data, &tmplist);
 	spin_unlock(&dlm_reco_state_lock);
 
-	list_for_each_safe(iter, iter2, &tmplist) {
-		ndata = list_entry (iter, struct dlm_reco_node_data, list);
+	list_for_each_entry_safe(ndata, next, &tmplist, list) {
 		list_del_init(&ndata->list);
 		kfree(ndata);
 	}
@@ -876,7 +868,6 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
 	struct dlm_lock_resource *res;
 	struct dlm_ctxt *dlm;
 	LIST_HEAD(resources);
-	struct list_head *iter;
 	int ret;
 	u8 dead_node, reco_master;
 	int skip_all_done = 0;
@@ -920,8 +911,7 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
 
 	/* any errors returned will be due to the new_master dying,
 	 * the dlm_reco_thread should detect this */
-	list_for_each(iter, &resources) {
-		res = list_entry (iter, struct dlm_lock_resource, recovering);
+	list_for_each_entry(res, &resources, recovering) {
 		ret = dlm_send_one_lockres(dlm, res, mres, reco_master,
 				   	DLM_MRES_RECOVERY);
 		if (ret < 0) {
@@ -983,7 +973,6 @@ int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
 {
 	struct dlm_ctxt *dlm = data;
 	struct dlm_reco_data_done *done = (struct dlm_reco_data_done *)msg->buf;
-	struct list_head *iter;
 	struct dlm_reco_node_data *ndata = NULL;
 	int ret = -EINVAL;
 
@@ -1000,8 +989,7 @@ int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
 			dlm->reco.dead_node, done->node_idx, dlm->node_num);
 
 	spin_lock(&dlm_reco_state_lock);
-	list_for_each(iter, &dlm->reco.node_data) {
-		ndata = list_entry (iter, struct dlm_reco_node_data, list);
+	list_for_each_entry(ndata, &dlm->reco.node_data, list) {
 		if (ndata->node_num != done->node_idx)
 			continue;
 
@@ -1049,13 +1037,11 @@ static void dlm_move_reco_locks_to_list(struct dlm_ctxt *dlm,
 					struct list_head *list,
 				       	u8 dead_node)
 {
-	struct dlm_lock_resource *res;
-	struct list_head *iter, *iter2;
+	struct dlm_lock_resource *res, *next;
 	struct dlm_lock *lock;
 
 	spin_lock(&dlm->spinlock);
-	list_for_each_safe(iter, iter2, &dlm->reco.resources) {
-		res = list_entry (iter, struct dlm_lock_resource, recovering);
+	list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
 		/* always prune any $RECOVERY entries for dead nodes,
 		 * otherwise hangs can occur during later recovery */
 		if (dlm_is_recovery_lock(res->lockname.name,
@@ -1169,7 +1155,7 @@ static void dlm_init_migratable_lockres(struct dlm_migratable_lockres *mres,
 					u8 flags, u8 master)
 {
 	/* mres here is one full page */
-	memset(mres, 0, PAGE_SIZE);
+	clear_page(mres);
 	mres->lockname_len = namelen;
 	memcpy(mres->lockname, lockname, namelen);
 	mres->num_locks = 0;
@@ -1252,7 +1238,7 @@ int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
 			 struct dlm_migratable_lockres *mres,
 			 u8 send_to, u8 flags)
 {
-	struct list_head *queue, *iter;
+	struct list_head *queue;
 	int total_locks, i;
 	u64 mig_cookie = 0;
 	struct dlm_lock *lock;
@@ -1278,9 +1264,7 @@ int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
 	total_locks = 0;
 	for (i=DLM_GRANTED_LIST; i<=DLM_BLOCKED_LIST; i++) {
 		queue = dlm_list_idx_to_ptr(res, i);
-		list_for_each(iter, queue) {
-			lock = list_entry (iter, struct dlm_lock, list);
-
+		list_for_each_entry(lock, queue, list) {
 			/* add another lock. */
 			total_locks++;
 			if (!dlm_add_lock_to_array(lock, mres, i))
@@ -1717,7 +1701,6 @@ static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
 	struct dlm_lockstatus *lksb = NULL;
 	int ret = 0;
 	int i, j, bad;
-	struct list_head *iter;
 	struct dlm_lock *lock = NULL;
 	u8 from = O2NM_MAX_NODES;
 	unsigned int added = 0;
@@ -1755,8 +1738,7 @@ static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
 			spin_lock(&res->spinlock);
 			for (j = DLM_GRANTED_LIST; j <= DLM_BLOCKED_LIST; j++) {
 				tmpq = dlm_list_idx_to_ptr(res, j);
-				list_for_each(iter, tmpq) {
-					lock = list_entry (iter, struct dlm_lock, list);
+				list_for_each_entry(lock, tmpq, list) {
 					if (lock->ml.cookie != ml->cookie)
 						lock = NULL;
 					else
@@ -1930,8 +1912,8 @@ void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
 				       struct dlm_lock_resource *res)
 {
 	int i;
-	struct list_head *queue, *iter, *iter2;
-	struct dlm_lock *lock;
+	struct list_head *queue;
+	struct dlm_lock *lock, *next;
 
 	res->state |= DLM_LOCK_RES_RECOVERING;
 	if (!list_empty(&res->recovering)) {
@@ -1947,8 +1929,7 @@ void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
 	/* find any pending locks and put them back on proper list */
 	for (i=DLM_BLOCKED_LIST; i>=DLM_GRANTED_LIST; i--) {
 		queue = dlm_list_idx_to_ptr(res, i);
-		list_for_each_safe(iter, iter2, queue) {
-			lock = list_entry (iter, struct dlm_lock, list);
+		list_for_each_entry_safe(lock, next, queue, list) {
 			dlm_lock_get(lock);
 			if (lock->convert_pending) {
 				/* move converting lock back to granted */
@@ -2013,18 +1994,15 @@ static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm,
 					      u8 dead_node, u8 new_master)
 {
 	int i;
-	struct list_head *iter, *iter2;
 	struct hlist_node *hash_iter;
 	struct hlist_head *bucket;
-
-	struct dlm_lock_resource *res;
+	struct dlm_lock_resource *res, *next;
 
 	mlog_entry_void();
 
 	assert_spin_locked(&dlm->spinlock);
 
-	list_for_each_safe(iter, iter2, &dlm->reco.resources) {
-		res = list_entry (iter, struct dlm_lock_resource, recovering);
+	list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
 		if (res->owner == dead_node) {
 			list_del_init(&res->recovering);
 			spin_lock(&res->spinlock);
@@ -2099,7 +2077,7 @@ static inline int dlm_lvb_needs_invalidation(struct dlm_lock *lock, int local)
 static void dlm_revalidate_lvb(struct dlm_ctxt *dlm,
 			       struct dlm_lock_resource *res, u8 dead_node)
 {
-	struct list_head *iter, *queue;
+	struct list_head *queue;
 	struct dlm_lock *lock;
 	int blank_lvb = 0, local = 0;
 	int i;
@@ -2121,8 +2099,7 @@ static void dlm_revalidate_lvb(struct dlm_ctxt *dlm,
 
 	for (i=DLM_GRANTED_LIST; i<=DLM_CONVERTING_LIST; i++) {
 		queue = dlm_list_idx_to_ptr(res, i);
-		list_for_each(iter, queue) {
-			lock = list_entry (iter, struct dlm_lock, list);
+		list_for_each_entry(lock, queue, list) {
 			if (lock->ml.node == search_node) {
 				if (dlm_lvb_needs_invalidation(lock, local)) {
 					/* zero the lksb lvb and lockres lvb */
@@ -2143,8 +2120,7 @@ static void dlm_revalidate_lvb(struct dlm_ctxt *dlm,
 static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
 				struct dlm_lock_resource *res, u8 dead_node)
 {
-	struct list_head *iter, *tmpiter;
-	struct dlm_lock *lock;
+	struct dlm_lock *lock, *next;
 	unsigned int freed = 0;
 
 	/* this node is the lockres master:
@@ -2155,24 +2131,21 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
 	assert_spin_locked(&res->spinlock);
 
