Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable into for-linus

Conflicts:
	fs/btrfs/super.c

fs/btrfs/async-thread.c

@@ -48,6 +48,9 @@ struct btrfs_worker_thread {
 	/* number of things on the pending list */
 	atomic_t num_pending;
 
+	/* reference counter for this struct */
+	atomic_t refs;
+
 	unsigned long sequence;
 
 	/* protects the pending list. */
@@ -71,7 +74,12 @@ static void check_idle_worker(struct btrfs_worker_thread *worker)
 		unsigned long flags;
 		spin_lock_irqsave(&worker->workers->lock, flags);
 		worker->idle = 1;
-		list_move(&worker->worker_list, &worker->workers->idle_list);
+
+		/* the list may be empty if the worker is just starting */
+		if (!list_empty(&worker->worker_list)) {
+			list_move(&worker->worker_list,
+				 &worker->workers->idle_list);
+		}
 		spin_unlock_irqrestore(&worker->workers->lock, flags);
 	}
 }
@@ -87,23 +95,49 @@ static void check_busy_worker(struct btrfs_worker_thread *worker)
 		unsigned long flags;
 		spin_lock_irqsave(&worker->workers->lock, flags);
 		worker->idle = 0;
-		list_move_tail(&worker->worker_list,
-			       &worker->workers->worker_list);
+
+		if (!list_empty(&worker->worker_list)) {
+			list_move_tail(&worker->worker_list,
+				      &worker->workers->worker_list);
+		}
 		spin_unlock_irqrestore(&worker->workers->lock, flags);
 	}
 }
 
-static noinline int run_ordered_completions(struct btrfs_workers *workers,
-					    struct btrfs_work *work)
+static void check_pending_worker_creates(struct btrfs_worker_thread *worker)
 {
+	struct btrfs_workers *workers = worker->workers;
 	unsigned long flags;
 
+	rmb();
+	if (!workers->atomic_start_pending)
+		return;
+
+	spin_lock_irqsave(&workers->lock, flags);
+	if (!workers->atomic_start_pending)
+		goto out;
+
+	workers->atomic_start_pending = 0;
+	if (workers->num_workers >= workers->max_workers)
+		goto out;
+
+	spin_unlock_irqrestore(&workers->lock, flags);
+	btrfs_start_workers(workers, 1);
+	return;
+
+out:
+	spin_unlock_irqrestore(&workers->lock, flags);
+}
+
+static noinline int run_ordered_completions(struct btrfs_workers *workers,
+					    struct btrfs_work *work)
+{
 	if (!workers->ordered)
 		return 0;
 
 	set_bit(WORK_DONE_BIT, &work->flags);
 
-	spin_lock_irqsave(&workers->lock, flags);
+	spin_lock(&workers->order_lock);
 
 	while (1) {
 		if (!list_empty(&workers->prio_order_list)) {
@@ -126,45 +160,118 @@ static noinline int run_ordered_completions(struct btrfs_workers *workers,
 		if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
 			break;
 
-		spin_unlock_irqrestore(&workers->lock, flags);
+		spin_unlock(&workers->order_lock);
 
 		work->ordered_func(work);
 
 		/* now take the lock again and call the freeing code */
-		spin_lock_irqsave(&workers->lock, flags);
+		spin_lock(&workers->order_lock);
 		list_del(&work->order_list);
 		work->ordered_free(work);
 	}
 
-	spin_unlock_irqrestore(&workers->lock, flags);
+	spin_unlock(&workers->order_lock);
 	return 0;
 }
 
+static void put_worker(struct btrfs_worker_thread *worker)
+{
+	if (atomic_dec_and_test(&worker->refs))
+		kfree(worker);
+}
+
+static int try_worker_shutdown(struct btrfs_worker_thread *worker)
+{
+	int freeit = 0;
+
+	spin_lock_irq(&worker->lock);
+	spin_lock(&worker->workers->lock);
+	if (worker->workers->num_workers > 1 &&
+	    worker->idle &&
+	    !worker->working &&
+	    !list_empty(&worker->worker_list) &&
+	    list_empty(&worker->prio_pending) &&
+	    list_empty(&worker->pending) &&
+	    atomic_read(&worker->num_pending) == 0) {
+		freeit = 1;
+		list_del_init(&worker->worker_list);
+		worker->workers->num_workers--;
+	}
+	spin_unlock(&worker->workers->lock);
+	spin_unlock_irq(&worker->lock);
+
+	if (freeit)
+		put_worker(worker);
+	return freeit;
+}
+
+static struct btrfs_work *get_next_work(struct btrfs_worker_thread *worker,
+					struct list_head *prio_head,
+					struct list_head *head)
+{
+	struct btrfs_work *work = NULL;
+	struct list_head *cur = NULL;
+
+	if(!list_empty(prio_head))
+		cur = prio_head->next;
+
+	smp_mb();
+	if (!list_empty(&worker->prio_pending))
+		goto refill;
+
+	if (!list_empty(head))
+		cur = head->next;
+
+	if (cur)
+		goto out;
+
+refill:
+	spin_lock_irq(&worker->lock);
+	list_splice_tail_init(&worker->prio_pending, prio_head);
+	list_splice_tail_init(&worker->pending, head);
+
+	if (!list_empty(prio_head))
+		cur = prio_head->next;
+	else if (!list_empty(head))
+		cur = head->next;
+	spin_unlock_irq(&worker->lock);
+
+	if (!cur)
+		goto out_fail;
+
+out:
+	work = list_entry(cur, struct btrfs_work, list);
+
+out_fail:
+	return work;
+}
+
 /*
  * main loop for servicing work items
  */
 static int worker_loop(void *arg)
 {
 	struct btrfs_worker_thread *worker = arg;
-	struct list_head *cur;
+	struct list_head head;
+	struct list_head prio_head;
 	struct btrfs_work *work;
+
+	INIT_LIST_HEAD(&head);
+	INIT_LIST_HEAD(&prio_head);
+
 	do {
-		spin_lock_irq(&worker->lock);
-again_locked:
+again:
 		while (1) {
-			if (!list_empty(&worker->prio_pending))
-				cur = worker->prio_pending.next;
-			else if (!list_empty(&worker->pending))
-				cur = worker->pending.next;
-			else
+
+
+			work = get_next_work(worker, &prio_head, &head);
+			if (!work)
 				break;
 
-			work = list_entry(cur, struct btrfs_work, list);
 			list_del(&work->list);
 			clear_bit(WORK_QUEUED_BIT, &work->flags);
 
 			work->worker = worker;
-			spin_unlock_irq(&worker->lock);
 
 			work->func(work);
 
@@ -175,9 +282,13 @@ again_locked:
 			 */
 			run_ordered_completions(worker->workers, work);
 
-			spin_lock_irq(&worker->lock);
-			check_idle_worker(worker);
+			check_pending_worker_creates(worker);
+
 		}
+
+		spin_lock_irq(&worker->lock);
+		check_idle_worker(worker);
+
 		if (freezing(current)) {
 			worker->working = 0;
 			spin_unlock_irq(&worker->lock);
@@ -216,8 +327,10 @@ again_locked:
 				spin_lock_irq(&worker->lock);
 				set_current_state(TASK_INTERRUPTIBLE);
 				if (!list_empty(&worker->pending) ||
-				    !list_empty(&worker->prio_pending))
-					goto again_locked;
+				    !list_empty(&worker->prio_pending)) {
+					spin_unlock_irq(&worker->lock);
+					goto again;
+				}
 
 				/*
 				 * this makes sure we get a wakeup when someone
@@ -226,8 +339,13 @@ again_locked:
 				worker->working = 0;
 				spin_unlock_irq(&worker->lock);
 
-				if (!kthread_should_stop())
-					schedule();
+				if (!kthread_should_stop()) {
+					schedule_timeout(HZ * 120);
+					if (!worker->working &&
+					    try_worker_shutdown(worker)) {
+						return 0;
+					}
+				}
 			}
 			__set_current_state(TASK_RUNNING);
 		}
@@ -242,16 +360,30 @@ int btrfs_stop_workers(struct btrfs_workers *workers)
 {
 	struct list_head *cur;
 	struct btrfs_worker_thread *worker;
+	int can_stop;
 
+	spin_lock_irq(&workers->lock);
 	list_splice_init(&workers->idle_list, &workers->worker_list);
 	while (!list_empty(&workers->worker_list)) {
 		cur = workers->worker_list.next;
 		worker = list_entry(cur, struct btrfs_worker_thread,
 				    worker_list);
-		kthread_stop(worker->task);
-		list_del(&worker->worker_list);
-		kfree(worker);
+
+		atomic_inc(&worker->refs);
+		workers->num_workers -= 1;
+		if (!list_empty(&worker->worker_list)) {
+			list_del_init(&worker->worker_list);
+			put_worker(worker);
+			can_stop = 1;
+		} else
+			can_stop = 0;
+		spin_unlock_irq(&workers->lock);
+		if (can_stop)
+			kthread_stop(worker->task);
+		spin_lock_irq(&workers->lock);
+		put_worker(worker);
 	}
+	spin_unlock_irq(&workers->lock);
 	return 0;
 }
 
@@ -266,10 +398,13 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
 	INIT_LIST_HEAD(&workers->order_list);
 	INIT_LIST_HEAD(&workers->prio_order_list);
 	spin_lock_init(&workers->lock);
+	spin_lock_init(&workers->order_lock);
 	workers->max_workers = max;
 	workers->idle_thresh = 32;
 	workers->name = name;
 	workers->ordered = 0;
+	workers->atomic_start_pending = 0;
+	workers->atomic_worker_start = 0;
 }
 
 /*
@@ -293,7 +428,9 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
 		INIT_LIST_HEAD(&worker->prio_pending);
 		INIT_LIST_HEAD(&worker->worker_list);
 		spin_lock_init(&worker->lock);
+
 		atomic_set(&worker->num_pending, 0);
+		atomic_set(&worker->refs, 1);
 		worker->workers = workers;
 		worker->task = kthread_run(worker_loop, worker,
 					   "btrfs-%s-%d", workers->name,
@@ -303,7 +440,6 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
 			kfree(worker);
 			goto fail;
 		}
-
 		spin_lock_irq(&workers->lock);
 		list_add_tail(&worker->worker_list, &workers->idle_list);
 		worker->idle = 1;
@@ -350,7 +486,6 @@ static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
 	 */
 	next = workers->worker_list.next;
 	worker = list_entry(next, struct btrfs_worker_thread, worker_list);
-	atomic_inc(&worker->num_pending);
 	worker->sequence++;
 
 	if (worker->sequence % workers->idle_thresh == 0)
@@ -367,28 +502,18 @@ static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
 {
 	struct btrfs_worker_thread *worker;
 	unsigned long flags;
+	struct list_head *fallback;
 
 again:
 	spin_lock_irqsave(&workers->lock, flags);
 	worker = next_worker(workers);
-	spin_unlock_irqrestore(&workers->lock, flags);
 
 	if (!worker) {
-		spin_lock_irqsave(&workers->lock, flags);
 		if (workers->num_workers >= workers->max_workers) {
-			struct list_head *fallback = NULL;
-			/*
-			 * we have failed to find any workers, just
-			 * return the force one
-			 */
-			if (!list_empty(&workers->worker_list))
-				fallback = workers->worker_list.next;
-			if (!list_empty(&workers->idle_list))
-				fallback = workers->idle_list.next;
-			BUG_ON(!fallback);
-			worker = list_entry(fallback,
-				  struct btrfs_worker_thread, worker_list);
-			spin_unlock_irqrestore(&workers->lock, flags);
+			goto fallback;
+		} else if (workers->atomic_worker_start) {
+			workers->atomic_start_pending = 1;
+			goto fallback;
 		} else {
 			spin_unlock_irqrestore(&workers->lock, flags);
 			/* we're below the limit, start another worker */
@@ -396,6 +521,28 @@ again:
 			goto again;
 		}
 	}
+	goto found;
+
+fallback:
+	fallback = NULL;
+	/*
+	 * we have failed to find any workers, just
+	 * return the first one we can find.
+	 */
+	if (!list_empty(&workers->worker_list))
+		fallback = workers->worker_list.next;
+	if (!list_empty(&workers->idle_list))
+		fallback = workers->idle_list.next;
+	BUG_ON(!fallback);
+	worker = list_entry(fallback,
+		  struct btrfs_worker_thread, worker_list);
+found:
+	/*
+	 * this makes sure the worker doesn't exit before it is placed
+	 * onto a busy/idle list
+	 */
+	atomic_inc(&worker->num_pending);
+	spin_unlock_irqrestore(&workers->lock, flags);
 	return worker;
 }
 
@@ -427,7 +574,7 @@ int btrfs_requeue_work(struct btrfs_work *work)
 		spin_lock(&worker->workers->lock);
 		worker->idle = 0;
 		list_move_tail(&worker->worker_list,
-			       &worker->workers->worker_list);
+			      &worker->workers->worker_list);
 		spin_unlock(&worker->workers->lock);
 	}
 	if (!worker->working) {
@@ -435,9 +582,9 @@ int btrfs_requeue_work(struct btrfs_work *work)
 		worker->working = 1;
 	}
 
-	spin_unlock_irqrestore(&worker->lock, flags);
 	if (wake)
 		wake_up_process(worker->task);
+	spin_unlock_irqrestore(&worker->lock, flags);
 out:
 
 	return 0;
@@ -463,14 +610,18 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
 
 	worker = find_worker(workers);
 	if (workers->ordered) {
-		spin_lock_irqsave(&workers->lock, flags);
+		/*
+		 * you're not allowed to do ordered queues from an
+		 * interrupt handler
+		 */
+		spin_lock(&workers->order_lock);
 		if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) {
 			list_add_tail(&work->order_list,
 				      &workers->prio_order_list);
 		} else {
 			list_add_tail(&work->order_list, &workers->order_list);
 		}
-		spin_unlock_irqrestore(&workers->lock, flags);
+		spin_unlock(&workers->order_lock);
 	} else {
 		INIT_LIST_HEAD(&work->order_list);
 	}
@@ -481,7 +632,6 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
 		list_add_tail(&work->list, &worker->prio_pending);
 	else
 		list_add_tail(&work->list, &worker->pending);
-	atomic_inc(&worker->num_pending);
 	check_busy_worker(worker);
 
 	/*
@@ -492,10 +642,10 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
 		wake = 1;
 	worker->working = 1;
 
-	spin_unlock_irqrestore(&worker->lock, flags);
-
 	if (wake)
 		wake_up_process(worker->task);
+	spin_unlock_irqrestore(&worker->lock, flags);
+
 out:
 	return 0;
 }

fs/btrfs/async-thread.h

@@ -73,6 +73,15 @@ struct btrfs_workers {
 	/* force completions in the order they were queued */
 	int ordered;
 
+	/* more workers required, but in an interrupt handler */
+	int atomic_start_pending;
+
+	/*
+	 * are we allowed to sleep while starting workers or are we required
+	 * to start them at a later time?
+	 */
+	int atomic_worker_start;
+
 	/* list with all the work threads.  The workers on the idle thread
 	 * may be actively servicing jobs, but they haven't yet hit the
 	 * idle thresh limit above.
@@ -90,6 +99,9 @@ struct btrfs_workers {
 	/* lock for finding the next worker thread to queue on */
 	spinlock_t lock;
 