 	/* TODO: check pending_asts, pending_basts here */
-	list_for_each_safe(iter, tmpiter, &res->granted) {
-		lock = list_entry (iter, struct dlm_lock, list);
+	list_for_each_entry_safe(lock, next, &res->granted, list) {
 		if (lock->ml.node == dead_node) {
 			list_del_init(&lock->list);
 			dlm_lock_put(lock);
 			freed++;
 		}
 	}
-	list_for_each_safe(iter, tmpiter, &res->converting) {
-		lock = list_entry (iter, struct dlm_lock, list);
+	list_for_each_entry_safe(lock, next, &res->converting, list) {
 		if (lock->ml.node == dead_node) {
 			list_del_init(&lock->list);
 			dlm_lock_put(lock);
 			freed++;
 		}
 	}
-	list_for_each_safe(iter, tmpiter, &res->blocked) {
-		lock = list_entry (iter, struct dlm_lock, list);
+	list_for_each_entry_safe(lock, next, &res->blocked, list) {
 		if (lock->ml.node == dead_node) {
 			list_del_init(&lock->list);
 			dlm_lock_put(lock);

fs/ocfs2/dlmglue.c

@@ -600,15 +600,13 @@ static inline int ocfs2_highest_compat_lock_level(int level)
 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
 			      unsigned long newflags)
 {
-	struct list_head *pos, *tmp;
-	struct ocfs2_mask_waiter *mw;
+	struct ocfs2_mask_waiter *mw, *tmp;
 
  	assert_spin_locked(&lockres->l_lock);
 
 	lockres->l_flags = newflags;
 
-	list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) {
-		mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item);
+	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
 			continue;
 

fs/ocfs2/endian.h

@@ -32,6 +32,11 @@ static inline void le32_add_cpu(__le32 *var, u32 val)
 	*var = cpu_to_le32(le32_to_cpu(*var) + val);
 }
 
+static inline void le64_add_cpu(__le64 *var, u64 val)
+{
+	*var = cpu_to_le64(le64_to_cpu(*var) + val);
+}
+
 static inline void le32_and_cpu(__le32 *var, u32 val)
 {
 	*var = cpu_to_le32(le32_to_cpu(*var) & val);

fs/ocfs2/extent_map.c

@@ -109,17 +109,14 @@ static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
  */
 void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
 {
-	struct list_head *p, *n;
-	struct ocfs2_extent_map_item *emi;
+	struct ocfs2_extent_map_item *emi, *n;
 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 	struct ocfs2_extent_map *em = &oi->ip_extent_map;
 	LIST_HEAD(tmp_list);
 	unsigned int range;
 
 	spin_lock(&oi->ip_lock);
-	list_for_each_safe(p, n, &em->em_list) {
-		emi = list_entry(p, struct ocfs2_extent_map_item, ei_list);
-
+	list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
 		if (emi->ei_cpos >= cpos) {
 			/* Full truncate of this record. */
 			list_move(&emi->ei_list, &tmp_list);
@@ -136,8 +133,7 @@ void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
 	}
 	spin_unlock(&oi->ip_lock);
 
-	list_for_each_safe(p, n, &tmp_list) {
-		emi = list_entry(p, struct ocfs2_extent_map_item, ei_list);
+	list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
 		list_del(&emi->ei_list);
 		kfree(emi);
 	}
@@ -377,37 +373,6 @@ out:
 	return ret;
 }
 
-/*
- * Return the index of the extent record which contains cluster #v_cluster.
- * -1 is returned if it was not found.
- *
- * Should work fine on interior and exterior nodes.
- */
-static int ocfs2_search_extent_list(struct ocfs2_extent_list *el,
-				    u32 v_cluster)
-{
-	int ret = -1;
-	int i;
-	struct ocfs2_extent_rec *rec;
-	u32 rec_end, rec_start, clusters;
-
-	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
-		rec = &el->l_recs[i];
-
-		rec_start = le32_to_cpu(rec->e_cpos);
-		clusters = ocfs2_rec_clusters(el, rec);
-
-		rec_end = rec_start + clusters;
-
-		if (v_cluster >= rec_start && v_cluster < rec_end) {
-			ret = i;
-			break;
-		}
-	}
-
-	return ret;
-}
-
 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
 		       u32 *p_cluster, u32 *num_clusters,
 		       unsigned int *extent_flags)

fs/ocfs2/file.c

@@ -263,6 +263,7 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 	int status;
 	handle_t *handle;
 	struct ocfs2_dinode *di;
+	u64 cluster_bytes;
 
 	mlog_entry_void();
 
@@ -286,7 +287,9 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 	/*
 	 * Do this before setting i_size.
 	 */
-	status = ocfs2_zero_tail_for_truncate(inode, handle, new_i_size);
+	cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
+	status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
+					       cluster_bytes);
 	if (status) {
 		mlog_errno(status);
 		goto out_commit;
@@ -326,9 +329,6 @@ static int ocfs2_truncate_file(struct inode *inode,
 		   (unsigned long long)OCFS2_I(inode)->ip_blkno,
 		   (unsigned long long)new_i_size);
 
-	unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
-	truncate_inode_pages(inode->i_mapping, new_i_size);
-
 	fe = (struct ocfs2_dinode *) di_bh->b_data;
 	if (!OCFS2_IS_VALID_DINODE(fe)) {
 		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
@@ -363,16 +363,23 @@ static int ocfs2_truncate_file(struct inode *inode,
 	if (new_i_size == le64_to_cpu(fe->i_size))
 		goto bail;
 
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
 	/* This forces other nodes to sync and drop their pages. Do
 	 * this even if we have a truncate without allocation change -
 	 * ocfs2 cluster sizes can be much greater than page size, so
 	 * we have to truncate them anyway.  */
 	status = ocfs2_data_lock(inode, 1);
 	if (status < 0) {
+		up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
 		mlog_errno(status);
 		goto bail;
 	}
 
+	unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
+	truncate_inode_pages(inode->i_mapping, new_i_size);
+
 	/* alright, we're going to need to do a full blown alloc size
 	 * change. Orphan the inode so that recovery can complete the
 	 * truncate if necessary. This does the task of marking
@@ -399,6 +406,8 @@ static int ocfs2_truncate_file(struct inode *inode,
 bail_unlock_data:
 	ocfs2_data_unlock(inode, 1);
 
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
 bail:
 
 	mlog_exit(status);
@@ -419,6 +428,7 @@ 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,
@@ -431,9 +441,13 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 	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, fe);
 	if (free_extents < 0) {
 		status = free_extents;
@@ -483,7 +497,7 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 	     num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
 	status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
 				     *logical_offset, block, num_bits,
-				     meta_ac);
+				     flags, meta_ac);
 	if (status < 0) {
 		mlog_errno(status);
 		goto leave;
@@ -516,25 +530,31 @@ leave:
  * For a given allocation, determine which allocators will need to be
  * accessed, and lock them, reserving the appropriate number of bits.
  *
- * Called from ocfs2_extend_allocation() for file systems which don't
- * support holes, and from ocfs2_write() for file systems which
- * understand sparse inodes.
+ * Sparse file systems call this from ocfs2_write_begin_nolock()
+ * and ocfs2_allocate_unwritten_extents().
+ *
+ * File systems which don't support holes call this from
+ * ocfs2_extend_allocation().
  */
 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
-			  u32 clusters_to_add,
+			  u32 clusters_to_add, u32 extents_to_split,
 			  struct ocfs2_alloc_context **data_ac,
 			  struct ocfs2_alloc_context **meta_ac)
 {
-	int ret, num_free_extents;
+	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);
 
 	*meta_ac = NULL;
-	*data_ac = NULL;
+	if (data_ac)
+		*data_ac = NULL;
+
+	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
 
 	mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
-	     "clusters_to_add = %u\n",
+	     "clusters_to_add = %u, extents_to_split = %u\n",
 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
-	     le32_to_cpu(di->i_clusters), clusters_to_add);
+	     le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
 
 	num_free_extents = ocfs2_num_free_extents(osb, inode, di);
 	if (num_free_extents < 0) {
@@ -552,9 +572,12 @@ int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
 	 *
 	 * Most of the time we'll only be seeing this 1 cluster at a time
 	 * anyway.
+	 *
+	 * Always lock for any unwritten extents - we might want to
+	 * add blocks during a split.
 	 */
 	if (!num_free_extents ||
-	    (ocfs2_sparse_alloc(osb) && num_free_extents < clusters_to_add)) {
+	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
 		ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
 		if (ret < 0) {
 			if (ret != -ENOSPC)
@@ -563,6 +586,9 @@ int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
 		}
 	}
 
+	if (clusters_to_add == 0)
+		goto out;
+
 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
 	if (ret < 0) {
 		if (ret != -ENOSPC)
@@ -585,14 +611,13 @@ out:
 	return ret;
 }
 
-static int ocfs2_extend_allocation(struct inode *inode,
-				   u32 clusters_to_add)
+static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+				     u32 clusters_to_add, int mark_unwritten)
 {
 	int status = 0;
 	int restart_func = 0;
-	int drop_alloc_sem = 0;
 	int credits;
-	u32 prev_clusters, logical_start;
+	u32 prev_clusters;
 	struct buffer_head *bh = NULL;
 	struct ocfs2_dinode *fe = NULL;
 	handle_t *handle = NULL;
@@ -607,7 +632,7 @@ static int ocfs2_extend_allocation(struct inode *inode,
 	 * This function only exists for file systems which don't
 	 * support holes.
 	 */
-	BUG_ON(ocfs2_sparse_alloc(osb));
+	BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 
 	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
 				  OCFS2_BH_CACHED, inode);
@@ -623,19 +648,10 @@ static int ocfs2_extend_allocation(struct inode *inode,
 		goto leave;
 	}
 
-	logical_start = OCFS2_I(inode)->ip_clusters;
-
 restart_all:
 	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 
-	/* blocks peope in read/write from reading our allocation
-	 * until we're done changing it. We depend on i_mutex to block
-	 * other extend/truncate calls while we're here. Ordering wrt
-	 * start_trans is important here -- always do it before! */
-	down_write(&OCFS2_I(inode)->ip_alloc_sem);
-	drop_alloc_sem = 1;
-
-	status = ocfs2_lock_allocators(inode, fe, clusters_to_add, &data_ac,
+	status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
 				       &meta_ac);
 	if (status) {
 		mlog_errno(status);
@@ -668,6 +684,7 @@ restarted_transaction:
 					    inode,
 					    &logical_start,
 					    clusters_to_add,
+					    mark_unwritten,
 					    bh,
 					    handle,
 					    data_ac,
@@ -720,10 +737,6 @@ restarted_transaction:
 	     OCFS2_I(inode)->ip_clusters, i_size_read(inode));
 
 leave:
-	if (drop_alloc_sem) {
-		up_write(&OCFS2_I(inode)->ip_alloc_sem);
-		drop_alloc_sem = 0;
-	}
 	if (handle) {
 		ocfs2_commit_trans(osb, handle);
 		handle = NULL;
@@ -749,6 +762,25 @@ leave:
 	return status;
 }
 
+static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+				   u32 clusters_to_add, int mark_unwritten)
+{
+	int ret;
+
+	/*
+	 * The alloc sem blocks peope in read/write from reading our
+	 * allocation until we're done changing it. We depend on
+	 * i_mutex to block other extend/truncate calls while we're
+	 * here.
+	 */
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+	ret = __ocfs2_extend_allocation(inode, logical_start, clusters_to_add,
+					mark_unwritten);
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	return ret;
+}
+
 /* Some parts of this taken from generic_cont_expand, which turned out
  * to be too fragile to do exactly what we need without us having to
  * worry about recursive locking in ->prepare_write() and
@@ -890,7 +922,9 @@ static int ocfs2_extend_file(struct inode *inode,
 	}
 
 	if (clusters_to_add) {
-		ret = ocfs2_extend_allocation(inode, clusters_to_add);
+		ret = ocfs2_extend_allocation(inode,
+					      OCFS2_I(inode)->ip_clusters,
+					      clusters_to_add, 0);
 		if (ret < 0) {
 			mlog_errno(ret);
 			goto out_unlock;
@@ -995,6 +1029,13 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
 		goto bail_unlock;
 	}
 
+	/*
+	 * This will intentionally not wind up calling vmtruncate(),
+	 * since all the work for a size change has been done above.
+	 * Otherwise, we could get into problems with truncate as
+	 * ip_alloc_sem is used there to protect against i_size
+	 * changes.
+	 */
 	status = inode_setattr(inode, attr);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1070,17 +1111,16 @@ out:
 	return ret;
 }
 
-static int ocfs2_write_remove_suid(struct inode *inode)
+static int __ocfs2_write_remove_suid(struct inode *inode,
+				     struct buffer_head *bh)
 {
 	int ret;
-	struct buffer_head *bh = NULL;
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 	handle_t *handle;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 	struct ocfs2_dinode *di;
 
 	mlog_entry("(Inode %llu, mode 0%o)\n",
-		   (unsigned long long)oi->ip_blkno, inode->i_mode);
+		   (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
 
 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 	if (handle == NULL) {
@@ -1089,17 +1129,11 @@ static int ocfs2_write_remove_suid(struct inode *inode)
 		goto out;
 	}
 
-	ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out_trans;
-	}
-
 	ret = ocfs2_journal_access(handle, inode, bh,
 				   OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret < 0) {
 		mlog_errno(ret);
-		goto out_bh;
+		goto out_trans;
 	}
 
 	inode->i_mode &= ~S_ISUID;
@@ -1112,8 +1146,7 @@ static int ocfs2_write_remove_suid(struct inode *inode)
 	ret = ocfs2_journal_dirty(handle, bh);
 	if (ret < 0)
 		mlog_errno(ret);
-out_bh:
-	brelse(bh);
+
 out_trans:
 	ocfs2_commit_trans(osb, handle);
 out:
@@ -1159,6 +1192,460 @@ out:
 	return ret;
 }
 