+	/* lock for the ordered lists */
+	spinlock_t order_lock;
+
 	/* extra name for this worker, used for current->name */
 	char *name;
 };

fs/btrfs/btrfs_inode.h

@@ -138,6 +138,7 @@ struct btrfs_inode {
 	 * of these.
 	 */
 	unsigned ordered_data_close:1;
+	unsigned dummy_inode:1;
 
 	struct inode vfs_inode;
 };

fs/btrfs/compression.c

@@ -506,10 +506,10 @@ static noinline int add_ra_bio_pages(struct inode *inode,
 		 */
 		set_page_extent_mapped(page);
 		lock_extent(tree, last_offset, end, GFP_NOFS);
-		spin_lock(&em_tree->lock);
+		read_lock(&em_tree->lock);
 		em = lookup_extent_mapping(em_tree, last_offset,
 					   PAGE_CACHE_SIZE);
-		spin_unlock(&em_tree->lock);
+		read_unlock(&em_tree->lock);
 
 		if (!em || last_offset < em->start ||
 		    (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
@@ -593,11 +593,11 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
 	em_tree = &BTRFS_I(inode)->extent_tree;
 
 	/* we need the actual starting offset of this extent in the file */
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree,
 				   page_offset(bio->bi_io_vec->bv_page),
 				   PAGE_CACHE_SIZE);
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 
 	compressed_len = em->block_len;
 	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);

fs/btrfs/ctree.c

@@ -2853,6 +2853,12 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
 	int split;
 	int num_doubles = 0;
 
+	l = path->nodes[0];
+	slot = path->slots[0];
+	if (extend && data_size + btrfs_item_size_nr(l, slot) +
+	    sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
+		return -EOVERFLOW;
+
 	/* first try to make some room by pushing left and right */
 	if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
 		wret = push_leaf_right(trans, root, path, data_size, 0);

fs/btrfs/ctree.h

@@ -114,6 +114,10 @@ struct btrfs_ordered_sum;
  */
 #define BTRFS_DEV_ITEMS_OBJECTID 1ULL
 
+#define BTRFS_BTREE_INODE_OBJECTID 1
+
+#define BTRFS_EMPTY_SUBVOL_DIR_OBJECTID 2
+
 /*
  * we can actually store much bigger names, but lets not confuse the rest
  * of linux
@@ -670,6 +674,7 @@ struct btrfs_space_info {
 	u64 bytes_reserved;	/* total bytes the allocator has reserved for
 				   current allocations */
 	u64 bytes_readonly;	/* total bytes that are read only */
+	u64 bytes_super;	/* total bytes reserved for the super blocks */
 
 	/* delalloc accounting */
 	u64 bytes_delalloc;	/* number of bytes reserved for allocation,
@@ -726,6 +731,15 @@ enum btrfs_caching_type {
 	BTRFS_CACHE_FINISHED	= 2,
 };
 
+struct btrfs_caching_control {
+	struct list_head list;
+	struct mutex mutex;
+	wait_queue_head_t wait;
+	struct btrfs_block_group_cache *block_group;
+	u64 progress;
+	atomic_t count;
+};
+
 struct btrfs_block_group_cache {
 	struct btrfs_key key;
 	struct btrfs_block_group_item item;
@@ -733,6 +747,7 @@ struct btrfs_block_group_cache {
 	spinlock_t lock;
 	u64 pinned;
 	u64 reserved;
+	u64 bytes_super;
 	u64 flags;
 	u64 sectorsize;
 	int extents_thresh;
@@ -742,8 +757,9 @@ struct btrfs_block_group_cache {
 	int dirty;
 
 	/* cache tracking stuff */
-	wait_queue_head_t caching_q;
 	int cached;
+	struct btrfs_caching_control *caching_ctl;
+	u64 last_byte_to_unpin;
 
 	struct btrfs_space_info *space_info;
 
@@ -782,13 +798,16 @@ struct btrfs_fs_info {
 
 	/* the log root tree is a directory of all the other log roots */
 	struct btrfs_root *log_root_tree;
+
+	spinlock_t fs_roots_radix_lock;
 	struct radix_tree_root fs_roots_radix;
 
 	/* block group cache stuff */
 	spinlock_t block_group_cache_lock;
 	struct rb_root block_group_cache_tree;
 
-	struct extent_io_tree pinned_extents;
+	struct extent_io_tree freed_extents[2];
+	struct extent_io_tree *pinned_extents;
 
 	/* logical->physical extent mapping */
 	struct btrfs_mapping_tree mapping_tree;
@@ -822,11 +841,7 @@ struct btrfs_fs_info {
 	struct mutex transaction_kthread_mutex;
 	struct mutex cleaner_mutex;
 	struct mutex chunk_mutex;
-	struct mutex drop_mutex;
 	struct mutex volume_mutex;
-	struct mutex tree_reloc_mutex;
-	struct rw_semaphore extent_commit_sem;
-
 	/*
 	 * this protects the ordered operations list only while we are
 	 * processing all of the entries on it.  This way we make
@@ -835,10 +850,16 @@ struct btrfs_fs_info {
 	 * before jumping into the main commit.
 	 */
 	struct mutex ordered_operations_mutex;
+	struct rw_semaphore extent_commit_sem;
+
+	struct rw_semaphore subvol_sem;
+
+	struct srcu_struct subvol_srcu;
 
 	struct list_head trans_list;
 	struct list_head hashers;
 	struct list_head dead_roots;
+	struct list_head caching_block_groups;
 
 	atomic_t nr_async_submits;
 	atomic_t async_submit_draining;
@@ -996,10 +1017,12 @@ struct btrfs_root {
 	u32 stripesize;
 
 	u32 type;
-	u64 highest_inode;
-	u64 last_inode_alloc;
+
+	u64 highest_objectid;
 	int ref_cows;
 	int track_dirty;
+	int in_radix;
+
 	u64 defrag_trans_start;
 	struct btrfs_key defrag_progress;
 	struct btrfs_key defrag_max;
@@ -1920,8 +1943,8 @@ void btrfs_put_block_group(struct btrfs_block_group_cache *cache);
 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root, unsigned long count);
 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len);
-int btrfs_update_pinned_extents(struct btrfs_root *root,
-				u64 bytenr, u64 num, int pin);
+int btrfs_pin_extent(struct btrfs_root *root,
+		     u64 bytenr, u64 num, int reserved);
 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root, struct extent_buffer *leaf);
 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
@@ -1971,9 +1994,10 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
 		      u64 root_objectid, u64 owner, u64 offset);
 
 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len);
+int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root);
 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
-			       struct btrfs_root *root,
-			       struct extent_io_tree *unpin);
+			       struct btrfs_root *root);
 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 			 struct btrfs_root *root,
 			 u64 bytenr, u64 num_bytes, u64 parent,
@@ -1984,6 +2008,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr);
 int btrfs_free_block_groups(struct btrfs_fs_info *info);
 int btrfs_read_block_groups(struct btrfs_root *root);
+int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr);
 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root, u64 bytes_used,
 			   u64 type, u64 chunk_objectid, u64 chunk_offset,
@@ -2006,7 +2031,6 @@ void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
 				 u64 bytes);
 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
 			      u64 bytes);
-void btrfs_free_pinned_extents(struct btrfs_fs_info *info);
 /* ctree.c */
 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 		     int level, int *slot);
@@ -2100,12 +2124,15 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 			struct extent_buffer *parent);
 /* root-item.c */
 int btrfs_find_root_ref(struct btrfs_root *tree_root,
-		   struct btrfs_path *path,
-		   u64 root_id, u64 ref_id);
+			struct btrfs_path *path,
+			u64 root_id, u64 ref_id);
 int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *tree_root,
-		       u64 root_id, u8 type, u64 ref_id,
-		       u64 dirid, u64 sequence,
+		       u64 root_id, u64 ref_id, u64 dirid, u64 sequence,
+		       const char *name, int name_len);
+int btrfs_del_root_ref(struct btrfs_trans_handle *trans,
+		       struct btrfs_root *tree_root,
+		       u64 root_id, u64 ref_id, u64 dirid, u64 *sequence,
 		       const char *name, int name_len);
 int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		   struct btrfs_key *key);
@@ -2120,6 +2147,7 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct
 int btrfs_search_root(struct btrfs_root *root, u64 search_start,
 		      u64 *found_objectid);
 int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid);
+int btrfs_find_orphan_roots(struct btrfs_root *tree_root);
 int btrfs_set_root_node(struct btrfs_root_item *item,
 			struct extent_buffer *node);
 /* dir-item.c */
@@ -2138,6 +2166,10 @@ btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
 			    struct btrfs_path *path, u64 dir,
 			    u64 objectid, const char *name, int name_len,
 			    int mod);
+struct btrfs_dir_item *
+btrfs_search_dir_index_item(struct btrfs_root *root,
+			    struct btrfs_path *path, u64 dirid,
+			    const char *name, int name_len);
 struct btrfs_dir_item *btrfs_match_dir_item_name(struct btrfs_root *root,
 			      struct btrfs_path *path,
 			      const char *name, int name_len);
@@ -2160,6 +2192,7 @@ int btrfs_insert_orphan_item(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root, u64 offset);
 int btrfs_del_orphan_item(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root, u64 offset);
+int btrfs_find_orphan_item(struct btrfs_root *root, u64 offset);
 
 /* inode-map.c */
 int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
@@ -2232,6 +2265,10 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
 int btrfs_add_link(struct btrfs_trans_handle *trans,
 		   struct inode *parent_inode, struct inode *inode,
 		   const char *name, int name_len, int add_backref, u64 index);
+int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct inode *dir, u64 objectid,
+			const char *name, int name_len);
 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root,
 			       struct inode *inode, u64 new_size,
@@ -2242,7 +2279,7 @@ int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end);
 int btrfs_writepages(struct address_space *mapping,
 		     struct writeback_control *wbc);
 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
-			     struct btrfs_root *new_root, struct dentry *dentry,
+			     struct btrfs_root *new_root,
 			     u64 new_dirid, u64 alloc_hint);
 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 			 size_t size, struct bio *bio, unsigned long bio_flags);
@@ -2258,6 +2295,7 @@ int btrfs_write_inode(struct inode *inode, int wait);
 void btrfs_dirty_inode(struct inode *inode);
 struct inode *btrfs_alloc_inode(struct super_block *sb);
 void btrfs_destroy_inode(struct inode *inode);
+void btrfs_drop_inode(struct inode *inode);
 int btrfs_init_cachep(void);
 void btrfs_destroy_cachep(void);
 long btrfs_ioctl_trans_end(struct file *file);
@@ -2275,6 +2313,8 @@ int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode);
 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode);
 void btrfs_orphan_cleanup(struct btrfs_root *root);
 int btrfs_cont_expand(struct inode *inode, loff_t size);
+int btrfs_invalidate_inodes(struct btrfs_root *root);
+extern struct dentry_operations btrfs_dentry_operations;
 
 /* ioctl.c */
 long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
@@ -2290,7 +2330,7 @@ extern struct file_operations btrfs_file_operations;
 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root, struct inode *inode,
 		       u64 start, u64 end, u64 locked_end,
-		       u64 inline_limit, u64 *hint_block);
+		       u64 inline_limit, u64 *hint_block, int drop_cache);
 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
 			      struct inode *inode, u64 start, u64 end);

fs/btrfs/dir-item.c

@@ -281,6 +281,53 @@ btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
 	return btrfs_match_dir_item_name(root, path, name, name_len);
 }
 
+struct btrfs_dir_item *
+btrfs_search_dir_index_item(struct btrfs_root *root,
+			    struct btrfs_path *path, u64 dirid,
+			    const char *name, int name_len)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_dir_item *di;
+	struct btrfs_key key;
+	u32 nritems;
+	int ret;
+
+	key.objectid = dirid;
+	key.type = BTRFS_DIR_INDEX_KEY;
+	key.offset = 0;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		return ERR_PTR(ret);
+
+	leaf = path->nodes[0];
+	nritems = btrfs_header_nritems(leaf);
+
+	while (1) {
+		if (path->slots[0] >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret < 0)
+				return ERR_PTR(ret);
+			if (ret > 0)
+				break;
+			leaf = path->nodes[0];
+			nritems = btrfs_header_nritems(leaf);
+			continue;
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		if (key.objectid != dirid || key.type != BTRFS_DIR_INDEX_KEY)
+			break;
+
+		di = btrfs_match_dir_item_name(root, path, name, name_len);
+		if (di)
+			return di;
+
+		path->slots[0]++;
+	}
+	return NULL;
+}
+
 struct btrfs_dir_item *btrfs_lookup_xattr(struct btrfs_trans_handle *trans,
 					  struct btrfs_root *root,
 					  struct btrfs_path *path, u64 dir,

fs/btrfs/disk-io.c

@@ -41,6 +41,7 @@
 
 static struct extent_io_ops btree_extent_io_ops;
 static void end_workqueue_fn(struct btrfs_work *work);
+static void free_fs_root(struct btrfs_root *root);
 
 static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
 
@@ -123,15 +124,15 @@ static struct extent_map *btree_get_extent(struct inode *inode,
 	struct extent_map *em;
 	int ret;
 
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree, start, len);
 	if (em) {
 		em->bdev =
 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
-		spin_unlock(&em_tree->lock);
+		read_unlock(&em_tree->lock);
 		goto out;
 	}
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 
 	em = alloc_extent_map(GFP_NOFS);
 	if (!em) {
@@ -144,7 +145,7 @@ static struct extent_map *btree_get_extent(struct inode *inode,
 	em->block_start = 0;
 	em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 
-	spin_lock(&em_tree->lock);
+	write_lock(&em_tree->lock);
 	ret = add_extent_mapping(em_tree, em);
 	if (ret == -EEXIST) {
 		u64 failed_start = em->start;
@@ -163,7 +164,7 @@ static struct extent_map *btree_get_extent(struct inode *inode,
 		free_extent_map(em);
 		em = NULL;
 	}
-	spin_unlock(&em_tree->lock);
+	write_unlock(&em_tree->lock);
 
 	if (ret)
 		em = ERR_PTR(ret);
@@ -895,8 +896,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 	root->fs_info = fs_info;
 	root->objectid = objectid;
 	root->last_trans = 0;
-	root->highest_inode = 0;
-	root->last_inode_alloc = 0;
+	root->highest_objectid = 0;
 	root->name = NULL;
 	root->in_sysfs = 0;
 	root->inode_tree.rb_node = NULL;
@@ -952,14 +952,16 @@ static int find_and_setup_root(struct btrfs_root *tree_root,
 		     root, fs_info, objectid);
 	ret = btrfs_find_last_root(tree_root, objectid,
 				   &root->root_item, &root->root_key);
+	if (ret > 0)
+		return -ENOENT;
 	BUG_ON(ret);
 
 	generation = btrfs_root_generation(&root->root_item);
 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 				     blocksize, generation);
-	root->commit_root = btrfs_root_node(root);
 	BUG_ON(!root->node);
+	root->commit_root = btrfs_root_node(root);
 	return 0;
 }
 
@@ -1095,7 +1097,6 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
 	struct btrfs_path *path;
 	struct extent_buffer *l;
-	u64 highest_inode;
 	u64 generation;
 	u32 blocksize;
 	int ret = 0;
@@ -1110,7 +1111,7 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 			kfree(root);
 			return ERR_PTR(ret);
 		}
-		goto insert;
+		goto out;
 	}
 