+static int ocfs2_write_remove_suid(struct inode *inode)
+{
+	int ret;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
+			       oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret =  __ocfs2_write_remove_suid(inode, bh);
+out:
+	brelse(bh);
+	return ret;
+}
+
+/*
+ * Allocate enough extents to cover the region starting at byte offset
+ * start for len bytes. Existing extents are skipped, any extents
+ * added are marked as "unwritten".
+ */
+static int ocfs2_allocate_unwritten_extents(struct inode *inode,
+					    u64 start, u64 len)
+{
+	int ret;
+	u32 cpos, phys_cpos, clusters, alloc_size;
+
+	/*
+	 * We consider both start and len to be inclusive.
+	 */
+	cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
+	clusters -= cpos;
+
+	while (clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
+					 &alloc_size, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * Hole or existing extent len can be arbitrary, so
+		 * cap it to our own allocation request.
+		 */
+		if (alloc_size > clusters)
+			alloc_size = clusters;
+
+		if (phys_cpos) {
+			/*
+			 * We already have an allocation at this
+			 * region so we can safely skip it.
+			 */
+			goto next;
+		}
+
+		ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
+		if (ret) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+
+next:
+		cpos += alloc_size;
+		clusters -= alloc_size;
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int __ocfs2_remove_inode_range(struct inode *inode,
+				      struct buffer_head *di_bh,
+				      u32 cpos, u32 phys_cpos, u32 len,
+				      struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *tl_inode = osb->osb_tl_inode;
+	handle_t *handle;
+	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);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	mutex_lock(&tl_inode->i_mutex);
+
+	if (ocfs2_truncate_log_needs_flush(osb)) {
+		ret = __ocfs2_flush_truncate_log(osb);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
+	if (handle == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access(handle, inode, di_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
+				  dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	OCFS2_I(inode)->ip_clusters -= len;
+	di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
+
+	ret = ocfs2_journal_dirty(handle, di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
+	if (ret)
+		mlog_errno(ret);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	mutex_unlock(&tl_inode->i_mutex);
+
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	return ret;
+}
+
+/*
+ * Truncate a byte range, avoiding pages within partial clusters. This
+ * preserves those pages for the zeroing code to write to.
+ */
+static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
+					 u64 byte_len)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	loff_t start, end;
+	struct address_space *mapping = inode->i_mapping;
+
+	start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
+	end = byte_start + byte_len;
+	end = end & ~(osb->s_clustersize - 1);
+
+	if (start < end) {
+		unmap_mapping_range(mapping, start, end - start, 0);
+		truncate_inode_pages_range(mapping, start, end - 1);
+	}
+}
+
+static int ocfs2_zero_partial_clusters(struct inode *inode,
+				       u64 start, u64 len)
+{
+	int ret = 0;
+	u64 tmpend, end = start + len;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	unsigned int csize = osb->s_clustersize;
+	handle_t *handle;
+
+	/*
+	 * The "start" and "end" values are NOT necessarily part of
+	 * the range whose allocation is being deleted. Rather, this
+	 * is what the user passed in with the request. We must zero
+	 * partial clusters here. There's no need to worry about
+	 * physical allocation - the zeroing code knows to skip holes.
+	 */
+	mlog(0, "byte start: %llu, end: %llu\n",
+	     (unsigned long long)start, (unsigned long long)end);
+
+	/*
+	 * If both edges are on a cluster boundary then there's no
+	 * zeroing required as the region is part of the allocation to
+	 * be truncated.
+	 */
+	if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
+		goto out;
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (handle == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * We want to get the byte offset of the end of the 1st cluster.
+	 */
+	tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
+	if (tmpend > end)
+		tmpend = end;
+
+	mlog(0, "1st range: start: %llu, tmpend: %llu\n",
+	     (unsigned long long)start, (unsigned long long)tmpend);
+
+	ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
+	if (ret)
+		mlog_errno(ret);
+
+	if (tmpend < end) {
+		/*
+		 * This may make start and end equal, but the zeroing
+		 * code will skip any work in that case so there's no
+		 * need to catch it up here.
+		 */
+		start = end & ~(osb->s_clustersize - 1);
+
+		mlog(0, "2nd range: start: %llu, end: %llu\n",
+		     (unsigned long long)start, (unsigned long long)end);
+
+		ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+	ocfs2_commit_trans(osb, handle);
+out:
+	return ret;
+}
+
+static int ocfs2_remove_inode_range(struct inode *inode,
+				    struct buffer_head *di_bh, u64 byte_start,
+				    u64 byte_len)
+{
+	int ret = 0;
+	u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	if (byte_len == 0)
+		return 0;
+
+	trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
+	trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
+	if (trunc_len >= trunc_start)
+		trunc_len -= trunc_start;
+	else
+		trunc_len = 0;
+
+	mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     (unsigned long long)byte_start,
+	     (unsigned long long)byte_len, trunc_start, trunc_len);
+
+	ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	cpos = trunc_start;
+	while (trunc_len) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
+					 &alloc_size, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (alloc_size > trunc_len)
+			alloc_size = trunc_len;
+
+		/* Only do work for non-holes */
+		if (phys_cpos != 0) {
+			ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
+							 phys_cpos, alloc_size,
+							 &dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		cpos += alloc_size;
+		trunc_len -= alloc_size;
+	}
+
+	ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
+
+out:
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+	ocfs2_run_deallocs(osb, &dealloc);
+
+	return ret;
+}
+
+/*
+ * Parts of this function taken from xfs_change_file_space()
+ */
+int ocfs2_change_file_space(struct file *file, unsigned int cmd,
+			    struct ocfs2_space_resv *sr)
+{
+	int ret;
+	s64 llen;
+	struct inode *inode = file->f_path.dentry->d_inode;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *di_bh = NULL;
+	handle_t *handle;
+	unsigned long long max_off = ocfs2_max_file_offset(inode->i_sb->s_blocksize_bits);
+
+	if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
+	    !ocfs2_writes_unwritten_extents(osb))
+		return -ENOTTY;
+	else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
+		 !ocfs2_sparse_alloc(osb))
+		return -ENOTTY;
+
+	if (!S_ISREG(inode->i_mode))
+		return -EINVAL;
+
+	if (!(file->f_mode & FMODE_WRITE))
+		return -EBADF;
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+		return -EROFS;
+
+	mutex_lock(&inode->i_mutex);
+
+	/*
+	 * This prevents concurrent writes on other nodes
+	 */
+	ret = ocfs2_rw_lock(inode, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_meta_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_rw_unlock;
+	}
+
+	if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
+		ret = -EPERM;
+		goto out_meta_unlock;
+	}
+
+	switch (sr->l_whence) {
+	case 0: /*SEEK_SET*/
+		break;
+	case 1: /*SEEK_CUR*/
+		sr->l_start += file->f_pos;
+		break;
+	case 2: /*SEEK_END*/
+		sr->l_start += i_size_read(inode);
+		break;
+	default:
+		ret = -EINVAL;
+		goto out_meta_unlock;
+	}
+	sr->l_whence = 0;
+
+	llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
+
+	if (sr->l_start < 0
+	    || sr->l_start > max_off
+	    || (sr->l_start + llen) < 0
+	    || (sr->l_start + llen) > max_off) {
+		ret = -EINVAL;
+		goto out_meta_unlock;
+	}
+
+	if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
+		if (sr->l_len <= 0) {
+			ret = -EINVAL;
+			goto out_meta_unlock;
+		}
+	}
+
+	if (should_remove_suid(file->f_path.dentry)) {
+		ret = __ocfs2_write_remove_suid(inode, di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_meta_unlock;
+		}
+	}
+
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+	switch (cmd) {
+	case OCFS2_IOC_RESVSP:
+	case OCFS2_IOC_RESVSP64:
+		/*
+		 * This takes unsigned offsets, but the signed ones we
+		 * pass have been checked against overflow above.
+		 */
+		ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
+						       sr->l_len);
+		break;
+	case OCFS2_IOC_UNRESVSP:
+	case OCFS2_IOC_UNRESVSP64:
+		ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
+					       sr->l_len);
+		break;
+	default:
+		ret = -EINVAL;
+	}
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_meta_unlock;
+	}
+
+	/*
+	 * We update c/mtime for these changes
+	 */
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_meta_unlock;
+	}
+
+	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+	ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
+	if (ret < 0)
+		mlog_errno(ret);
+
+	ocfs2_commit_trans(osb, handle);
+
+out_meta_unlock:
+	brelse(di_bh);
+	ocfs2_meta_unlock(inode, 1);
+out_rw_unlock:
+	ocfs2_rw_unlock(inode, 1);
+
+	mutex_unlock(&inode->i_mutex);
+out:
+	return ret;
+}
+
 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 					 loff_t *ppos,
 					 size_t count,
@@ -1329,15 +1816,16 @@ ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
 	*basep = base;
 }
 
-static struct page * ocfs2_get_write_source(struct ocfs2_buffered_write_priv *bp,
+static struct page * ocfs2_get_write_source(char **ret_src_buf,
 					    const struct iovec *cur_iov,
 					    size_t iov_offset)
 {
 	int ret;
-	char *buf;
+	char *buf = cur_iov->iov_base + iov_offset;
 	struct page *src_page = NULL;
+	unsigned long off;
 
-	buf = cur_iov->iov_base + iov_offset;
+	off = (unsigned long)(buf) & ~PAGE_CACHE_MASK;
 
 	if (!segment_eq(get_fs(), KERNEL_DS)) {
 		/*
@@ -1349,18 +1837,17 @@ static struct page * ocfs2_get_write_source(struct ocfs2_buffered_write_priv *bp
 				     (unsigned long)buf & PAGE_CACHE_MASK, 1,
 				     0, 0, &src_page, NULL);
 		if (ret == 1)
-			bp->b_src_buf = kmap(src_page);
+			*ret_src_buf = kmap(src_page) + off;
 		else
 			src_page = ERR_PTR(-EFAULT);
 	} else {
-		bp->b_src_buf = buf;
+		*ret_src_buf = buf;
 	}
 
 	return src_page;
 }
 
-static void ocfs2_put_write_source(struct ocfs2_buffered_write_priv *bp,
-				   struct page *page)
+static void ocfs2_put_write_source(struct page *page)
 {
 	if (page) {
 		kunmap(page);
@@ -1376,10 +1863,12 @@ static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
 {
 	int ret = 0;
 	ssize_t copied, total = 0;
-	size_t iov_offset = 0;
+	size_t iov_offset = 0, bytes;
+	loff_t pos;
 	const struct iovec *cur_iov = iov;
-	struct ocfs2_buffered_write_priv bp;
-	struct page *page;
+	struct page *user_page, *page;
+	char *buf, *dst;
+	void *fsdata;
 
 	/*
 	 * handle partial DIO write.  Adjust cur_iov if needed.
@@ -1387,21 +1876,38 @@ static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
 	ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
 
 	do {
-		bp.b_cur_off = iov_offset;
-		bp.b_cur_iov = cur_iov;
+		pos = *ppos;
 
-		page = ocfs2_get_write_source(&bp, cur_iov, iov_offset);
-		if (IS_ERR(page)) {
-			ret = PTR_ERR(page);
+		user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset);
+		if (IS_ERR(user_page)) {
+			ret = PTR_ERR(user_page);
 			goto out;
 		}
 
-		copied = ocfs2_buffered_write_cluster(file, *ppos, count,
-						      ocfs2_map_and_write_user_data,
-						      &bp);
+		/* Stay within our page boundaries */
+		bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)),
+			    (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK)));
+		/* Stay within the vector boundary */
+		bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset);
+		/* Stay within count */
+		bytes = min(bytes, count);
+
+		page = NULL;
+		ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0,
+					&page, &fsdata);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
 
-		ocfs2_put_write_source(&bp, page);
+		dst = kmap_atomic(page, KM_USER0);
+		memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes);
+		kunmap_atomic(dst, KM_USER0);
+		flush_dcache_page(page);
+		ocfs2_put_write_source(user_page);
 
+		copied = ocfs2_write_end(file, file->f_mapping, pos, bytes,
+					 bytes, page, fsdata);
 		if (copied < 0) {
 			mlog_errno(copied);
 			ret = copied;
@@ -1409,7 +1915,7 @@ static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
 		}
 
 		total += copied;
-		*ppos = *ppos + copied;
+		*ppos = pos + copied;
 		count -= copied;
 
 		ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
@@ -1579,52 +2085,46 @@ static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
 				    struct pipe_buffer *buf,
 				    struct splice_desc *sd)
 {
-	int ret, count, total = 0;
+	int ret, count;
 	ssize_t copied = 0;
-	struct ocfs2_splice_write_priv sp;
+	struct file *file = sd->u.file;
+	unsigned int offset;
+	struct page *page = NULL;
+	void *fsdata;
+	char *src, *dst;
 
 	ret = buf->ops->confirm(pipe, buf);
 	if (ret)
 		goto out;
 
-	sp.s_sd = sd;
-	sp.s_buf = buf;
-	sp.s_pipe = pipe;
-	sp.s_offset = sd->pos & ~PAGE_CACHE_MASK;
-	sp.s_buf_offset = buf->offset;
-
+	offset = sd->pos & ~PAGE_CACHE_MASK;
 	count = sd->len;
-	if (count + sp.s_offset > PAGE_CACHE_SIZE)
-		count = PAGE_CACHE_SIZE - sp.s_offset;
+	if (count + offset > PAGE_CACHE_SIZE)
+		count = PAGE_CACHE_SIZE - offset;
 
-	do {
-		/*
-		 * splice wants us to copy up to one page at a
-		 * time. For pagesize > cluster size, this means we
-		 * might enter ocfs2_buffered_write_cluster() more
-		 * than once, so keep track of our progress here.
-		 */
-		copied = ocfs2_buffered_write_cluster(sd->u.file,
-						      (loff_t)sd->pos + total,
-						      count,
-						      ocfs2_map_and_write_splice_data,
-						      &sp);
-		if (copied < 0) {
-			mlog_errno(copied);
-			ret = copied;
-			goto out;
-		}
+	ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0,
+				&page, &fsdata);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
 
-		count -= copied;
-		sp.s_offset += copied;
-		sp.s_buf_offset += copied;
-		total += copied;
-	} while (count);
+	src = buf->ops->map(pipe, buf, 1);
+	dst = kmap_atomic(page, KM_USER1);
+	memcpy(dst + offset, src + buf->offset, count);
+	kunmap_atomic(page, KM_USER1);
+	buf->ops->unmap(pipe, buf, src);
 
-	ret = 0;
+	copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count,
+				 page, fsdata);
+	if (copied < 0) {
+		mlog_errno(copied);
+		ret = copied;
+		goto out;
+	}
 out:
 
-	return total ? total : ret;
+	return copied ? copied : ret;
 }
 
 static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,

fs/ocfs2/file.h

@@ -39,15 +39,16 @@ enum ocfs2_alloc_restarted {
 };
 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 			       struct inode *inode,
-			       u32 *cluster_start,
+			       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);
+			       enum ocfs2_alloc_restarted *reason_ret);
 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
-			  u32 clusters_to_add,
+			  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);
@@ -61,4 +62,7 @@ int ocfs2_should_update_atime(struct inode *inode,
 int ocfs2_update_inode_atime(struct inode *inode,
 			     struct buffer_head *bh);
 
+int ocfs2_change_file_space(struct file *file, unsigned int cmd,
+			    struct ocfs2_space_resv *sr);
+
 #endif /* OCFS2_FILE_H */

fs/ocfs2/heartbeat.c

@@ -157,16 +157,16 @@ int ocfs2_register_hb_callbacks(struct ocfs2_super *osb)
 	if (ocfs2_mount_local(osb))
 		return 0;
 
-	status = o2hb_register_callback(&osb->osb_hb_down);
+	status = o2hb_register_callback(osb->uuid_str, &osb->osb_hb_down);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
 	}
 
-	status = o2hb_register_callback(&osb->osb_hb_up);
+	status = o2hb_register_callback(osb->uuid_str, &osb->osb_hb_up);
 	if (status < 0) {
 		mlog_errno(status);
-		o2hb_unregister_callback(&osb->osb_hb_down);
+		o2hb_unregister_callback(osb->uuid_str, &osb->osb_hb_down);
 	}
 
 bail:
@@ -178,8 +178,8 @@ void ocfs2_clear_hb_callbacks(struct ocfs2_super *osb)
 	if (ocfs2_mount_local(osb))
 		return;
 
-	o2hb_unregister_callback(&osb->osb_hb_down);
-	o2hb_unregister_callback(&osb->osb_hb_up);
+	o2hb_unregister_callback(osb->uuid_str, &osb->osb_hb_down);
+	o2hb_unregister_callback(osb->uuid_str, &osb->osb_hb_up);
 }
 
 void ocfs2_stop_heartbeat(struct ocfs2_super *osb)

fs/ocfs2/ioctl.c

@@ -14,6 +14,7 @@
 #include "ocfs2.h"
 #include "alloc.h"
 #include "dlmglue.h"
+#include "file.h"
 #include "inode.h"
 #include "journal.h"
 
@@ -115,6 +116,7 @@ int ocfs2_ioctl(struct inode * inode, struct file * filp,
 {
 	unsigned int flags;
 	int status;
+	struct ocfs2_space_resv sr;
 
 	switch (cmd) {
 	case OCFS2_IOC_GETFLAGS:
@@ -130,6 +132,14 @@ int ocfs2_ioctl(struct inode * inode, struct file * filp,
 
 		return ocfs2_set_inode_attr(inode, flags,
 			OCFS2_FL_MODIFIABLE);
+	case OCFS2_IOC_RESVSP:
+	case OCFS2_IOC_RESVSP64:
+	case OCFS2_IOC_UNRESVSP:
+	case OCFS2_IOC_UNRESVSP64:
+		if (copy_from_user(&sr, (int __user *) arg, sizeof(sr)))
+			return -EFAULT;
+
+		return ocfs2_change_file_space(filp, cmd, &sr);
 	default:
 		return -ENOTTY;
 	}
@@ -148,6 +158,11 @@ long ocfs2_compat_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 	case OCFS2_IOC32_SETFLAGS:
 		cmd = OCFS2_IOC_SETFLAGS;
 		break;
+	case OCFS2_IOC_RESVSP:
+	case OCFS2_IOC_RESVSP64:
+	case OCFS2_IOC_UNRESVSP:
+	case OCFS2_IOC_UNRESVSP64:
+		break;
 	default:
 		return -ENOIOCTLCMD;
 	}

fs/ocfs2/journal.c

@@ -722,8 +722,7 @@ void ocfs2_complete_recovery(struct work_struct *work)
 		container_of(work, struct ocfs2_journal, j_recovery_work);
 	struct ocfs2_super *osb = journal->j_osb;
 	struct ocfs2_dinode *la_dinode, *tl_dinode;
-	struct ocfs2_la_recovery_item *item;
-	struct list_head *p, *n;
+	struct ocfs2_la_recovery_item *item, *n;
 	LIST_HEAD(tmp_la_list);
 
 	mlog_entry_void();
@@ -734,8 +733,7 @@ void ocfs2_complete_recovery(struct work_struct *work)
 	list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
 	spin_unlock(&journal->j_lock);
 