 	__setup_root(tree_root->nodesize, tree_root->leafsize,
@@ -1120,39 +1121,30 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 	ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
-	if (ret != 0) {
-		if (ret > 0)
-			ret = -ENOENT;
-		goto out;
+	if (ret == 0) {
+		l = path->nodes[0];
+		read_extent_buffer(l, &root->root_item,
+				btrfs_item_ptr_offset(l, path->slots[0]),
+				sizeof(root->root_item));
+		memcpy(&root->root_key, location, sizeof(*location));
 	}
-	l = path->nodes[0];
-	read_extent_buffer(l, &root->root_item,
-	       btrfs_item_ptr_offset(l, path->slots[0]),
-	       sizeof(root->root_item));
-	memcpy(&root->root_key, location, sizeof(*location));
-	ret = 0;
-out:
-	btrfs_release_path(root, path);
 	btrfs_free_path(path);
 	if (ret) {
-		kfree(root);
+		if (ret > 0)
+			ret = -ENOENT;
 		return ERR_PTR(ret);
 	}
+
 	generation = btrfs_root_generation(&root->root_item);
 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 				     blocksize, generation);
 	root->commit_root = btrfs_root_node(root);
 	BUG_ON(!root->node);
-insert:
-	if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
+out:
+	if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
 		root->ref_cows = 1;
-		ret = btrfs_find_highest_inode(root, &highest_inode);
-		if (ret == 0) {
-			root->highest_inode = highest_inode;
-			root->last_inode_alloc = highest_inode;
-		}
-	}
+
 	return root;
 }
 
@@ -1187,39 +1179,66 @@ struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 		return fs_info->dev_root;
 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
 		return fs_info->csum_root;
-
+again:
+	spin_lock(&fs_info->fs_roots_radix_lock);
 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
 				 (unsigned long)location->objectid);
+	spin_unlock(&fs_info->fs_roots_radix_lock);
 	if (root)
 		return root;
 
+	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
+	if (ret == 0)
+		ret = -ENOENT;
+	if (ret < 0)
+		return ERR_PTR(ret);
+
 	root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
 	if (IS_ERR(root))
 		return root;
 
+	WARN_ON(btrfs_root_refs(&root->root_item) == 0);
 	set_anon_super(&root->anon_super, NULL);
 
+	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
+	if (ret)
+		goto fail;
+
+	spin_lock(&fs_info->fs_roots_radix_lock);
 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
 				(unsigned long)root->root_key.objectid,
 				root);
+	if (ret == 0)
+		root->in_radix = 1;
+	spin_unlock(&fs_info->fs_roots_radix_lock);
+	radix_tree_preload_end();
 	if (ret) {
-		free_extent_buffer(root->node);
-		kfree(root);
-		return ERR_PTR(ret);
+		if (ret == -EEXIST) {
+			free_fs_root(root);
+			goto again;
+		}
+		goto fail;
 	}
-	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
-		ret = btrfs_find_dead_roots(fs_info->tree_root,
-					    root->root_key.objectid);
-		BUG_ON(ret);
+
+	ret = btrfs_find_dead_roots(fs_info->tree_root,
+				    root->root_key.objectid);
+	WARN_ON(ret);
+
+	if (!(fs_info->sb->s_flags & MS_RDONLY))
 		btrfs_orphan_cleanup(root);
-	}
+
 	return root;
+fail:
+	free_fs_root(root);
+	return ERR_PTR(ret);
 }
 
 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 				      struct btrfs_key *location,
 				      const char *name, int namelen)
 {
+	return btrfs_read_fs_root_no_name(fs_info, location);
+#if 0
 	struct btrfs_root *root;
 	int ret;
 
@@ -1236,7 +1255,7 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 		kfree(root);
 		return ERR_PTR(ret);
 	}
-#if 0
+
 	ret = btrfs_sysfs_add_root(root);
 	if (ret) {
 		free_extent_buffer(root->node);
@@ -1244,9 +1263,9 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 		kfree(root);
 		return ERR_PTR(ret);
 	}
-#endif
 	root->in_sysfs = 1;
 	return root;
+#endif
 }
 
 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
@@ -1325,9 +1344,9 @@ static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 	offset = page_offset(page);
 
 	em_tree = &BTRFS_I(inode)->extent_tree;
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 	if (!em) {
 		__unplug_io_fn(bdi, page);
 		return;
@@ -1360,8 +1379,10 @@ static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 
 	err = bdi_register(bdi, NULL, "btrfs-%d",
 				atomic_inc_return(&btrfs_bdi_num));
-	if (err)
+	if (err) {
+		bdi_destroy(bdi);
 		return err;
+	}
 
 	bdi->ra_pages	= default_backing_dev_info.ra_pages;
 	bdi->unplug_io_fn	= btrfs_unplug_io_fn;
@@ -1451,9 +1472,12 @@ static int cleaner_kthread(void *arg)
 			break;
 
 		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
-		mutex_lock(&root->fs_info->cleaner_mutex);
-		btrfs_clean_old_snapshots(root);
-		mutex_unlock(&root->fs_info->cleaner_mutex);
+
+		if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
+		    mutex_trylock(&root->fs_info->cleaner_mutex)) {
+			btrfs_clean_old_snapshots(root);
+			mutex_unlock(&root->fs_info->cleaner_mutex);
+		}
 
 		if (freezing(current)) {
 			refrigerator();
@@ -1558,15 +1582,36 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 		err = -ENOMEM;
 		goto fail;
 	}
-	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
+
+	ret = init_srcu_struct(&fs_info->subvol_srcu);
+	if (ret) {
+		err = ret;
+		goto fail;
+	}
+
+	ret = setup_bdi(fs_info, &fs_info->bdi);
+	if (ret) {
+		err = ret;
+		goto fail_srcu;
+	}
+
+	fs_info->btree_inode = new_inode(sb);
+	if (!fs_info->btree_inode) {
+		err = -ENOMEM;
+		goto fail_bdi;
+	}
+
+	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
 	INIT_LIST_HEAD(&fs_info->trans_list);
 	INIT_LIST_HEAD(&fs_info->dead_roots);
 	INIT_LIST_HEAD(&fs_info->hashers);
 	INIT_LIST_HEAD(&fs_info->delalloc_inodes);
 	INIT_LIST_HEAD(&fs_info->ordered_operations);
+	INIT_LIST_HEAD(&fs_info->caching_block_groups);
 	spin_lock_init(&fs_info->delalloc_lock);
 	spin_lock_init(&fs_info->new_trans_lock);
 	spin_lock_init(&fs_info->ref_cache_lock);
+	spin_lock_init(&fs_info->fs_roots_radix_lock);
 
 	init_completion(&fs_info->kobj_unregister);
 	fs_info->tree_root = tree_root;
@@ -1585,11 +1630,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	fs_info->sb = sb;
 	fs_info->max_extent = (u64)-1;
 	fs_info->max_inline = 8192 * 1024;
-	if (setup_bdi(fs_info, &fs_info->bdi))
-		goto fail_bdi;
-	fs_info->btree_inode = new_inode(sb);
-	fs_info->btree_inode->i_ino = 1;
-	fs_info->btree_inode->i_nlink = 1;
 	fs_info->metadata_ratio = 8;
 
 	fs_info->thread_pool_size = min_t(unsigned long,
@@ -1602,6 +1642,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	sb->s_blocksize_bits = blksize_bits(4096);
 	sb->s_bdi = &fs_info->bdi;
 
+	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
+	fs_info->btree_inode->i_nlink = 1;
 	/*
 	 * we set the i_size on the btree inode to the max possible int.
 	 * the real end of the address space is determined by all of
@@ -1620,28 +1662,32 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 
 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
 
+	BTRFS_I(fs_info->btree_inode)->root = tree_root;
+	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
+	       sizeof(struct btrfs_key));
+	BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
+	insert_inode_hash(fs_info->btree_inode);
+
 	spin_lock_init(&fs_info->block_group_cache_lock);
 	fs_info->block_group_cache_tree.rb_node = NULL;
 
-	extent_io_tree_init(&fs_info->pinned_extents,
+	extent_io_tree_init(&fs_info->freed_extents[0],
 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
+	extent_io_tree_init(&fs_info->freed_extents[1],
+			     fs_info->btree_inode->i_mapping, GFP_NOFS);
+	fs_info->pinned_extents = &fs_info->freed_extents[0];
 	fs_info->do_barriers = 1;
 
-	BTRFS_I(fs_info->btree_inode)->root = tree_root;
-	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
-	       sizeof(struct btrfs_key));
-	insert_inode_hash(fs_info->btree_inode);
 
 	mutex_init(&fs_info->trans_mutex);
 	mutex_init(&fs_info->ordered_operations_mutex);
 	mutex_init(&fs_info->tree_log_mutex);
-	mutex_init(&fs_info->drop_mutex);
 	mutex_init(&fs_info->chunk_mutex);
 	mutex_init(&fs_info->transaction_kthread_mutex);
 	mutex_init(&fs_info->cleaner_mutex);
 	mutex_init(&fs_info->volume_mutex);
-	mutex_init(&fs_info->tree_reloc_mutex);
 	init_rwsem(&fs_info->extent_commit_sem);
+	init_rwsem(&fs_info->subvol_sem);
 
 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
@@ -1700,7 +1746,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 		err = -EINVAL;
 		goto fail_iput;
 	}
-
+printk("thread pool is %d\n", fs_info->thread_pool_size);
 	/*
 	 * we need to start all the end_io workers up front because the
 	 * queue work function gets called at interrupt time, and so it
@@ -1745,20 +1791,22 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	fs_info->endio_workers.idle_thresh = 4;
 	fs_info->endio_meta_workers.idle_thresh = 4;
 
-	fs_info->endio_write_workers.idle_thresh = 64;
-	fs_info->endio_meta_write_workers.idle_thresh = 64;
+	fs_info->endio_write_workers.idle_thresh = 2;
+	fs_info->endio_meta_write_workers.idle_thresh = 2;
+
+	fs_info->endio_workers.atomic_worker_start = 1;
+	fs_info->endio_meta_workers.atomic_worker_start = 1;
+	fs_info->endio_write_workers.atomic_worker_start = 1;
+	fs_info->endio_meta_write_workers.atomic_worker_start = 1;
 
 	btrfs_start_workers(&fs_info->workers, 1);
 	btrfs_start_workers(&fs_info->submit_workers, 1);
 	btrfs_start_workers(&fs_info->delalloc_workers, 1);
 	btrfs_start_workers(&fs_info->fixup_workers, 1);
-	btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
-	btrfs_start_workers(&fs_info->endio_meta_workers,
-			    fs_info->thread_pool_size);
-	btrfs_start_workers(&fs_info->endio_meta_write_workers,
-			    fs_info->thread_pool_size);
-	btrfs_start_workers(&fs_info->endio_write_workers,
-			    fs_info->thread_pool_size);
+	btrfs_start_workers(&fs_info->endio_workers, 1);
+	btrfs_start_workers(&fs_info->endio_meta_workers, 1);
+	btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
+	btrfs_start_workers(&fs_info->endio_write_workers, 1);
 
 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
@@ -1918,6 +1966,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 		}
 	}
 
+	ret = btrfs_find_orphan_roots(tree_root);
+	BUG_ON(ret);
+
 	if (!(sb->s_flags & MS_RDONLY)) {
 		ret = btrfs_recover_relocation(tree_root);
 		BUG_ON(ret);
@@ -1977,6 +2028,8 @@ fail_iput:
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
 fail_bdi:
 	bdi_destroy(&fs_info->bdi);
+fail_srcu:
+	cleanup_srcu_struct(&fs_info->subvol_srcu);
 fail:
 	kfree(extent_root);
 	kfree(tree_root);
@@ -2236,20 +2289,29 @@ int write_ctree_super(struct btrfs_trans_handle *trans,
 
 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 {
-	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
+	spin_lock(&fs_info->fs_roots_radix_lock);
 	radix_tree_delete(&fs_info->fs_roots_radix,
 			  (unsigned long)root->root_key.objectid);
+	spin_unlock(&fs_info->fs_roots_radix_lock);
+
+	if (btrfs_root_refs(&root->root_item) == 0)
+		synchronize_srcu(&fs_info->subvol_srcu);
+
+	free_fs_root(root);
+	return 0;
+}
+
+static void free_fs_root(struct btrfs_root *root)
+{
+	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
 	if (root->anon_super.s_dev) {
 		down_write(&root->anon_super.s_umount);
 		kill_anon_super(&root->anon_super);
 	}
-	if (root->node)
-		free_extent_buffer(root->node);
-	if (root->commit_root)
-		free_extent_buffer(root->commit_root);
+	free_extent_buffer(root->node);
+	free_extent_buffer(root->commit_root);
 	kfree(root->name);
 	kfree(root);
-	return 0;
 }
 
 static int del_fs_roots(struct btrfs_fs_info *fs_info)
@@ -2258,6 +2320,20 @@ static int del_fs_roots(struct btrfs_fs_info *fs_info)
 	struct btrfs_root *gang[8];
 	int i;
 
+	while (!list_empty(&fs_info->dead_roots)) {
+		gang[0] = list_entry(fs_info->dead_roots.next,
+				     struct btrfs_root, root_list);
+		list_del(&gang[0]->root_list);
+
+		if (gang[0]->in_radix) {
+			btrfs_free_fs_root(fs_info, gang[0]);
+		} else {
+			free_extent_buffer(gang[0]->node);
+			free_extent_buffer(gang[0]->commit_root);
+			kfree(gang[0]);
+		}
+	}
+
 	while (1) {
 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 					     (void **)gang, 0,
@@ -2287,9 +2363,6 @@ int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
 		for (i = 0; i < ret; i++) {
 			root_objectid = gang[i]->root_key.objectid;
-			ret = btrfs_find_dead_roots(fs_info->tree_root,
-						    root_objectid);
-			BUG_ON(ret);
 			btrfs_orphan_cleanup(gang[i]);
 		}
 		root_objectid++;
@@ -2359,7 +2432,6 @@ int close_ctree(struct btrfs_root *root)
 	free_extent_buffer(root->fs_info->csum_root->commit_root);
 
 	btrfs_free_block_groups(root->fs_info);
-	btrfs_free_pinned_extents(root->fs_info);
 
 	del_fs_roots(fs_info);
 
@@ -2378,6 +2450,7 @@ int close_ctree(struct btrfs_root *root)
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
 
 	bdi_destroy(&fs_info->bdi);
+	cleanup_srcu_struct(&fs_info->subvol_srcu);
 
 	kfree(fs_info->extent_root);
 	kfree(fs_info->tree_root);

fs/btrfs/export.c

@@ -28,7 +28,7 @@ static int btrfs_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 	len  = BTRFS_FID_SIZE_NON_CONNECTABLE;
 	type = FILEID_BTRFS_WITHOUT_PARENT;
 
-	fid->objectid = BTRFS_I(inode)->location.objectid;
+	fid->objectid = inode->i_ino;
 	fid->root_objectid = BTRFS_I(inode)->root->objectid;
 	fid->gen = inode->i_generation;
 
@@ -60,34 +60,61 @@ static int btrfs_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 }
 
 static struct dentry *btrfs_get_dentry(struct super_block *sb, u64 objectid,
-				       u64 root_objectid, u32 generation)
+				       u64 root_objectid, u32 generation,
+				       int check_generation)
 {
+	struct btrfs_fs_info *fs_info = btrfs_sb(sb)->fs_info;
 	struct btrfs_root *root;
+	struct dentry *dentry;
 	struct inode *inode;
 	struct btrfs_key key;
+	int index;
+	int err = 0;
+
+	if (objectid < BTRFS_FIRST_FREE_OBJECTID)
+		return ERR_PTR(-ESTALE);
 
 	key.objectid = root_objectid;
 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 	key.offset = (u64)-1;
 
-	root = btrfs_read_fs_root_no_name(btrfs_sb(sb)->fs_info, &key);
-	if (IS_ERR(root))
-		return ERR_CAST(root);
+	index = srcu_read_lock(&fs_info->subvol_srcu);
+
+	root = btrfs_read_fs_root_no_name(fs_info, &key);
+	if (IS_ERR(root)) {
+		err = PTR_ERR(root);
+		goto fail;
+	}
+
+	if (btrfs_root_refs(&root->root_item) == 0) {
+		err = -ENOENT;
+		goto fail;
+	}
 
 	key.objectid = objectid;
 	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 	key.offset = 0;
 
 	inode = btrfs_iget(sb, &key, root);
-	if (IS_ERR(inode))
-		return (void *)inode;
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
+		goto fail;
+	}
+
+	srcu_read_unlock(&fs_info->subvol_srcu, index);
 
-	if (generation != inode->i_generation) {
+	if (check_generation && generation != inode->i_generation) {
 		iput(inode);
 		return ERR_PTR(-ESTALE);
 	}
 
-	return d_obtain_alias(inode);
+	dentry = d_obtain_alias(inode);
+	if (!IS_ERR(dentry))
+		dentry->d_op = &btrfs_dentry_operations;
+	return dentry;
+fail:
+	srcu_read_unlock(&fs_info->subvol_srcu, index);
+	return ERR_PTR(err);
 }
 
 static struct dentry *btrfs_fh_to_parent(struct super_block *sb, struct fid *fh,
@@ -111,7 +138,7 @@ static struct dentry *btrfs_fh_to_parent(struct super_block *sb, struct fid *fh,
 	objectid = fid->parent_objectid;
 	generation = fid->parent_gen;
 
-	return btrfs_get_dentry(sb, objectid, root_objectid, generation);
+	return btrfs_get_dentry(sb, objectid, root_objectid, generation, 1);
 }
 
 static struct dentry *btrfs_fh_to_dentry(struct super_block *sb, struct fid *fh,
@@ -133,66 +160,76 @@ static struct dentry *btrfs_fh_to_dentry(struct super_block *sb, struct fid *fh,
 	root_objectid = fid->root_objectid;
 	generation = fid->gen;
 
-	return btrfs_get_dentry(sb, objectid, root_objectid, generation);
+	return btrfs_get_dentry(sb, objectid, root_objectid, generation, 1);
 }
 
 static struct dentry *btrfs_get_parent(struct dentry *child)
 {
 	struct inode *dir = child->d_inode;
+	static struct dentry *dentry;
 	struct btrfs_root *root = BTRFS_I(dir)->root;
-	struct btrfs_key key;
 	struct btrfs_path *path;
 	struct extent_buffer *leaf;
-	int slot;
-	u64 objectid;
+	struct btrfs_root_ref *ref;
+	struct btrfs_key key;
+	struct btrfs_key found_key;
 	int ret;
 
 	path = btrfs_alloc_path();
 
-	key.objectid = dir->i_ino;
-	btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
-	key.offset = (u64)-1;
+	if (dir->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
+		key.objectid = root->root_key.objectid;
+		key.type = BTRFS_ROOT_BACKREF_KEY;
+		key.offset = (u64)-1;
+		root = root->fs_info->tree_root;
+	} else {
+		key.objectid = dir->i_ino;
+		key.type = BTRFS_INODE_REF_KEY;
+		key.offset = (u64)-1;
+	}
 
 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
-	if (ret < 0) {
-		/* Error */
-		btrfs_free_path(path);
-		return ERR_PTR(ret);
+	if (ret < 0)
+		goto fail;
+
+	BUG_ON(ret == 0);
+	if (path->slots[0] == 0) {
+		ret = -ENOENT;
+		goto fail;
 	}
+
+	path->slots[0]--;
 	leaf = path->nodes[0];
-	slot = path->slots[0];
-	if (ret) {
-		/* btrfs_search_slot() returns the slot where we'd want to
-		   insert a backref for parent inode #0xFFFFFFFFFFFFFFFF.
-		   The _real_ backref, telling us what the parent inode
-		   _actually_ is, will be in the slot _before_ the one
-		   that btrfs_search_slot() returns. */
-		if (!slot) {
-			/* Unless there is _no_ key in the tree before... */
-			btrfs_free_path(path);
-			return ERR_PTR(-EIO);
-		}
-		slot--;
+
+	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+	if (found_key.objectid != key.objectid || found_key.type != key.type) {
+		ret = -ENOENT;
+		goto fail;
 	}
 
-	btrfs_item_key_to_cpu(leaf, &key, slot);
+	if (found_key.type == BTRFS_ROOT_BACKREF_KEY) {
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_root_ref);
+		key.objectid = btrfs_root_ref_dirid(leaf, ref);
+	} else {
+		key.objectid = found_key.offset;
+	}
 	btrfs_free_path(path);
 
-	if (key.objectid != dir->i_ino || key.type != BTRFS_INODE_REF_KEY)
-		return ERR_PTR(-EINVAL);
-
-	objectid = key.offset;
-
-	/* If we are already at the root of a subvol, return the real root */
-	if (objectid == dir->i_ino)
-		return dget(dir->i_sb->s_root);
+	if (found_key.type == BTRFS_ROOT_BACKREF_KEY) {
+		return btrfs_get_dentry(root->fs_info->sb, key.objectid,
+					found_key.offset, 0, 0);
+	}
 
-	/* Build a new key for the inode item */
-	key.objectid = objectid;
-	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
+	key.type = BTRFS_INODE_ITEM_KEY;
 	key.offset = 0;
-
-	return d_obtain_alias(btrfs_iget(root->fs_info->sb, &key, root));
+	dentry = d_obtain_alias(btrfs_iget(root->fs_info->sb, &key, root));
+	if (!IS_ERR(dentry))
+		dentry->d_op = &btrfs_dentry_operations;
+	return dentry;
+fail:
+	btrfs_free_path(path);
+	return ERR_PTR(ret);
 }
 
 const struct export_operations btrfs_export_ops = {

fs/btrfs/extent-tree.c

@@ -32,12 +32,12 @@
 #include "locking.h"
 #include "free-space-cache.h"
 
-static int update_reserved_extents(struct btrfs_root *root,
-				   u64 bytenr, u64 num, int reserve);
 static int update_block_group(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
 			      u64 bytenr, u64 num_bytes, int alloc,
 			      int mark_free);
+static int update_reserved_extents(struct btrfs_block_group_cache *cache,
+				   u64 num_bytes, int reserve);
 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root,
 				u64 bytenr, u64 num_bytes, u64 parent,
@@ -57,10 +57,17 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 				     u64 parent, u64 root_objectid,
 				     u64 flags, struct btrfs_disk_key *key,
 				     int level, struct btrfs_key *ins);
-
 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *extent_root, u64 alloc_bytes,
 			  u64 flags, int force);
+static int pin_down_bytes(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  struct btrfs_path *path,
+			  u64 bytenr, u64 num_bytes,
+			  int is_data, int reserved,
+			  struct extent_buffer **must_clean);
+static int find_next_key(struct btrfs_path *path, int level,
+			 struct btrfs_key *key);
 
 static noinline int
 block_group_cache_done(struct btrfs_block_group_cache *cache)
@@ -153,34 +160,34 @@ block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 	return ret;
 }
 
-/*
- * We always set EXTENT_LOCKED for the super mirror extents so we don't
- * overwrite them, so those bits need to be unset.  Also, if we are unmounting
- * with pinned extents still sitting there because we had a block group caching,
- * we need to clear those now, since we are done.
- */
-void btrfs_free_pinned_extents(struct btrfs_fs_info *info)
+static int add_excluded_extent(struct btrfs_root *root,
+			       u64 start, u64 num_bytes)
 {
-	u64 start, end, last = 0;
-	int ret;
+	u64 end = start + num_bytes - 1;
+	set_extent_bits(&root->fs_info->freed_extents[0],
+			start, end, EXTENT_UPTODATE, GFP_NOFS);
+	set_extent_bits(&root->fs_info->freed_extents[1],
+			start, end, EXTENT_UPTODATE, GFP_NOFS);
+	return 0;
+}
 
-	while (1) {
-		ret = find_first_extent_bit(&info->pinned_extents, last,
-					    &start, &end,
-					    EXTENT_LOCKED|EXTENT_DIRTY);
-		if (ret)
-			break;
+static void free_excluded_extents(struct btrfs_root *root,
+				  struct btrfs_block_group_cache *cache)
+{
+	u64 start, end;
 
-		clear_extent_bits(&info->pinned_extents, start, end,
-				  EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS);
-		last = end+1;
-	}
+	start = cache->key.objectid;
+	end = start + cache->key.offset - 1;
+
+	clear_extent_bits(&root->fs_info->freed_extents[0],
+			  start, end, EXTENT_UPTODATE, GFP_NOFS);
+	clear_extent_bits(&root->fs_info->freed_extents[1],
+			  start, end, EXTENT_UPTODATE, GFP_NOFS);
 }
 
-static int remove_sb_from_cache(struct btrfs_root *root,
-				struct btrfs_block_group_cache *cache)
+static int exclude_super_stripes(struct btrfs_root *root,
+				 struct btrfs_block_group_cache *cache)
 {
-	struct btrfs_fs_info *fs_info = root->fs_info;
 	u64 bytenr;
 	u64 *logical;
 	int stripe_len;
@@ -192,17 +199,42 @@ static int remove_sb_from_cache(struct btrfs_root *root,
 				       cache->key.objectid, bytenr,
 				       0, &logical, &nr, &stripe_len);
 		BUG_ON(ret);
+
 		while (nr--) {
-			try_lock_extent(&fs_info->pinned_extents,
-					logical[nr],
-					logical[nr] + stripe_len - 1, GFP_NOFS);
+			cache->bytes_super += stripe_len;
+			ret = add_excluded_extent(root, logical[nr],
+						  stripe_len);
+			BUG_ON(ret);
 		}
+
 		kfree(logical);
 	}
-
 	return 0;
 }
 
+static struct btrfs_caching_control *
+get_caching_control(struct btrfs_block_group_cache *cache)
+{
+	struct btrfs_caching_control *ctl;
+
+	spin_lock(&cache->lock);
+	if (cache->cached != BTRFS_CACHE_STARTED) {
+		spin_unlock(&cache->lock);
+		return NULL;
+	}
+
+	ctl = cache->caching_ctl;
+	atomic_inc(&ctl->count);
+	spin_unlock(&cache->lock);
+	return ctl;
+}
+
+static void put_caching_control(struct btrfs_caching_control *ctl)
+{
+	if (atomic_dec_and_test(&ctl->count))
+		kfree(ctl);
+}
+
 /*
  * this is only called by cache_block_group, since we could have freed extents
  * we need to check the pinned_extents for any extents that can't be used yet
@@ -215,9 +247,9 @@ static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 	int ret;
 
 	while (start < end) {
-		ret = find_first_extent_bit(&info->pinned_extents, start,
+		ret = find_first_extent_bit(info->pinned_extents, start,
 					    &extent_start, &extent_end,
-					    EXTENT_DIRTY|EXTENT_LOCKED);
+					    EXTENT_DIRTY | EXTENT_UPTODATE);
 		if (ret)
 			break;
 
@@ -249,22 +281,27 @@ static int caching_kthread(void *data)
 {
 	struct btrfs_block_group_cache *block_group = data;
 	struct btrfs_fs_info *fs_info = block_group->fs_info;
-	u64 last = 0;
+	struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
+	struct btrfs_root *extent_root = fs_info->extent_root;
 	struct btrfs_path *path;
-	int ret = 0;
-	struct btrfs_key key;
 	struct extent_buffer *leaf;
-	int slot;
+	struct btrfs_key key;
 	u64 total_found = 0;
-
-	BUG_ON(!fs_info);
+	u64 last = 0;
+	u32 nritems;
+	int ret = 0;
 
 	path = btrfs_alloc_path();
 	if (!path)
 		return -ENOMEM;
 
-	atomic_inc(&block_group->space_info->caching_threads);
+	exclude_super_stripes(extent_root, block_group);
+	spin_lock(&block_group->space_info->lock);
+	block_group->space_info->bytes_super += block_group->bytes_super;
+	spin_unlock(&block_group->space_info->lock);
+
 	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
+
 	/*
 	 * We don't want to deadlock with somebody trying to allocate a new
 	 * extent for the extent root while also trying to search the extent
@@ -277,74 +314,64 @@ static int caching_kthread(void *data)
 
 	key.objectid = last;
 	key.offset = 0;
-	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+	key.type = BTRFS_EXTENT_ITEM_KEY;
 again:
+	mutex_lock(&caching_ctl->mutex);
 	/* need to make sure the commit_root doesn't disappear */
 	down_read(&fs_info->extent_commit_sem);
 
-	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
+	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 	if (ret < 0)
 		goto err;
 
+	leaf = path->nodes[0];
+	nritems = btrfs_header_nritems(leaf);
+
 	while (1) {
 		smp_mb();
-		if (block_group->fs_info->closing > 1) {
+		if (fs_info->closing > 1) {
 			last = (u64)-1;
 			break;
 		}
 
-		leaf = path->nodes[0];
-		slot = path->slots[0];
-		if (slot >= btrfs_header_nritems(leaf)) {
-			ret = btrfs_next_leaf(fs_info->extent_root, path);
-			if (ret < 0)
-				goto err;
-			else if (ret)
+		if (path->slots[0] < nritems) {
+			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		} else {
+			ret = find_next_key(path, 0, &key);
+			if (ret)
 				break;
 
-			if (need_resched() ||
-			    btrfs_transaction_in_commit(fs_info)) {
-				leaf = path->nodes[0];
-
-				/* this shouldn't happen, but if the
-				 * leaf is empty just move on.
-				 */
-				if (btrfs_header_nritems(leaf) == 0)
-					break;
-				/*
-				 * we need to copy the key out so that
-				 * we are sure the next search advances
-				 * us forward in the btree.
-				 */
-				btrfs_item_key_to_cpu(leaf, &key, 0);
-				btrfs_release_path(fs_info->extent_root, path);
-				up_read(&fs_info->extent_commit_sem);
+			caching_ctl->progress = last;
+			btrfs_release_path(extent_root, path);
+			up_read(&fs_info->extent_commit_sem);
+			mutex_unlock(&caching_ctl->mutex);
+			if (btrfs_transaction_in_commit(fs_info))
 				schedule_timeout(1);
-				goto again;
-			}
+			else
+				cond_resched();
+			goto again;
+		}
 
+		if (key.objectid < block_group->key.objectid) {
+			path->slots[0]++;
 			continue;
 		}
-		btrfs_item_key_to_cpu(leaf, &key, slot);
-		if (key.objectid < block_group->key.objectid)
-			goto next;
 
 		if (key.objectid >= block_group->key.objectid +
 		    block_group->key.offset)
 			break;
 
-		if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
+		if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 			total_found += add_new_free_space(block_group,
 							  fs_info, last,
 							  key.objectid);
 			last = key.objectid + key.offset;
-		}
 
-		if (total_found > (1024 * 1024 * 2)) {
-			total_found = 0;
-			wake_up(&block_group->caching_q);
+			if (total_found > (1024 * 1024 * 2)) {
+				total_found = 0;
+				wake_up(&caching_ctl->wait);
+			}
 		}
-next:
 		path->slots[0]++;
 	}
 	ret = 0;
@@ -352,33 +379,65 @@ next:
 	total_found += add_new_free_space(block_group, fs_info, last,
 					  block_group->key.objectid +
 					  block_group->key.offset);
+	caching_ctl->progress = (u64)-1;
 
 	spin_lock(&block_group->lock);
+	block_group->caching_ctl = NULL;
 	block_group->cached = BTRFS_CACHE_FINISHED;
 	spin_unlock(&block_group->lock);
 
 err:
 	btrfs_free_path(path);
 	up_read(&fs_info->extent_commit_sem);
-	atomic_dec(&block_group->space_info->caching_threads);
-	wake_up(&block_group->caching_q);
 