-	list_for_each_safe(p, n, &tmp_la_list) {
-		item = list_entry(p, struct ocfs2_la_recovery_item, lri_list);
+	list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
 		list_del_init(&item->lri_list);
 
 		mlog(0, "Complete recovery for slot %d\n", item->lri_slot);

fs/ocfs2/journal.h

@@ -289,6 +289,8 @@ int                  ocfs2_journal_dirty_data(handle_t *handle,
 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE 		      \
 					 + OCFS2_TRUNCATE_LOG_UPDATE)
 
+#define OCFS2_REMOVE_EXTENT_CREDITS (OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS)
+
 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe +
  * bitmap block for the new bit) */
 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2)

fs/ocfs2/mmap.c

@@ -37,11 +37,29 @@
 
 #include "ocfs2.h"
 
+#include "aops.h"
 #include "dlmglue.h"
 #include "file.h"
 #include "inode.h"
 #include "mmap.h"
 
+static inline int ocfs2_vm_op_block_sigs(sigset_t *blocked, sigset_t *oldset)
+{
+	/* The best way to deal with signals in the vm path is
+	 * to block them upfront, rather than allowing the
+	 * locking paths to return -ERESTARTSYS. */
+	sigfillset(blocked);
+
+	/* We should technically never get a bad return value
+	 * from sigprocmask */
+	return sigprocmask(SIG_BLOCK, blocked, oldset);
+}
+
+static inline int ocfs2_vm_op_unblock_sigs(sigset_t *oldset)
+{
+	return sigprocmask(SIG_SETMASK, oldset, NULL);
+}
+
 static struct page *ocfs2_nopage(struct vm_area_struct * area,
 				 unsigned long address,
 				 int *type)
@@ -53,14 +71,7 @@ static struct page *ocfs2_nopage(struct vm_area_struct * area,
 	mlog_entry("(area=%p, address=%lu, type=%p)\n", area, address,
 		   type);
 
-	/* The best way to deal with signals in this path is
-	 * to block them upfront, rather than allowing the
-	 * locking paths to return -ERESTARTSYS. */
-	sigfillset(&blocked);
-
-	/* We should technically never get a bad ret return
-	 * from sigprocmask */
-	ret = sigprocmask(SIG_BLOCK, &blocked, &oldset);
+	ret = ocfs2_vm_op_block_sigs(&blocked, &oldset);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out;
@@ -68,7 +79,7 @@ static struct page *ocfs2_nopage(struct vm_area_struct * area,
 
 	page = filemap_nopage(area, address, type);
 
-	ret = sigprocmask(SIG_SETMASK, &oldset, NULL);
+	ret = ocfs2_vm_op_unblock_sigs(&oldset);
 	if (ret < 0)
 		mlog_errno(ret);
 out:
@@ -76,28 +87,136 @@ out:
 	return page;
 }
 
-static struct vm_operations_struct ocfs2_file_vm_ops = {
-	.nopage = ocfs2_nopage,
-};
+static int __ocfs2_page_mkwrite(struct inode *inode, struct buffer_head *di_bh,
+				struct page *page)
+{
+	int ret;
+	struct address_space *mapping = inode->i_mapping;
+	loff_t pos = page->index << PAGE_CACHE_SHIFT;
+	unsigned int len = PAGE_CACHE_SIZE;
+	pgoff_t last_index;
+	struct page *locked_page = NULL;
+	void *fsdata;
+	loff_t size = i_size_read(inode);
 
-int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
+	/*
+	 * Another node might have truncated while we were waiting on
+	 * cluster locks.
+	 */
+	last_index = size >> PAGE_CACHE_SHIFT;
+	if (page->index > last_index) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * The i_size check above doesn't catch the case where nodes
+	 * truncated and then re-extended the file. We'll re-check the
+	 * page mapping after taking the page lock inside of
+	 * ocfs2_write_begin_nolock().
+	 */
+	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * Call ocfs2_write_begin() and ocfs2_write_end() to take
+	 * advantage of the allocation code there. We pass a write
+	 * length of the whole page (chopped to i_size) to make sure
+	 * the whole thing is allocated.
+	 *
+	 * Since we know the page is up to date, we don't have to
+	 * worry about ocfs2_write_begin() skipping some buffer reads
+	 * because the "write" would invalidate their data.
+	 */
+	if (page->index == last_index)
+		len = size & ~PAGE_CACHE_MASK;
+
+	ret = ocfs2_write_begin_nolock(mapping, pos, len, 0, &locked_page,
+				       &fsdata, di_bh, page);
+	if (ret) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page,
+				     fsdata);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+	BUG_ON(ret != len);
+	ret = 0;
+out:
+	return ret;
+}
+
+static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 {
-	int ret = 0, lock_level = 0;
-	struct ocfs2_super *osb = OCFS2_SB(file->f_dentry->d_inode->i_sb);
+	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
+	struct buffer_head *di_bh = NULL;
+	sigset_t blocked, oldset;
+	int ret, ret2;
+
+	ret = ocfs2_vm_op_block_sigs(&blocked, &oldset);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/*
+	 * The cluster locks taken will block a truncate from another
+	 * node. Taking the data lock will also ensure that we don't
+	 * attempt page truncation as part of a downconvert.
+	 */
+	ret = ocfs2_meta_lock(inode, &di_bh, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
 
 	/*
-	 * Only support shared writeable mmap for local mounts which
-	 * don't know about holes.
+	 * The alloc sem should be enough to serialize with
+	 * ocfs2_truncate_file() changing i_size as well as any thread
+	 * modifying the inode btree.
 	 */
-	if ((!ocfs2_mount_local(osb) || ocfs2_sparse_alloc(osb)) &&
-	    ((vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_MAYSHARE)) &&
-	    ((vma->vm_flags & VM_WRITE) || (vma->vm_flags & VM_MAYWRITE))) {
-		mlog(0, "disallow shared writable mmaps %lx\n", vma->vm_flags);
-		/* This is -EINVAL because generic_file_readonly_mmap
-		 * returns it in a similar situation. */
-		return -EINVAL;
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ret = ocfs2_data_lock(inode, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_meta_unlock;
 	}
 
+	ret = __ocfs2_page_mkwrite(inode, di_bh, page);
+
+	ocfs2_data_unlock(inode, 1);
+
+out_meta_unlock:
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	brelse(di_bh);
+	ocfs2_meta_unlock(inode, 1);
+
+out:
+	ret2 = ocfs2_vm_op_unblock_sigs(&oldset);
+	if (ret2 < 0)
+		mlog_errno(ret2);
+
+	return ret;
+}
+
+static struct vm_operations_struct ocfs2_file_vm_ops = {
+	.nopage		= ocfs2_nopage,
+	.page_mkwrite	= ocfs2_page_mkwrite,
+};
+
+int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	int ret = 0, lock_level = 0;
+
 	ret = ocfs2_meta_lock_atime(file->f_dentry->d_inode,
 				    file->f_vfsmnt, &lock_level);
 	if (ret < 0) {

fs/ocfs2/namei.c

@@ -1674,7 +1674,7 @@ 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,
+		status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
 						    new_fe_bh,
 						    handle, data_ac, NULL,
 						    NULL);

fs/ocfs2/ocfs2.h

@@ -219,6 +219,7 @@ struct ocfs2_super
 	u16 max_slots;
 	s16 node_num;
 	s16 slot_num;
+	s16 preferred_slot;
 	int s_sectsize_bits;
 	int s_clustersize;
 	int s_clustersize_bits;
@@ -305,6 +306,19 @@ static inline int ocfs2_sparse_alloc(struct ocfs2_super *osb)
 	return 0;
 }
 
+static inline int ocfs2_writes_unwritten_extents(struct ocfs2_super *osb)
+{
+	/*
+	 * Support for sparse files is a pre-requisite
+	 */
+	if (!ocfs2_sparse_alloc(osb))
+		return 0;
+
+	if (osb->s_feature_ro_compat & OCFS2_FEATURE_RO_COMPAT_UNWRITTEN)
+		return 1;
+	return 0;
+}
+
 /* set / clear functions because cluster events can make these happen
  * in parallel so we want the transitions to be atomic. this also
  * means that any future flags osb_flags must be protected by spinlock

fs/ocfs2/ocfs2_fs.h

@@ -88,7 +88,7 @@
 #define OCFS2_FEATURE_COMPAT_SUPP	OCFS2_FEATURE_COMPAT_BACKUP_SB
 #define OCFS2_FEATURE_INCOMPAT_SUPP	(OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT \
 					 | OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC)
-#define OCFS2_FEATURE_RO_COMPAT_SUPP	0
+#define OCFS2_FEATURE_RO_COMPAT_SUPP	OCFS2_FEATURE_RO_COMPAT_UNWRITTEN
 