+	free_excluded_extents(extent_root, block_group);
+
+	mutex_unlock(&caching_ctl->mutex);
+	wake_up(&caching_ctl->wait);
+
+	put_caching_control(caching_ctl);
+	atomic_dec(&block_group->space_info->caching_threads);
 	return 0;
 }
 
 static int cache_block_group(struct btrfs_block_group_cache *cache)
 {
+	struct btrfs_fs_info *fs_info = cache->fs_info;
+	struct btrfs_caching_control *caching_ctl;
 	struct task_struct *tsk;
 	int ret = 0;
 
+	smp_mb();
+	if (cache->cached != BTRFS_CACHE_NO)
+		return 0;
+
+	caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
+	BUG_ON(!caching_ctl);
+
+	INIT_LIST_HEAD(&caching_ctl->list);
+	mutex_init(&caching_ctl->mutex);
+	init_waitqueue_head(&caching_ctl->wait);
+	caching_ctl->block_group = cache;
+	caching_ctl->progress = cache->key.objectid;
+	/* one for caching kthread, one for caching block group list */
+	atomic_set(&caching_ctl->count, 2);
+
 	spin_lock(&cache->lock);
 	if (cache->cached != BTRFS_CACHE_NO) {
 		spin_unlock(&cache->lock);
-		return ret;
+		kfree(caching_ctl);
+		return 0;
 	}
+	cache->caching_ctl = caching_ctl;
 	cache->cached = BTRFS_CACHE_STARTED;
 	spin_unlock(&cache->lock);
 
+	down_write(&fs_info->extent_commit_sem);
+	list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
+	up_write(&fs_info->extent_commit_sem);
+
+	atomic_inc(&cache->space_info->caching_threads);
+
 	tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 			  cache->key.objectid);
 	if (IS_ERR(tsk)) {
@@ -1657,7 +1716,6 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 						 parent, ref_root, flags,
 						 ref->objectid, ref->offset,
 						 &ins, node->ref_mod);
-		update_reserved_extents(root, ins.objectid, ins.offset, 0);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
 		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
 					     node->num_bytes, parent,
@@ -1783,7 +1841,6 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 						extent_op->flags_to_set,
 						&extent_op->key,
 						ref->level, &ins);
-		update_reserved_extents(root, ins.objectid, ins.offset, 0);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
 		ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
 					     node->num_bytes, parent, ref_root,
@@ -1818,16 +1875,32 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 		BUG_ON(extent_op);
 		head = btrfs_delayed_node_to_head(node);
 		if (insert_reserved) {
+			int mark_free = 0;
+			struct extent_buffer *must_clean = NULL;
+
+			ret = pin_down_bytes(trans, root, NULL,
+					     node->bytenr, node->num_bytes,
+					     head->is_data, 1, &must_clean);
+			if (ret > 0)
+				mark_free = 1;
+
+			if (must_clean) {
+				clean_tree_block(NULL, root, must_clean);
+				btrfs_tree_unlock(must_clean);
+				free_extent_buffer(must_clean);
+			}
 			if (head->is_data) {
 				ret = btrfs_del_csums(trans, root,
 						      node->bytenr,
 						      node->num_bytes);
 				BUG_ON(ret);
 			}
-			btrfs_update_pinned_extents(root, node->bytenr,
-						    node->num_bytes, 1);
-			update_reserved_extents(root, node->bytenr,
-						node->num_bytes, 0);
+			if (mark_free) {
+				ret = btrfs_free_reserved_extent(root,
+							node->bytenr,
+							node->num_bytes);
+				BUG_ON(ret);
+			}
 		}
 		mutex_unlock(&head->mutex);
 		return 0;
@@ -2706,6 +2779,8 @@ int btrfs_check_metadata_free_space(struct btrfs_root *root)
 	/* get the space info for where the metadata will live */
 	alloc_target = btrfs_get_alloc_profile(root, 0);
 	meta_sinfo = __find_space_info(info, alloc_target);
+	if (!meta_sinfo)
+		goto alloc;
 
 again:
 	spin_lock(&meta_sinfo->lock);
@@ -2717,12 +2792,13 @@ again:
 	do_div(thresh, 100);
 
 	if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
-	    meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
+	    meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+	    meta_sinfo->bytes_super > thresh) {
 		struct btrfs_trans_handle *trans;
 		if (!meta_sinfo->full) {
 			meta_sinfo->force_alloc = 1;
 			spin_unlock(&meta_sinfo->lock);
-
+alloc:
 			trans = btrfs_start_transaction(root, 1);
 			if (!trans)
 				return -ENOMEM;
@@ -2730,6 +2806,10 @@ again:
 			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
 					     2 * 1024 * 1024, alloc_target, 0);
 			btrfs_end_transaction(trans, root);
+			if (!meta_sinfo) {
+				meta_sinfo = __find_space_info(info,
+							       alloc_target);
+			}
 			goto again;
 		}
 		spin_unlock(&meta_sinfo->lock);
@@ -2765,13 +2845,16 @@ int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
 	bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 
 	data_sinfo = BTRFS_I(inode)->space_info;
+	if (!data_sinfo)
+		goto alloc;
+
 again:
 	/* make sure we have enough space to handle the data first */
 	spin_lock(&data_sinfo->lock);
 	if (data_sinfo->total_bytes - data_sinfo->bytes_used -
 	    data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
 	    data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
-	    data_sinfo->bytes_may_use < bytes) {
+	    data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
 		struct btrfs_trans_handle *trans;
 
 		/*
@@ -2783,7 +2866,7 @@ again:
 
 			data_sinfo->force_alloc = 1;
 			spin_unlock(&data_sinfo->lock);
-
+alloc:
 			alloc_target = btrfs_get_alloc_profile(root, 1);
 			trans = btrfs_start_transaction(root, 1);
 			if (!trans)
@@ -2795,6 +2878,11 @@ again:
 			btrfs_end_transaction(trans, root);
 			if (ret)
 				return ret;
+
+			if (!data_sinfo) {
+				btrfs_set_inode_space_info(root, inode);
+				data_sinfo = BTRFS_I(inode)->space_info;
+			}
 			goto again;
 		}
 		spin_unlock(&data_sinfo->lock);
@@ -3009,10 +3097,12 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 		num_bytes = min(total, cache->key.offset - byte_in_group);
 		if (alloc) {
 			old_val += num_bytes;
+			btrfs_set_block_group_used(&cache->item, old_val);
+			cache->reserved -= num_bytes;
 			cache->space_info->bytes_used += num_bytes;
+			cache->space_info->bytes_reserved -= num_bytes;
 			if (cache->ro)
 				cache->space_info->bytes_readonly -= num_bytes;
-			btrfs_set_block_group_used(&cache->item, old_val);
 			spin_unlock(&cache->lock);
 			spin_unlock(&cache->space_info->lock);
 		} else {
@@ -3057,127 +3147,136 @@ static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
 	return bytenr;
 }
 
-int btrfs_update_pinned_extents(struct btrfs_root *root,
-				u64 bytenr, u64 num, int pin)
+/*
+ * this function must be called within transaction
+ */
+int btrfs_pin_extent(struct btrfs_root *root,
+		     u64 bytenr, u64 num_bytes, int reserved)
 {
-	u64 len;
-	struct btrfs_block_group_cache *cache;
 	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_block_group_cache *cache;
 
-	if (pin)
-		set_extent_dirty(&fs_info->pinned_extents,
-				bytenr, bytenr + num - 1, GFP_NOFS);
-
-	while (num > 0) {
-		cache = btrfs_lookup_block_group(fs_info, bytenr);
-		BUG_ON(!cache);
-		len = min(num, cache->key.offset -
-			  (bytenr - cache->key.objectid));
-		if (pin) {
-			spin_lock(&cache->space_info->lock);
-			spin_lock(&cache->lock);
-			cache->pinned += len;
-			cache->space_info->bytes_pinned += len;
-			spin_unlock(&cache->lock);
-			spin_unlock(&cache->space_info->lock);
-			fs_info->total_pinned += len;
-		} else {
-			int unpin = 0;
+	cache = btrfs_lookup_block_group(fs_info, bytenr);
+	BUG_ON(!cache);
 
-			/*
-			 * in order to not race with the block group caching, we
-			 * only want to unpin the extent if we are cached.  If
-			 * we aren't cached, we want to start async caching this
-			 * block group so we can free the extent the next time
-			 * around.
-			 */
-			spin_lock(&cache->space_info->lock);
-			spin_lock(&cache->lock);
-			unpin = (cache->cached == BTRFS_CACHE_FINISHED);
-			if (likely(unpin)) {
-				cache->pinned -= len;
-				cache->space_info->bytes_pinned -= len;
-				fs_info->total_pinned -= len;
-			}
-			spin_unlock(&cache->lock);
-			spin_unlock(&cache->space_info->lock);
+	spin_lock(&cache->space_info->lock);
+	spin_lock(&cache->lock);
+	cache->pinned += num_bytes;
+	cache->space_info->bytes_pinned += num_bytes;
+	if (reserved) {
+		cache->reserved -= num_bytes;
+		cache->space_info->bytes_reserved -= num_bytes;
+	}
+	spin_unlock(&cache->lock);
+	spin_unlock(&cache->space_info->lock);
 
-			if (likely(unpin))
-				clear_extent_dirty(&fs_info->pinned_extents,
-						   bytenr, bytenr + len -1,
-						   GFP_NOFS);
-			else
-				cache_block_group(cache);
+	btrfs_put_block_group(cache);
 
-			if (unpin)
-				btrfs_add_free_space(cache, bytenr, len);
-		}
-		btrfs_put_block_group(cache);
-		bytenr += len;
-		num -= len;
+	set_extent_dirty(fs_info->pinned_extents,
+			 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
+	return 0;
+}
+
+static int update_reserved_extents(struct btrfs_block_group_cache *cache,
+				   u64 num_bytes, int reserve)
+{
+	spin_lock(&cache->space_info->lock);
+	spin_lock(&cache->lock);
+	if (reserve) {
+		cache->reserved += num_bytes;
+		cache->space_info->bytes_reserved += num_bytes;
+	} else {
+		cache->reserved -= num_bytes;
+		cache->space_info->bytes_reserved -= num_bytes;
 	}
+	spin_unlock(&cache->lock);
+	spin_unlock(&cache->space_info->lock);
 	return 0;
 }
 
-static int update_reserved_extents(struct btrfs_root *root,
-				   u64 bytenr, u64 num, int reserve)
+int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root)
 {
-	u64 len;
-	struct btrfs_block_group_cache *cache;
 	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_caching_control *next;
+	struct btrfs_caching_control *caching_ctl;
+	struct btrfs_block_group_cache *cache;
 
-	while (num > 0) {
-		cache = btrfs_lookup_block_group(fs_info, bytenr);
-		BUG_ON(!cache);
-		len = min(num, cache->key.offset -
-			  (bytenr - cache->key.objectid));
+	down_write(&fs_info->extent_commit_sem);
 
-		spin_lock(&cache->space_info->lock);
-		spin_lock(&cache->lock);
-		if (reserve) {
-			cache->reserved += len;
-			cache->space_info->bytes_reserved += len;
+	list_for_each_entry_safe(caching_ctl, next,
+				 &fs_info->caching_block_groups, list) {
+		cache = caching_ctl->block_group;
+		if (block_group_cache_done(cache)) {
+			cache->last_byte_to_unpin = (u64)-1;
+			list_del_init(&caching_ctl->list);
+			put_caching_control(caching_ctl);
 		} else {
-			cache->reserved -= len;
-			cache->space_info->bytes_reserved -= len;
+			cache->last_byte_to_unpin = caching_ctl->progress;
 		}
-		spin_unlock(&cache->lock);
-		spin_unlock(&cache->space_info->lock);
-		btrfs_put_block_group(cache);
-		bytenr += len;
-		num -= len;
 	}
+
+	if (fs_info->pinned_extents == &fs_info->freed_extents[0])
+		fs_info->pinned_extents = &fs_info->freed_extents[1];
+	else
+		fs_info->pinned_extents = &fs_info->freed_extents[0];
+
+	up_write(&fs_info->extent_commit_sem);
 	return 0;
 }
 
-int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
+static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
 {
-	u64 last = 0;
-	u64 start;
-	u64 end;
-	struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
-	int ret;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_block_group_cache *cache = NULL;
+	u64 len;
 
-	while (1) {
-		ret = find_first_extent_bit(pinned_extents, last,
-					    &start, &end, EXTENT_DIRTY);
-		if (ret)
-			break;
+	while (start <= end) {
+		if (!cache ||
+		    start >= cache->key.objectid + cache->key.offset) {
+			if (cache)
+				btrfs_put_block_group(cache);
+			cache = btrfs_lookup_block_group(fs_info, start);
+			BUG_ON(!cache);
+		}
+
+		len = cache->key.objectid + cache->key.offset - start;
+		len = min(len, end + 1 - start);
+
+		if (start < cache->last_byte_to_unpin) {
+			len = min(len, cache->last_byte_to_unpin - start);
+			btrfs_add_free_space(cache, start, len);
+		}
+
+		spin_lock(&cache->space_info->lock);
+		spin_lock(&cache->lock);
+		cache->pinned -= len;
+		cache->space_info->bytes_pinned -= len;
+		spin_unlock(&cache->lock);
+		spin_unlock(&cache->space_info->lock);
 
-		set_extent_dirty(copy, start, end, GFP_NOFS);
-		last = end + 1;
+		start += len;
 	}
+
+	if (cache)
+		btrfs_put_block_group(cache);
 	return 0;
 }
 
 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
-			       struct btrfs_root *root,
-			       struct extent_io_tree *unpin)
+			       struct btrfs_root *root)
 {
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct extent_io_tree *unpin;
 	u64 start;
 	u64 end;
 	int ret;
 
+	if (fs_info->pinned_extents == &fs_info->freed_extents[0])
+		unpin = &fs_info->freed_extents[1];
+	else
+		unpin = &fs_info->freed_extents[0];
+
 	while (1) {
 		ret = find_first_extent_bit(unpin, 0, &start, &end,
 					    EXTENT_DIRTY);
@@ -3186,10 +3285,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 
 		ret = btrfs_discard_extent(root, start, end + 1 - start);
 
-		/* unlocks the pinned mutex */
-		btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
-
+		unpin_extent_range(root, start, end);
 		cond_resched();
 	}
 
@@ -3199,7 +3296,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 static int pin_down_bytes(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *root,
 			  struct btrfs_path *path,
-			  u64 bytenr, u64 num_bytes, int is_data,
+			  u64 bytenr, u64 num_bytes,
+			  int is_data, int reserved,
 			  struct extent_buffer **must_clean)
 {
 	int err = 0;
@@ -3231,15 +3329,15 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
 	}
 	free_extent_buffer(buf);
 pinit:
-	btrfs_set_path_blocking(path);
+	if (path)
+		btrfs_set_path_blocking(path);
 	/* unlocks the pinned mutex */
-	btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
+	btrfs_pin_extent(root, bytenr, num_bytes, reserved);
 
 	BUG_ON(err < 0);
 	return 0;
 }
 
-
 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root,
 				u64 bytenr, u64 num_bytes, u64 parent,
@@ -3413,7 +3511,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 		}
 
 		ret = pin_down_bytes(trans, root, path, bytenr,
-				     num_bytes, is_data, &must_clean);
+				     num_bytes, is_data, 0, &must_clean);
 		if (ret > 0)
 			mark_free = 1;
 		BUG_ON(ret < 0);
@@ -3544,8 +3642,7 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
 	if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
 		WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
 		/* unlocks the pinned mutex */
-		btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
-		update_reserved_extents(root, bytenr, num_bytes, 0);
+		btrfs_pin_extent(root, bytenr, num_bytes, 1);
 		ret = 0;
 	} else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 		ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
@@ -3585,19 +3682,33 @@ static noinline int
 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
 				u64 num_bytes)
 {
+	struct btrfs_caching_control *caching_ctl;
 	DEFINE_WAIT(wait);
 