 /*
  * Heartbeat-only devices are missing journals and other files.  The
@@ -116,6 +116,11 @@
  */
 #define OCFS2_FEATURE_COMPAT_BACKUP_SB		0x0001
 
+/*
+ * Unwritten extents support.
+ */
+#define OCFS2_FEATURE_RO_COMPAT_UNWRITTEN	0x0001
+
 /* The byte offset of the first backup block will be 1G.
  * The following will be 4G, 16G, 64G, 256G and 1T.
  */
@@ -169,6 +174,32 @@
 #define OCFS2_IOC32_GETFLAGS	_IOR('f', 1, int)
 #define OCFS2_IOC32_SETFLAGS	_IOW('f', 2, int)
 
+/*
+ * Space reservation / allocation / free ioctls and argument structure
+ * are designed to be compatible with XFS.
+ *
+ * ALLOCSP* and FREESP* are not and will never be supported, but are
+ * included here for completeness.
+ */
+struct ocfs2_space_resv {
+	__s16		l_type;
+	__s16		l_whence;
+	__s64		l_start;
+	__s64		l_len;		/* len == 0 means until end of file */
+	__s32		l_sysid;
+	__u32		l_pid;
+	__s32		l_pad[4];	/* reserve area			    */
+};
+
+#define OCFS2_IOC_ALLOCSP		_IOW ('X', 10, struct ocfs2_space_resv)
+#define OCFS2_IOC_FREESP		_IOW ('X', 11, struct ocfs2_space_resv)
+#define OCFS2_IOC_RESVSP		_IOW ('X', 40, struct ocfs2_space_resv)
+#define OCFS2_IOC_UNRESVSP	_IOW ('X', 41, struct ocfs2_space_resv)
+#define OCFS2_IOC_ALLOCSP64	_IOW ('X', 36, struct ocfs2_space_resv)
+#define OCFS2_IOC_FREESP64	_IOW ('X', 37, struct ocfs2_space_resv)
+#define OCFS2_IOC_RESVSP64	_IOW ('X', 42, struct ocfs2_space_resv)
+#define OCFS2_IOC_UNRESVSP64	_IOW ('X', 43, struct ocfs2_space_resv)
+
 /*
  * Journal Flags (ocfs2_dinode.id1.journal1.i_flags)
  */

fs/ocfs2/slot_map.c

@@ -121,17 +121,25 @@ static s16 __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
 	return ret;
 }
 
-static s16 __ocfs2_find_empty_slot(struct ocfs2_slot_info *si)
+static s16 __ocfs2_find_empty_slot(struct ocfs2_slot_info *si, s16 preferred)
 {
 	int i;
 	s16 ret = OCFS2_INVALID_SLOT;
 
+	if (preferred >= 0 && preferred < si->si_num_slots) {
+		if (OCFS2_INVALID_SLOT == si->si_global_node_nums[preferred]) {
+			ret = preferred;
+			goto out;
+		}
+	}
+
 	for(i = 0; i < si->si_num_slots; i++) {
 		if (OCFS2_INVALID_SLOT == si->si_global_node_nums[i]) {
 			ret = (s16) i;
 			break;
 		}
 	}
+out:
 	return ret;
 }
 
@@ -248,7 +256,7 @@ int ocfs2_find_slot(struct ocfs2_super *osb)
 	if (slot == OCFS2_INVALID_SLOT) {
 		/* if no slot yet, then just take 1st available
 		 * one. */
-		slot = __ocfs2_find_empty_slot(si);
+		slot = __ocfs2_find_empty_slot(si, osb->preferred_slot);
 		if (slot == OCFS2_INVALID_SLOT) {
 			spin_unlock(&si->si_lock);
 			mlog(ML_ERROR, "no free slots available!\n");

fs/ocfs2/suballoc.c

@@ -98,14 +98,6 @@ static int ocfs2_relink_block_group(handle_t *handle,
 				    u16 chain);
 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
 						     u32 wanted);
-static int ocfs2_free_suballoc_bits(handle_t *handle,
-				    struct inode *alloc_inode,
-				    struct buffer_head *alloc_bh,
-				    unsigned int start_bit,
-				    u64 bg_blkno,
-				    unsigned int count);
-static inline u64 ocfs2_which_suballoc_group(u64 block,
-					     unsigned int bit);
 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
 						   u64 bg_blkno,
 						   u16 bg_bit_off);
@@ -496,13 +488,7 @@ int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
 
 	(*ac)->ac_bits_wanted = ocfs2_extend_meta_needed(fe);
 	(*ac)->ac_which = OCFS2_AC_USE_META;
-
-#ifndef OCFS2_USE_ALL_METADATA_SUBALLOCATORS
-	slot = 0;
-#else
 	slot = osb->slot_num;
-#endif
-
 	(*ac)->ac_group_search = ocfs2_block_group_search;
 
 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
@@ -1626,12 +1612,12 @@ bail:
 /*
  * expects the suballoc inode to already be locked.
  */
-static int ocfs2_free_suballoc_bits(handle_t *handle,
-				    struct inode *alloc_inode,
-				    struct buffer_head *alloc_bh,
-				    unsigned int start_bit,
-				    u64 bg_blkno,
-				    unsigned int count)
+int ocfs2_free_suballoc_bits(handle_t *handle,
+			     struct inode *alloc_inode,
+			     struct buffer_head *alloc_bh,
+			     unsigned int start_bit,
+			     u64 bg_blkno,
+			     unsigned int count)
 {
 	int status = 0;
 	u32 tmp_used;
@@ -1703,13 +1689,6 @@ bail:
 	return status;
 }
 
-static inline u64 ocfs2_which_suballoc_group(u64 block, unsigned int bit)
-{
-	u64 group = block - (u64) bit;
-
-	return group;
-}
-
 int ocfs2_free_dinode(handle_t *handle,
 		      struct inode *inode_alloc_inode,
 		      struct buffer_head *inode_alloc_bh,
@@ -1723,19 +1702,6 @@ int ocfs2_free_dinode(handle_t *handle,
 					inode_alloc_bh, bit, bg_blkno, 1);
 }
 
-int ocfs2_free_extent_block(handle_t *handle,
-			    struct inode *eb_alloc_inode,
-			    struct buffer_head *eb_alloc_bh,
-			    struct ocfs2_extent_block *eb)
-{
-	u64 blk = le64_to_cpu(eb->h_blkno);
-	u16 bit = le16_to_cpu(eb->h_suballoc_bit);
-	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
-
-	return ocfs2_free_suballoc_bits(handle, eb_alloc_inode, eb_alloc_bh,
-					bit, bg_blkno, 1);
-}
-
 int ocfs2_free_clusters(handle_t *handle,
 		       struct inode *bitmap_inode,
 		       struct buffer_head *bitmap_bh,

fs/ocfs2/suballoc.h

@@ -86,20 +86,29 @@ int ocfs2_claim_clusters(struct ocfs2_super *osb,
 			 u32 *cluster_start,
 			 u32 *num_clusters);
 
+int ocfs2_free_suballoc_bits(handle_t *handle,
+			     struct inode *alloc_inode,
+			     struct buffer_head *alloc_bh,
+			     unsigned int start_bit,
+			     u64 bg_blkno,
+			     unsigned int count);
 int ocfs2_free_dinode(handle_t *handle,
 		      struct inode *inode_alloc_inode,
 		      struct buffer_head *inode_alloc_bh,
 		      struct ocfs2_dinode *di);
-int ocfs2_free_extent_block(handle_t *handle,
-			    struct inode *eb_alloc_inode,
-			    struct buffer_head *eb_alloc_bh,
-			    struct ocfs2_extent_block *eb);
 int ocfs2_free_clusters(handle_t *handle,
 			struct inode *bitmap_inode,
 			struct buffer_head *bitmap_bh,
 			u64 start_blk,
 			unsigned int num_clusters);
 