-	prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE);
-
-	if (block_group_cache_done(cache)) {
-		finish_wait(&cache->caching_q, &wait);
+	caching_ctl = get_caching_control(cache);
+	if (!caching_ctl)
 		return 0;
-	}
-	schedule();
-	finish_wait(&cache->caching_q, &wait);
 
-	wait_event(cache->caching_q, block_group_cache_done(cache) ||
+	wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
 		   (cache->free_space >= num_bytes));
+
+	put_caching_control(caching_ctl);
+	return 0;
+}
+
+static noinline int
+wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
+{
+	struct btrfs_caching_control *caching_ctl;
+	DEFINE_WAIT(wait);
+
+	caching_ctl = get_caching_control(cache);
+	if (!caching_ctl)
+		return 0;
+
+	wait_event(caching_ctl->wait, block_group_cache_done(cache));
+
+	put_caching_control(caching_ctl);
 	return 0;
 }
 
@@ -3635,6 +3746,7 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
 	int last_ptr_loop = 0;
 	int loop = 0;
 	bool found_uncached_bg = false;
+	bool failed_cluster_refill = false;
 
 	WARN_ON(num_bytes < root->sectorsize);
 	btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
@@ -3732,7 +3844,16 @@ have_block_group:
 		if (unlikely(block_group->ro))
 			goto loop;
 
-		if (last_ptr) {
+		/*
+		 * Ok we want to try and use the cluster allocator, so lets look
+		 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
+		 * have tried the cluster allocator plenty of times at this
+		 * point and not have found anything, so we are likely way too
+		 * fragmented for the clustering stuff to find anything, so lets
+		 * just skip it and let the allocator find whatever block it can
+		 * find
+		 */
+		if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
 			/*
 			 * the refill lock keeps out other
 			 * people trying to start a new cluster
@@ -3807,9 +3928,11 @@ refill_cluster:
 					spin_unlock(&last_ptr->refill_lock);
 					goto checks;
 				}
-			} else if (!cached && loop > LOOP_CACHING_NOWAIT) {
+			} else if (!cached && loop > LOOP_CACHING_NOWAIT
+				   && !failed_cluster_refill) {
 				spin_unlock(&last_ptr->refill_lock);
 
+				failed_cluster_refill = true;
 				wait_block_group_cache_progress(block_group,
 				       num_bytes + empty_cluster + empty_size);
 				goto have_block_group;
@@ -3821,13 +3944,9 @@ refill_cluster:
 			 * cluster.  Free the cluster we've been trying
 			 * to use, and go to the next block group
 			 */
-			if (loop < LOOP_NO_EMPTY_SIZE) {
-				btrfs_return_cluster_to_free_space(NULL,
-								   last_ptr);
-				spin_unlock(&last_ptr->refill_lock);
-				goto loop;
-			}
+			btrfs_return_cluster_to_free_space(NULL, last_ptr);
 			spin_unlock(&last_ptr->refill_lock);
+			goto loop;
 		}
 
 		offset = btrfs_find_space_for_alloc(block_group, search_start,
@@ -3881,9 +4000,12 @@ checks:
 					     search_start - offset);
 		BUG_ON(offset > search_start);
 
+		update_reserved_extents(block_group, num_bytes, 1);
+
 		/* we are all good, lets return */
 		break;
 loop:
+		failed_cluster_refill = false;
 		btrfs_put_block_group(block_group);
 	}
 	up_read(&space_info->groups_sem);
@@ -3973,12 +4095,12 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
 	up_read(&info->groups_sem);
 }
 
-static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
-				  struct btrfs_root *root,
-				  u64 num_bytes, u64 min_alloc_size,
-				  u64 empty_size, u64 hint_byte,
-				  u64 search_end, struct btrfs_key *ins,
-				  u64 data)
+int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+			 struct btrfs_root *root,
+			 u64 num_bytes, u64 min_alloc_size,
+			 u64 empty_size, u64 hint_byte,
+			 u64 search_end, struct btrfs_key *ins,
+			 u64 data)
 {
 	int ret;
 	u64 search_start = 0;
@@ -4044,25 +4166,8 @@ int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
 	ret = btrfs_discard_extent(root, start, len);
 
 	btrfs_add_free_space(cache, start, len);
+	update_reserved_extents(cache, len, 0);
 	btrfs_put_block_group(cache);
-	update_reserved_extents(root, start, len, 0);
-
-	return ret;
-}
-
-int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
-				  struct btrfs_root *root,
-				  u64 num_bytes, u64 min_alloc_size,
-				  u64 empty_size, u64 hint_byte,
-				  u64 search_end, struct btrfs_key *ins,
-				  u64 data)
-{
-	int ret;
-	ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
-				     empty_size, hint_byte, search_end, ins,
-				     data);
-	if (!ret)
-		update_reserved_extents(root, ins->objectid, ins->offset, 1);
 
 	return ret;
 }
@@ -4223,15 +4328,46 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 {
 	int ret;
 	struct btrfs_block_group_cache *block_group;
+	struct btrfs_caching_control *caching_ctl;
+	u64 start = ins->objectid;
+	u64 num_bytes = ins->offset;
 
 	block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
 	cache_block_group(block_group);
-	wait_event(block_group->caching_q,
-		   block_group_cache_done(block_group));
+	caching_ctl = get_caching_control(block_group);
 
-	ret = btrfs_remove_free_space(block_group, ins->objectid,
-				      ins->offset);
-	BUG_ON(ret);
+	if (!caching_ctl) {
+		BUG_ON(!block_group_cache_done(block_group));
+		ret = btrfs_remove_free_space(block_group, start, num_bytes);
+		BUG_ON(ret);
+	} else {
+		mutex_lock(&caching_ctl->mutex);
+
+		if (start >= caching_ctl->progress) {
+			ret = add_excluded_extent(root, start, num_bytes);
+			BUG_ON(ret);
+		} else if (start + num_bytes <= caching_ctl->progress) {
+			ret = btrfs_remove_free_space(block_group,
+						      start, num_bytes);
+			BUG_ON(ret);
+		} else {
+			num_bytes = caching_ctl->progress - start;
+			ret = btrfs_remove_free_space(block_group,
+						      start, num_bytes);
+			BUG_ON(ret);
+
+			start = caching_ctl->progress;
+			num_bytes = ins->objectid + ins->offset -
+				    caching_ctl->progress;
+			ret = add_excluded_extent(root, start, num_bytes);
+			BUG_ON(ret);
+		}
+
+		mutex_unlock(&caching_ctl->mutex);
+		put_caching_control(caching_ctl);
+	}
+
+	update_reserved_extents(block_group, ins->offset, 1);
 	btrfs_put_block_group(block_group);
 	ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
 					 0, owner, offset, ins, 1);
@@ -4255,9 +4391,9 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
 	int ret;
 	u64 flags = 0;
 
-	ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
-				     empty_size, hint_byte, search_end,
-				     ins, 0);
+	ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
+				   empty_size, hint_byte, search_end,
+				   ins, 0);
 	if (ret)
 		return ret;
 
@@ -4268,7 +4404,6 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
 	} else
 		BUG_ON(parent > 0);
 
-	update_reserved_extents(root, ins->objectid, ins->offset, 1);
 	if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
 		struct btrfs_delayed_extent_op *extent_op;
 		extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
@@ -4347,452 +4482,99 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 	return buf;
 }
 
-#if 0
-int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
-			struct btrfs_root *root, struct extent_buffer *leaf)
+struct walk_control {
+	u64 refs[BTRFS_MAX_LEVEL];
+	u64 flags[BTRFS_MAX_LEVEL];
+	struct btrfs_key update_progress;
+	int stage;
+	int level;
+	int shared_level;
+	int update_ref;
+	int keep_locks;
+	int reada_slot;
+	int reada_count;
+};
+
+#define DROP_REFERENCE	1
+#define UPDATE_BACKREF	2
+
+static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
+				     struct btrfs_root *root,
+				     struct walk_control *wc,
+				     struct btrfs_path *path)
 {
-	u64 disk_bytenr;
-	u64 num_bytes;
-	struct btrfs_key key;
-	struct btrfs_file_extent_item *fi;
+	u64 bytenr;
+	u64 generation;
+	u64 refs;
+	u64 last = 0;
 	u32 nritems;
-	int i;
+	u32 blocksize;
+	struct btrfs_key key;
+	struct extent_buffer *eb;
 	int ret;
+	int slot;
+	int nread = 0;
 
-	BUG_ON(!btrfs_is_leaf(leaf));
-	nritems = btrfs_header_nritems(leaf);
+	if (path->slots[wc->level] < wc->reada_slot) {
+		wc->reada_count = wc->reada_count * 2 / 3;
+		wc->reada_count = max(wc->reada_count, 2);
+	} else {
+		wc->reada_count = wc->reada_count * 3 / 2;
+		wc->reada_count = min_t(int, wc->reada_count,
+					BTRFS_NODEPTRS_PER_BLOCK(root));
+	}
 
-	for (i = 0; i < nritems; i++) {
-		cond_resched();
-		btrfs_item_key_to_cpu(leaf, &key, i);
+	eb = path->nodes[wc->level];
+	nritems = btrfs_header_nritems(eb);
+	blocksize = btrfs_level_size(root, wc->level - 1);
 
-		/* only extents have references, skip everything else */
-		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
-			continue;
-
-		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
-
-		/* inline extents live in the btree, they don't have refs */
-		if (btrfs_file_extent_type(leaf, fi) ==
-		    BTRFS_FILE_EXTENT_INLINE)
-			continue;
-
-		disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
-
-		/* holes don't have refs */
-		if (disk_bytenr == 0)
-			continue;
-
-		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
-		ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
-					leaf->start, 0, key.objectid, 0);
-		BUG_ON(ret);
-	}
-	return 0;
-}
-
-static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
-					struct btrfs_root *root,
-					struct btrfs_leaf_ref *ref)
-{
-	int i;
-	int ret;
-	struct btrfs_extent_info *info;
-	struct refsort *sorted;
-
-	if (ref->nritems == 0)
-		return 0;
-
-	sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
-	for (i = 0; i < ref->nritems; i++) {
-		sorted[i].bytenr = ref->extents[i].bytenr;
-		sorted[i].slot = i;
-	}
-	sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
+	for (slot = path->slots[wc->level]; slot < nritems; slot++) {
+		if (nread >= wc->reada_count)
+			break;
 
-	/*
-	 * the items in the ref were sorted when the ref was inserted
-	 * into the ref cache, so this is already in order
-	 */
-	for (i = 0; i < ref->nritems; i++) {
-		info = ref->extents + sorted[i].slot;
-		ret = btrfs_free_extent(trans, root, info->bytenr,
-					  info->num_bytes, ref->bytenr,
-					  ref->owner, ref->generation,
-					  info->objectid, 0);
-
-		atomic_inc(&root->fs_info->throttle_gen);
-		wake_up(&root->fs_info->transaction_throttle);
 		cond_resched();
+		bytenr = btrfs_node_blockptr(eb, slot);
+		generation = btrfs_node_ptr_generation(eb, slot);
 
-		BUG_ON(ret);
-		info++;
-	}
-
-	kfree(sorted);
-	return 0;
-}
-
-
-static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
-				     struct btrfs_root *root, u64 start,
-				     u64 len, u32 *refs)
-{
-	int ret;
-
-	ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
-	BUG_ON(ret);
-
-#if 0 /* some debugging code in case we see problems here */
-	/* if the refs count is one, it won't get increased again.  But
-	 * if the ref count is > 1, someone may be decreasing it at
-	 * the same time we are.
-	 */
-	if (*refs != 1) {
-		struct extent_buffer *eb = NULL;
-		eb = btrfs_find_create_tree_block(root, start, len);
-		if (eb)
-			btrfs_tree_lock(eb);
-
-		mutex_lock(&root->fs_info->alloc_mutex);
-		ret = lookup_extent_ref(NULL, root, start, len, refs);
-		BUG_ON(ret);
-		mutex_unlock(&root->fs_info->alloc_mutex);
-
-		if (eb) {
-			btrfs_tree_unlock(eb);
-			free_extent_buffer(eb);
-		}
-		if (*refs == 1) {
-			printk(KERN_ERR "btrfs block %llu went down to one "
-			       "during drop_snap\n", (unsigned long long)start);
-		}
-
-	}
-#endif
-
-	cond_resched();
-	return ret;
-}
-
+		if (slot == path->slots[wc->level])
+			goto reada;
 
-/*
- * this is used while deleting old snapshots, and it drops the refs
- * on a whole subtree starting from a level 1 node.
- *
- * The idea is to sort all the leaf pointers, and then drop the
- * ref on all the leaves in order.  Most of the time the leaves
- * will have ref cache entries, so no leaf IOs will be required to
- * find the extents they have references on.
- *
- * For each leaf, any references it has are also dropped in order
- *
- * This ends up dropping the references in something close to optimal
- * order for reading and modifying the extent allocation tree.
- */
-static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
-					struct btrfs_root *root,
-					struct btrfs_path *path)
-{
-	u64 bytenr;
-	u64 root_owner;
-	u64 root_gen;
-	struct extent_buffer *eb = path->nodes[1];
-	struct extent_buffer *leaf;
-	struct btrfs_leaf_ref *ref;
-	struct refsort *sorted = NULL;
-	int nritems = btrfs_header_nritems(eb);
-	int ret;
-	int i;
-	int refi = 0;
-	int slot = path->slots[1];
-	u32 blocksize = btrfs_level_size(root, 0);
-	u32 refs;
-
-	if (nritems == 0)
-		goto out;
-
-	root_owner = btrfs_header_owner(eb);
-	root_gen = btrfs_header_generation(eb);
-	sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
-
-	/*
-	 * step one, sort all the leaf pointers so we don't scribble
-	 * randomly into the extent allocation tree
-	 */
-	for (i = slot; i < nritems; i++) {
-		sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
-		sorted[refi].slot = i;
-		refi++;
-	}
-
-	/*
-	 * nritems won't be zero, but if we're picking up drop_snapshot
-	 * after a crash, slot might be > 0, so double check things
-	 * just in case.
-	 */
-	if (refi == 0)
-		goto out;
-
-	sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
-
-	/*
-	 * the first loop frees everything the leaves point to
-	 */
-	for (i = 0; i < refi; i++) {
-		u64 ptr_gen;
-
-		bytenr = sorted[i].bytenr;
-
-		/*
-		 * check the reference count on this leaf.  If it is > 1
-		 * we just decrement it below and don't update any
-		 * of the refs the leaf points to.
-		 */
-		ret = drop_snap_lookup_refcount(trans, root, bytenr,
-						blocksize, &refs);
-		BUG_ON(ret);
-		if (refs != 1)
+		if (wc->stage == UPDATE_BACKREF &&
+		    generation <= root->root_key.offset)
 			continue;
 
-		ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
-
-		/*
-		 * the leaf only had one reference, which means the
-		 * only thing pointing to this leaf is the snapshot
-		 * we're deleting.  It isn't possible for the reference
-		 * count to increase again later
-		 *
-		 * The reference cache is checked for the leaf,
-		 * and if found we'll be able to drop any refs held by
-		 * the leaf without needing to read it in.
-		 */
-		ref = btrfs_lookup_leaf_ref(root, bytenr);
-		if (ref && ref->generation != ptr_gen) {
-			btrfs_free_leaf_ref(root, ref);
-			ref = NULL;
-		}
-		if (ref) {
-			ret = cache_drop_leaf_ref(trans, root, ref);
-			BUG_ON(ret);
-			btrfs_remove_leaf_ref(root, ref);
-			btrfs_free_leaf_ref(root, ref);
-		} else {
-			/*
-			 * the leaf wasn't in the reference cache, so
-			 * we have to read it.
-			 */
-			leaf = read_tree_block(root, bytenr, blocksize,
-					       ptr_gen);
-			ret = btrfs_drop_leaf_ref(trans, root, leaf);
+		if (wc->stage == DROP_REFERENCE) {
+			ret = btrfs_lookup_extent_info(trans, root,
+						bytenr, blocksize,
+						&refs, NULL);
 			BUG_ON(ret);
-			free_extent_buffer(leaf);
-		}
-		atomic_inc(&root->fs_info->throttle_gen);
-		wake_up(&root->fs_info->transaction_throttle);
-		cond_resched();
-	}
-
-	/*
-	 * run through the loop again to free the refs on the leaves.
-	 * This is faster than doing it in the loop above because
-	 * the leaves are likely to be clustered together.  We end up
-	 * working in nice chunks on the extent allocation tree.
-	 */
-	for (i = 0; i < refi; i++) {
-		bytenr = sorted[i].bytenr;
-		ret = btrfs_free_extent(trans, root, bytenr,
-					blocksize, eb->start,
-					root_owner, root_gen, 0, 1);
-		BUG_ON(ret);
+			BUG_ON(refs == 0);
+			if (refs == 1)
+				goto reada;
 
-		atomic_inc(&root->fs_info->throttle_gen);
-		wake_up(&root->fs_info->transaction_throttle);
-		cond_resched();
-	}
-out:
-	kfree(sorted);
-
-	/*
-	 * update the path to show we've processed the entire level 1
-	 * node.  This will get saved into the root's drop_snapshot_progress
-	 * field so these drops are not repeated again if this transaction
-	 * commits.
-	 */
-	path->slots[1] = nritems;
-	return 0;
-}
-
-/*
- * helper function for drop_snapshot, this walks down the tree dropping ref
- * counts as it goes.
- */
-static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
-				   struct btrfs_root *root,
-				   struct btrfs_path *path, int *level)
-{
-	u64 root_owner;
-	u64 root_gen;
-	u64 bytenr;
-	u64 ptr_gen;
-	struct extent_buffer *next;
-	struct extent_buffer *cur;
-	struct extent_buffer *parent;
-	u32 blocksize;
-	int ret;
-	u32 refs;
-
-	WARN_ON(*level < 0);
-	WARN_ON(*level >= BTRFS_MAX_LEVEL);
-	ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
-				path->nodes[*level]->len, &refs);
-	BUG_ON(ret);
-	if (refs > 1)
-		goto out;
-
-	/*
-	 * walk down to the last node level and free all the leaves
-	 */
-	while (*level >= 0) {
-		WARN_ON(*level < 0);
-		WARN_ON(*level >= BTRFS_MAX_LEVEL);
-		cur = path->nodes[*level];
-
-		if (btrfs_header_level(cur) != *level)
-			WARN_ON(1);
-
-		if (path->slots[*level] >=
-		    btrfs_header_nritems(cur))
-			break;
-
-		/* the new code goes down to level 1 and does all the
-		 * leaves pointed to that node in bulk.  So, this check
-		 * for level 0 will always be false.
-		 *
-		 * But, the disk format allows the drop_snapshot_progress
-		 * field in the root to leave things in a state where
-		 * a leaf will need cleaning up here.  If someone crashes
-		 * with the old code and then boots with the new code,
-		 * we might find a leaf here.
-		 */
-		if (*level == 0) {
-			ret = btrfs_drop_leaf_ref(trans, root, cur);
-			BUG_ON(ret);
-			break;
+			if (!wc->update_ref ||
+			    generation <= root->root_key.offset)
+				continue;
+			btrfs_node_key_to_cpu(eb, &key, slot);
+			ret = btrfs_comp_cpu_keys(&key,
+						  &wc->update_progress);
+			if (ret < 0)
+				continue;
 		}
-
-		/*
-		 * once we get to level one, process the whole node
-		 * at once, including everything below it.
-		 */
-		if (*level == 1) {
-			ret = drop_level_one_refs(trans, root, path);
-			BUG_ON(ret);
+reada:
+		ret = readahead_tree_block(root, bytenr, blocksize,
+					   generation);
+		if (ret)
 			break;
-		}
-
-		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
-		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
-		blocksize = btrfs_level_size(root, *level - 1);
-
-		ret = drop_snap_lookup_refcount(trans, root, bytenr,
-						blocksize, &refs);
-		BUG_ON(ret);
-
-		/*
-		 * if there is more than one reference, we don't need
-		 * to read that node to drop any references it has.  We
-		 * just drop the ref we hold on that node and move on to the
-		 * next slot in this level.
-		 */
-		if (refs != 1) {
-			parent = path->nodes[*level];
-			root_owner = btrfs_header_owner(parent);
-			root_gen = btrfs_header_generation(parent);
-			path->slots[*level]++;
-
-			ret = btrfs_free_extent(trans, root, bytenr,
-						blocksize, parent->start,
-						root_owner, root_gen,
-						*level - 1, 1);
-			BUG_ON(ret);
-
-			atomic_inc(&root->fs_info->throttle_gen);
-			wake_up(&root->fs_info->transaction_throttle);
-			cond_resched();
-
-			continue;
-		}
-
-		/*
-		 * we need to keep freeing things in the next level down.
-		 * read the block and loop around to process it
-		 */
-		next = read_tree_block(root, bytenr, blocksize, ptr_gen);
-		WARN_ON(*level <= 0);
-		if (path->nodes[*level-1])
-			free_extent_buffer(path->nodes[*level-1]);
-		path->nodes[*level-1] = next;
-		*level = btrfs_header_level(next);
-		path->slots[*level] = 0;
-		cond_resched();
+		last = bytenr + blocksize;
+		nread++;
 	}
-out:
-	WARN_ON(*level < 0);
-	WARN_ON(*level >= BTRFS_MAX_LEVEL);
-
-	if (path->nodes[*level] == root->node) {
-		parent = path->nodes[*level];
-		bytenr = path->nodes[*level]->start;
-	} else {
-		parent = path->nodes[*level + 1];
-		bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
-	}
-
-	blocksize = btrfs_level_size(root, *level);
-	root_owner = btrfs_header_owner(parent);
-	root_gen = btrfs_header_generation(parent);
-
-	/*
-	 * cleanup and free the reference on the last node
-	 * we processed
-	 */
-	ret = btrfs_free_extent(trans, root, bytenr, blocksize,
-				  parent->start, root_owner, root_gen,
-				  *level, 1);
-	free_extent_buffer(path->nodes[*level]);
-	path->nodes[*level] = NULL;
-
-	*level += 1;
-	BUG_ON(ret);
-
-	cond_resched();
-	return 0;
+	wc->reada_slot = slot;
 }
-#endif
-
-struct walk_control {
-	u64 refs[BTRFS_MAX_LEVEL];
-	u64 flags[BTRFS_MAX_LEVEL];
-	struct btrfs_key update_progress;
-	int stage;
-	int level;
-	int shared_level;
-	int update_ref;
-	int keep_locks;
-};
-
-#define DROP_REFERENCE	1
-#define UPDATE_BACKREF	2
 
 /*
  * hepler to process tree block while walking down the tree.
  *
- * when wc->stage == DROP_REFERENCE, this function checks
- * reference count of the block. if the block is shared and
- * we need update back refs for the subtree rooted at the
- * block, this function changes wc->stage to UPDATE_BACKREF
- *
  * when wc->stage == UPDATE_BACKREF, this function updates
  * back refs for pointers in the block.
  *
@@ -4805,7 +4587,6 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
 {
 	int level = wc->level;
 	struct extent_buffer *eb = path->nodes[level];
-	struct btrfs_key key;
 	u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 	int ret;
 
@@ -4828,21 +4609,6 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
 		BUG_ON(wc->refs[level] == 0);
 	}
 
-	if (wc->stage == DROP_REFERENCE &&
-	    wc->update_ref && wc->refs[level] > 1) {
-		BUG_ON(eb == root->node);
-		BUG_ON(path->slots[level] > 0);
-		if (level == 0)
-			btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
-		else
-			btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
-		if (btrfs_header_owner(eb) == root->root_key.objectid &&
-		    btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
-			wc->stage = UPDATE_BACKREF;
-			wc->shared_level = level;
-		}
-	}
-
 	if (wc->stage == DROP_REFERENCE) {
 		if (wc->refs[level] > 1)
 			return 1;
@@ -4878,6 +4644,123 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+/*
+ * hepler to process tree block pointer.
+ *
+ * when wc->stage == DROP_REFERENCE, this function checks
+ * reference count of the block pointed to. if the block
+ * is shared and we need update back refs for the subtree
+ * rooted at the block, this function changes wc->stage to
+ * UPDATE_BACKREF. if the block is shared and there is no
+ * need to update back, this function drops the reference
+ * to the block.
+ *
+ * NOTE: return value 1 means we should stop walking down.
+ */
+static noinline int do_walk_down(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root,
+				 struct btrfs_path *path,
+				 struct walk_control *wc)
+{
+	u64 bytenr;
+	u64 generation;
+	u64 parent;
+	u32 blocksize;
+	struct btrfs_key key;
+	struct extent_buffer *next;
+	int level = wc->level;
+	int reada = 0;
+	int ret = 0;
+
+	generation = btrfs_node_ptr_generation(path->nodes[level],
+					       path->slots[level]);
+	/*
+	 * if the lower level block was created before the snapshot
+	 * was created, we know there is no need to update back refs
+	 * for the subtree
+	 */
+	if (wc->stage == UPDATE_BACKREF &&
+	    generation <= root->root_key.offset)
+		return 1;
+
+	bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
+	blocksize = btrfs_level_size(root, level - 1);
+
+	next = btrfs_find_tree_block(root, bytenr, blocksize);
+	if (!next) {
+		next = btrfs_find_create_tree_block(root, bytenr, blocksize);
+		reada = 1;
+	}
+	btrfs_tree_lock(next);
+	btrfs_set_lock_blocking(next);
+
+	if (wc->stage == DROP_REFERENCE) {
+		ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
+					       &wc->refs[level - 1],
+					       &wc->flags[level - 1]);
+		BUG_ON(ret);
+		BUG_ON(wc->refs[level - 1] == 0);
+
+		if (wc->refs[level - 1] > 1) {
+			if (!wc->update_ref ||
+			    generation <= root->root_key.offset)
+				goto skip;
+
+			btrfs_node_key_to_cpu(path->nodes[level], &key,
+					      path->slots[level]);
+			ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
+			if (ret < 0)
+				goto skip;
+
+			wc->stage = UPDATE_BACKREF;
+			wc->shared_level = level - 1;
+		}
+	}
+
+	if (!btrfs_buffer_uptodate(next, generation)) {
+		btrfs_tree_unlock(next);
+		free_extent_buffer(next);
+		next = NULL;
+	}
+
+	if (!next) {
+		if (reada && level == 1)
+			reada_walk_down(trans, root, wc, path);
+		next = read_tree_block(root, bytenr, blocksize, generation);
+		btrfs_tree_lock(next);
+		btrfs_set_lock_blocking(next);
+	}
+
+	level--;
+	BUG_ON(level != btrfs_header_level(next));
+	path->nodes[level] = next;
+	path->slots[level] = 0;
+	path->locks[level] = 1;
+	wc->level = level;
+	if (wc->level == 1)
+		wc->reada_slot = 0;
+	return 0;
+skip:
+	wc->refs[level - 1] = 0;
+	wc->flags[level - 1] = 0;
+
+	if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
+		parent = path->nodes[level]->start;
+	} else {
+		BUG_ON(root->root_key.objectid !=
+		       btrfs_header_owner(path->nodes[level]));
+		parent = 0;
+	}
+
+	ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
+				root->root_key.objectid, level - 1, 0);
+	BUG_ON(ret);
+
+	btrfs_tree_unlock(next);
+	free_extent_buffer(next);
+	return 1;
+}
+
 /*
  * hepler to process tree block while walking up the tree.
  *
@@ -4905,7 +4788,6 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 		if (level < wc->shared_level)
 			goto out;
 
-		BUG_ON(wc->refs[level] <= 1);
 		ret = find_next_key(path, level + 1, &wc->update_progress);
 		if (ret > 0)
 			wc->update_ref = 0;
@@ -4936,8 +4818,6 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 				path->locks[level] = 0;
 				return 1;
 			}
-		} else {
-			BUG_ON(level != 0);
 		}
 	}
 
@@ -4990,17 +4870,13 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
 				   struct btrfs_path *path,
 				   struct walk_control *wc)
 {
-	struct extent_buffer *next;
-	struct extent_buffer *cur;
-	u64 bytenr;
-	u64 ptr_gen;
-	u32 blocksize;
 	int level = wc->level;
 	int ret;
 
 	while (level >= 0) {
-		cur = path->nodes[level];
-		BUG_ON(path->slots[level] >= btrfs_header_nritems(cur));
+		if (path->slots[level] >=
+		    btrfs_header_nritems(path->nodes[level]))
+			break;
 
 		ret = walk_down_proc(trans, root, path, wc);
 		if (ret > 0)
@@ -5009,20 +4885,12 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
 		if (level == 0)
 			break;
 
-		bytenr = btrfs_node_blockptr(cur, path->slots[level]);
-		blocksize = btrfs_level_size(root, level - 1);
-		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]);
-
-		next = read_tree_block(root, bytenr, blocksize, ptr_gen);
-		btrfs_tree_lock(next);
-		btrfs_set_lock_blocking(next);
-
-		level--;
-		BUG_ON(level != btrfs_header_level(next));
-		path->nodes[level] = next;
-		path->slots[level] = 0;
-		path->locks[level] = 1;
-		wc->level = level;
+		ret = do_walk_down(trans, root, path, wc);
+		if (ret > 0) {
+			path->slots[level]++;
+			continue;
+		}
+		level = wc->level;
 	}
 	return 0;
 }
@@ -5112,9 +4980,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
 			err = ret;
 			goto out;
 		}
-		btrfs_node_key_to_cpu(path->nodes[level], &key,
-				      path->slots[level]);
-		WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
+		WARN_ON(ret > 0);
 
 		/*
 		 * unlock our path, this is safe because only this
@@ -5149,6 +5015,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
 	wc->stage = DROP_REFERENCE;
 	wc->update_ref = update_ref;
 	wc->keep_locks = 0;
+	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
 
 	while (1) {
 		ret = walk_down_tree(trans, root, path, wc);
@@ -5201,9 +5068,24 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
 	ret = btrfs_del_root(trans, tree_root, &root->root_key);
 	BUG_ON(ret);
 
-	free_extent_buffer(root->node);
-	free_extent_buffer(root->commit_root);
-	kfree(root);
+	if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
+		ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
+					   NULL, NULL);
+		BUG_ON(ret < 0);
+		if (ret > 0) {
+			ret = btrfs_del_orphan_item(trans, tree_root,
+						    root->root_key.objectid);
+			BUG_ON(ret);
+		}
+	}
+
+	if (root->in_radix) {
+		btrfs_free_fs_root(tree_root->fs_info, root);
+	} else {
+		free_extent_buffer(root->node);
+		free_extent_buffer(root->commit_root);
+		kfree(root);
+	}
 out:
 	btrfs_end_transaction(trans, tree_root);
 	kfree(wc);
@@ -5255,6 +5137,7 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 	wc->stage = DROP_REFERENCE;
 	wc->update_ref = 0;
 	wc->keep_locks = 1;
+	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
 
 	while (1) {
 		wret = walk_down_tree(trans, root, path, wc);
@@ -5397,9 +5280,9 @@ static noinline int relocate_data_extent(struct inode *reloc_inode,
 	lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
 	while (1) {
 		int ret;
-		spin_lock(&em_tree->lock);
+		write_lock(&em_tree->lock);
 		ret = add_extent_mapping(em_tree, em);
-		spin_unlock(&em_tree->lock);
+		write_unlock(&em_tree->lock);
 		if (ret != -EEXIST) {
 			free_extent_map(em);
 			break;
@@ -6842,287 +6725,86 @@ int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
 	return 0;
 }
 