+static inline u64 ocfs2_which_suballoc_group(u64 block, unsigned int bit)
+{
+	u64 group = block - (u64) bit;
+
+	return group;
+}
+
 static inline u32 ocfs2_cluster_from_desc(struct ocfs2_super *osb,
 					  u64 bg_blkno)
 {

fs/ocfs2/super.c

@@ -82,7 +82,8 @@ MODULE_AUTHOR("Oracle");
 MODULE_LICENSE("GPL");
 
 static int ocfs2_parse_options(struct super_block *sb, char *options,
-			       unsigned long *mount_opt, int is_remount);
+			       unsigned long *mount_opt, s16 *slot,
+			       int is_remount);
 static void ocfs2_put_super(struct super_block *sb);
 static int ocfs2_mount_volume(struct super_block *sb);
 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
@@ -114,8 +115,6 @@ static void ocfs2_write_super(struct super_block *sb);
 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 static void ocfs2_destroy_inode(struct inode *inode);
 
-static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
-
 static const struct super_operations ocfs2_sops = {
 	.statfs		= ocfs2_statfs,
 	.alloc_inode	= ocfs2_alloc_inode,
@@ -140,6 +139,7 @@ enum {
 	Opt_data_ordered,
 	Opt_data_writeback,
 	Opt_atime_quantum,
+	Opt_slot,
 	Opt_err,
 };
 
@@ -154,6 +154,7 @@ static match_table_t tokens = {
 	{Opt_data_ordered, "data=ordered"},
 	{Opt_data_writeback, "data=writeback"},
 	{Opt_atime_quantum, "atime_quantum=%u"},
+	{Opt_slot, "preferred_slot=%u"},
 	{Opt_err, NULL}
 };
 
@@ -318,7 +319,7 @@ static void ocfs2_destroy_inode(struct inode *inode)
 /* From xfs_super.c:xfs_max_file_offset
  * Copyright (c) 2000-2004 Silicon Graphics, Inc.
  */
-static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
+unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
 {
 	unsigned int pagefactor = 1;
 	unsigned int bitshift = BITS_PER_LONG - 1;
@@ -355,9 +356,10 @@ static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 	int incompat_features;
 	int ret = 0;
 	unsigned long parsed_options;
+	s16 slot;
 	struct ocfs2_super *osb = OCFS2_SB(sb);
 
-	if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
+	if (!ocfs2_parse_options(sb, data, &parsed_options, &slot, 1)) {
 		ret = -EINVAL;
 		goto out;
 	}
@@ -534,6 +536,7 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 	struct dentry *root;
 	int status, sector_size;
 	unsigned long parsed_opt;
+	s16 slot;
 	struct inode *inode = NULL;
 	struct ocfs2_super *osb = NULL;
 	struct buffer_head *bh = NULL;
@@ -541,7 +544,7 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 
 	mlog_entry("%p, %p, %i", sb, data, silent);
 
-	if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
+	if (!ocfs2_parse_options(sb, data, &parsed_opt, &slot, 0)) {
 		status = -EINVAL;
 		goto read_super_error;
 	}
@@ -571,6 +574,7 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 	brelse(bh);
 	bh = NULL;
 	osb->s_mount_opt = parsed_opt;
+	osb->preferred_slot = slot;
 
 	sb->s_magic = OCFS2_SUPER_MAGIC;
 
@@ -713,6 +717,7 @@ static struct file_system_type ocfs2_fs_type = {
 static int ocfs2_parse_options(struct super_block *sb,
 			       char *options,
 			       unsigned long *mount_opt,
+			       s16 *slot,
 			       int is_remount)
 {
 	int status;
@@ -722,6 +727,7 @@ static int ocfs2_parse_options(struct super_block *sb,
 		   options ? options : "(none)");
 
 	*mount_opt = 0;
+	*slot = OCFS2_INVALID_SLOT;
 
 	if (!options) {
 		status = 1;
@@ -782,6 +788,15 @@ static int ocfs2_parse_options(struct super_block *sb,
 			else
 				osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
 			break;
+		case Opt_slot:
+			option = 0;
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option)
+				*slot = (s16)option;
+			break;
 		default:
 			mlog(ML_ERROR,
 			     "Unrecognized mount option \"%s\" "

fs/ocfs2/super.h

@@ -45,4 +45,6 @@ void __ocfs2_abort(struct super_block *sb,
 
 #define ocfs2_abort(sb, fmt, args...) __ocfs2_abort(sb, __PRETTY_FUNCTION__, fmt, ##args)
 
+unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
+
 #endif /* OCFS2_SUPER_H */

include/linux/configfs.h

@@ -40,9 +40,9 @@
 #include <linux/types.h>
 #include <linux/list.h>
 #include <linux/kref.h>
+#include <linux/mutex.h>
 
 #include <asm/atomic.h>
-#include <asm/semaphore.h>
 
 #define CONFIGFS_ITEM_NAME_LEN	20
 
@@ -75,7 +75,6 @@ extern void config_item_init(struct config_item *);
 extern void config_item_init_type_name(struct config_item *item,
 				       const char *name,
 				       struct config_item_type *type);
-extern void config_item_cleanup(struct config_item *);
 
 extern struct config_item * config_item_get(struct config_item *);
 extern void config_item_put(struct config_item *);
@@ -87,12 +86,10 @@ struct config_item_type {
 	struct configfs_attribute		**ct_attrs;
 };
 
-
 /**
  *	group - a group of config_items of a specific type, belonging
  *	to a specific subsystem.
  */
-
 struct config_group {
 	struct config_item		cg_item;
 	struct list_head		cg_children;
@@ -100,13 +97,11 @@ struct config_group {
 	struct config_group		**default_groups;
 };
 
-
 extern void config_group_init(struct config_group *group);
 extern void config_group_init_type_name(struct config_group *group,
 					const char *name,
 					struct config_item_type *type);
 
-
 static inline struct config_group *to_config_group(struct config_item *item)
 {
 	return item ? container_of(item,struct config_group,cg_item) : NULL;
@@ -122,7 +117,8 @@ static inline void config_group_put(struct config_group *group)
 	config_item_put(&group->cg_item);
 }
 
-extern struct config_item *config_group_find_obj(struct config_group *, const char *);
+extern struct config_item *config_group_find_item(struct config_group *,
+						  const char *);
 
 
 struct configfs_attribute {
@@ -131,6 +127,22 @@ struct configfs_attribute {
 	mode_t			ca_mode;
 };
 
+/*
+ * Users often need to create attribute structures for their configurable
+ * attributes, containing a configfs_attribute member and function pointers
+ * for the show() and store() operations on that attribute. They can use
+ * this macro (similar to sysfs' __ATTR) to make defining attributes easier.
+ */
+#define __CONFIGFS_ATTR(_name, _mode, _show, _store)			\
+{									\
+	.attr	= {							\
+			.ca_name = __stringify(_name),			\
+			.ca_mode = _mode,				\
+			.ca_owner = THIS_MODULE,			\
+	},								\
+	.show	= _show,						\
+	.store	= _store,						\
+}
 
 /*
  * If allow_link() exists, the item can symlink(2) out to other
@@ -157,12 +169,13 @@ struct configfs_group_operations {
 	struct config_item *(*make_item)(struct config_group *group, const char *name);
 	struct config_group *(*make_group)(struct config_group *group, const char *name);
 	int (*commit_item)(struct config_item *item);
+	void (*disconnect_notify)(struct config_group *group, struct config_item *item);
 	void (*drop_item)(struct config_group *group, struct config_item *item);
 };
 
 struct configfs_subsystem {
 	struct config_group	su_group;
-	struct semaphore	su_sem;
+	struct mutex		su_mutex;
 };
 
 static inline struct configfs_subsystem *to_configfs_subsystem(struct config_group *group)
@@ -175,6 +188,11 @@ static inline struct configfs_subsystem *to_configfs_subsystem(struct config_gro
 int configfs_register_subsystem(struct configfs_subsystem *subsys);
 void configfs_unregister_subsystem(struct configfs_subsystem *subsys);
 
+/* These functions can sleep and can alloc with GFP_KERNEL */
+/* WARNING: These cannot be called underneath configfs callbacks!! */
+int configfs_depend_item(struct configfs_subsystem *subsys, struct config_item *target);
+void configfs_undepend_item(struct configfs_subsystem *subsys, struct config_item *target);
+
 #endif  /* __KERNEL__ */
 
 #endif /* _CONFIGFS_H_ */