-#if 0
-static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
-				 struct btrfs_root *root,
-				 u64 objectid, u64 size)
-{
-	struct btrfs_path *path;
-	struct btrfs_inode_item *item;
-	struct extent_buffer *leaf;
-	int ret;
-
-	path = btrfs_alloc_path();
-	if (!path)
-		return -ENOMEM;
-
-	path->leave_spinning = 1;
-	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
-	if (ret)
-		goto out;
-
-	leaf = path->nodes[0];
-	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
-	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
-	btrfs_set_inode_generation(leaf, item, 1);
-	btrfs_set_inode_size(leaf, item, size);
-	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
-	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
-	btrfs_mark_buffer_dirty(leaf);
-	btrfs_release_path(root, path);
-out:
-	btrfs_free_path(path);
-	return ret;
-}
-
-static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
-					struct btrfs_block_group_cache *group)
+/*
+ * checks to see if its even possible to relocate this block group.
+ *
+ * @return - -1 if it's not a good idea to relocate this block group, 0 if its
+ * ok to go ahead and try.
+ */
+int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
 {
-	struct inode *inode = NULL;
-	struct btrfs_trans_handle *trans;
-	struct btrfs_root *root;
-	struct btrfs_key root_key;
-	u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
-	int err = 0;
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_space_info *space_info;
+	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
+	struct btrfs_device *device;
+	int full = 0;
+	int ret = 0;
 
-	root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
-	root_key.type = BTRFS_ROOT_ITEM_KEY;
-	root_key.offset = (u64)-1;
-	root = btrfs_read_fs_root_no_name(fs_info, &root_key);
-	if (IS_ERR(root))
-		return ERR_CAST(root);
+	block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
 
-	trans = btrfs_start_transaction(root, 1);
-	BUG_ON(!trans);
+	/* odd, couldn't find the block group, leave it alone */
+	if (!block_group)
+		return -1;
 
-	err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
-	if (err)
+	/* no bytes used, we're good */
+	if (!btrfs_block_group_used(&block_group->item))
 		goto out;
 
-	err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
-	BUG_ON(err);
-
-	err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
-				       group->key.offset, 0, group->key.offset,
-				       0, 0, 0);
-	BUG_ON(err);
-
-	inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
-	if (inode->i_state & I_NEW) {
-		BTRFS_I(inode)->root = root;
-		BTRFS_I(inode)->location.objectid = objectid;
-		BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
-		BTRFS_I(inode)->location.offset = 0;
-		btrfs_read_locked_inode(inode);
-		unlock_new_inode(inode);
-		BUG_ON(is_bad_inode(inode));
-	} else {
-		BUG_ON(1);
-	}
-	BTRFS_I(inode)->index_cnt = group->key.objectid;
-
-	err = btrfs_orphan_add(trans, inode);
-out:
-	btrfs_end_transaction(trans, root);
-	if (err) {
-		if (inode)
-			iput(inode);
-		inode = ERR_PTR(err);
-	}
-	return inode;
-}
-
-int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
-{
-
-	struct btrfs_ordered_sum *sums;
-	struct btrfs_sector_sum *sector_sum;
-	struct btrfs_ordered_extent *ordered;
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct list_head list;
-	size_t offset;
-	int ret;
-	u64 disk_bytenr;
-
-	INIT_LIST_HEAD(&list);
-
-	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
-	BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
-
-	disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
-	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
-				       disk_bytenr + len - 1, &list);
-
-	while (!list_empty(&list)) {
-		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
-		list_del_init(&sums->list);
-
-		sector_sum = sums->sums;
-		sums->bytenr = ordered->start;
+	space_info = block_group->space_info;
+	spin_lock(&space_info->lock);
 
-		offset = 0;
-		while (offset < sums->len) {
-			sector_sum->bytenr += ordered->start - disk_bytenr;
-			sector_sum++;
-			offset += root->sectorsize;
-		}
+	full = space_info->full;
 
-		btrfs_add_ordered_sum(inode, ordered, sums);
+	/*
+	 * if this is the last block group we have in this space, we can't
+	 * relocate it unless we're able to allocate a new chunk below.
+	 *
+	 * Otherwise, we need to make sure we have room in the space to handle
+	 * all of the extents from this block group.  If we can, we're good
+	 */
+	if ((space_info->total_bytes != block_group->key.offset) &&
+	   (space_info->bytes_used + space_info->bytes_reserved +
+	    space_info->bytes_pinned + space_info->bytes_readonly +
+	    btrfs_block_group_used(&block_group->item) <
+	    space_info->total_bytes)) {
+		spin_unlock(&space_info->lock);
+		goto out;
 	}
-	btrfs_put_ordered_extent(ordered);
-	return 0;
-}
-
-int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
-{
-	struct btrfs_trans_handle *trans;
-	struct btrfs_path *path;
-	struct btrfs_fs_info *info = root->fs_info;
-	struct extent_buffer *leaf;
-	struct inode *reloc_inode;
-	struct btrfs_block_group_cache *block_group;
-	struct btrfs_key key;
-	u64 skipped;
-	u64 cur_byte;
-	u64 total_found;
-	u32 nritems;
-	int ret;
-	int progress;
-	int pass = 0;
-
-	root = root->fs_info->extent_root;
-
-	block_group = btrfs_lookup_block_group(info, group_start);
-	BUG_ON(!block_group);
-
-	printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
-	       (unsigned long long)block_group->key.objectid,
-	       (unsigned long long)block_group->flags);
-
-	path = btrfs_alloc_path();
-	BUG_ON(!path);
-
-	reloc_inode = create_reloc_inode(info, block_group);
-	BUG_ON(IS_ERR(reloc_inode));
-
-	__alloc_chunk_for_shrink(root, block_group, 1);
-	set_block_group_readonly(block_group);
-
-	btrfs_start_delalloc_inodes(info->tree_root);
-	btrfs_wait_ordered_extents(info->tree_root, 0);
-again:
-	skipped = 0;
-	total_found = 0;
-	progress = 0;
-	key.objectid = block_group->key.objectid;
-	key.offset = 0;
-	key.type = 0;
-	cur_byte = key.objectid;
-
-	trans = btrfs_start_transaction(info->tree_root, 1);
-	btrfs_commit_transaction(trans, info->tree_root);
+	spin_unlock(&space_info->lock);
 
-	mutex_lock(&root->fs_info->cleaner_mutex);
-	btrfs_clean_old_snapshots(info->tree_root);
-	btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
-	mutex_unlock(&root->fs_info->cleaner_mutex);
+	/*
+	 * ok we don't have enough space, but maybe we have free space on our
+	 * devices to allocate new chunks for relocation, so loop through our
+	 * alloc devices and guess if we have enough space.  However, if we
+	 * were marked as full, then we know there aren't enough chunks, and we
+	 * can just return.
+	 */
+	ret = -1;
+	if (full)
+		goto out;
 
-	trans = btrfs_start_transaction(info->tree_root, 1);
-	btrfs_commit_transaction(trans, info->tree_root);
+	mutex_lock(&root->fs_info->chunk_mutex);
+	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
+		u64 min_free = btrfs_block_group_used(&block_group->item);
+		u64 dev_offset, max_avail;
 
-	while (1) {
-		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
-		if (ret < 0)
-			goto out;
-next:
-		leaf = path->nodes[0];
-		nritems = btrfs_header_nritems(leaf);
-		if (path->slots[0] >= nritems) {
-			ret = btrfs_next_leaf(root, path);
-			if (ret < 0)
-				goto out;
-			if (ret == 1) {
-				ret = 0;
+		/*
+		 * check to make sure we can actually find a chunk with enough
+		 * space to fit our block group in.
+		 */
+		if (device->total_bytes > device->bytes_used + min_free) {
+			ret = find_free_dev_extent(NULL, device, min_free,
+						   &dev_offset, &max_avail);
+			if (!ret)
 				break;
-			}
-			leaf = path->nodes[0];
-			nritems = btrfs_header_nritems(leaf);
-		}
-
-		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
-
-		if (key.objectid >= block_group->key.objectid +
-		    block_group->key.offset)
-			break;
-
-		if (progress && need_resched()) {
-			btrfs_release_path(root, path);
-			cond_resched();
-			progress = 0;
-			continue;
+			ret = -1;
 		}
-		progress = 1;
-
-		if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
-		    key.objectid + key.offset <= cur_byte) {
-			path->slots[0]++;
-			goto next;
-		}
-
-		total_found++;
-		cur_byte = key.objectid + key.offset;
-		btrfs_release_path(root, path);
-
-		__alloc_chunk_for_shrink(root, block_group, 0);
-		ret = relocate_one_extent(root, path, &key, block_group,
-					  reloc_inode, pass);
-		BUG_ON(ret < 0);
-		if (ret > 0)
-			skipped++;
-
-		key.objectid = cur_byte;
-		key.type = 0;
-		key.offset = 0;
 	}
-
-	btrfs_release_path(root, path);
-
-	if (pass == 0) {
-		btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
-		invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
-	}
-
-	if (total_found > 0) {
-		printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
-		       (unsigned long long)total_found, pass);
-		pass++;
-		if (total_found == skipped && pass > 2) {
-			iput(reloc_inode);
-			reloc_inode = create_reloc_inode(info, block_group);
-			pass = 0;
-		}
-		goto again;
-	}
-
-	/* delete reloc_inode */
-	iput(reloc_inode);
-
-	/* unpin extents in this range */
-	trans = btrfs_start_transaction(info->tree_root, 1);
-	btrfs_commit_transaction(trans, info->tree_root);
-
-	spin_lock(&block_group->lock);
-	WARN_ON(block_group->pinned > 0);
-	WARN_ON(block_group->reserved > 0);
-	WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
-	spin_unlock(&block_group->lock);
-	btrfs_put_block_group(block_group);
-	ret = 0;
+	mutex_unlock(&root->fs_info->chunk_mutex);
 out:
-	btrfs_free_path(path);
+	btrfs_put_block_group(block_group);
 	return ret;
 }
-#endif
 
 static int find_first_block_group(struct btrfs_root *root,
 		struct btrfs_path *path, struct btrfs_key *key)
@@ -7165,8 +6847,18 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
 {
 	struct btrfs_block_group_cache *block_group;
 	struct btrfs_space_info *space_info;
+	struct btrfs_caching_control *caching_ctl;
 	struct rb_node *n;
 
+	down_write(&info->extent_commit_sem);
+	while (!list_empty(&info->caching_block_groups)) {
+		caching_ctl = list_entry(info->caching_block_groups.next,
+					 struct btrfs_caching_control, list);
+		list_del(&caching_ctl->list);
+		put_caching_control(caching_ctl);
+	}
+	up_write(&info->extent_commit_sem);
+
 	spin_lock(&info->block_group_cache_lock);
 	while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
 		block_group = rb_entry(n, struct btrfs_block_group_cache,
@@ -7180,8 +6872,7 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
 		up_write(&block_group->space_info->groups_sem);
 
 		if (block_group->cached == BTRFS_CACHE_STARTED)
-			wait_event(block_group->caching_q,
-				   block_group_cache_done(block_group));
+			wait_block_group_cache_done(block_group);
 
 		btrfs_remove_free_space_cache(block_group);
 
@@ -7251,7 +6942,6 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 		spin_lock_init(&cache->lock);
 		spin_lock_init(&cache->tree_lock);
 		cache->fs_info = info;
-		init_waitqueue_head(&cache->caching_q);
 		INIT_LIST_HEAD(&cache->list);
 		INIT_LIST_HEAD(&cache->cluster_list);
 
@@ -7273,8 +6963,6 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 		cache->flags = btrfs_block_group_flags(&cache->item);
 		cache->sectorsize = root->sectorsize;
 
-		remove_sb_from_cache(root, cache);
-
 		/*
 		 * check for two cases, either we are full, and therefore
 		 * don't need to bother with the caching work since we won't
@@ -7283,13 +6971,19 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 		 * time, particularly in the full case.
 		 */
 		if (found_key.offset == btrfs_block_group_used(&cache->item)) {
+			exclude_super_stripes(root, cache);
+			cache->last_byte_to_unpin = (u64)-1;
 			cache->cached = BTRFS_CACHE_FINISHED;
+			free_excluded_extents(root, cache);
 		} else if (btrfs_block_group_used(&cache->item) == 0) {
+			exclude_super_stripes(root, cache);
+			cache->last_byte_to_unpin = (u64)-1;
 			cache->cached = BTRFS_CACHE_FINISHED;
 			add_new_free_space(cache, root->fs_info,
 					   found_key.objectid,
 					   found_key.objectid +
 					   found_key.offset);
+			free_excluded_extents(root, cache);
 		}
 
 		ret = update_space_info(info, cache->flags, found_key.offset,
@@ -7297,6 +6991,10 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 					&space_info);
 		BUG_ON(ret);
 		cache->space_info = space_info;
+		spin_lock(&cache->space_info->lock);
+		cache->space_info->bytes_super += cache->bytes_super;
+		spin_unlock(&cache->space_info->lock);
+
 		down_write(&space_info->groups_sem);
 		list_add_tail(&cache->list, &space_info->block_groups);
 		up_write(&space_info->groups_sem);
@@ -7346,7 +7044,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 	atomic_set(&cache->count, 1);
 	spin_lock_init(&cache->lock);
 	spin_lock_init(&cache->tree_lock);
-	init_waitqueue_head(&cache->caching_q);
 	INIT_LIST_HEAD(&cache->list);
 	INIT_LIST_HEAD(&cache->cluster_list);
 
@@ -7355,15 +7052,23 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 	cache->flags = type;
 	btrfs_set_block_group_flags(&cache->item, type);
 
+	cache->last_byte_to_unpin = (u64)-1;
 	cache->cached = BTRFS_CACHE_FINISHED;
-	remove_sb_from_cache(root, cache);
+	exclude_super_stripes(root, cache);
 
 	add_new_free_space(cache, root->fs_info, chunk_offset,
 			   chunk_offset + size);
 
+	free_excluded_extents(root, cache);
+
 	ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
 				&cache->space_info);
 	BUG_ON(ret);
+
+	spin_lock(&cache->space_info->lock);
+	cache->space_info->bytes_super += cache->bytes_super;
+	spin_unlock(&cache->space_info->lock);
+
 	down_write(&cache->space_info->groups_sem);
 	list_add_tail(&cache->list, &cache->space_info->block_groups);
 	up_write(&cache->space_info->groups_sem);
@@ -7429,8 +7134,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 	up_write(&block_group->space_info->groups_sem);
 
 	if (block_group->cached == BTRFS_CACHE_STARTED)
-		wait_event(block_group->caching_q,
-			   block_group_cache_done(block_group));
+		wait_block_group_cache_done(block_group);
 
 	btrfs_remove_free_space_cache(block_group);
 

fs/btrfs/extent_io.c

@@ -367,10 +367,10 @@ static int insert_state(struct extent_io_tree *tree,
 	}
 	if (bits & EXTENT_DIRTY)
 		tree->dirty_bytes += end - start + 1;
-	set_state_cb(tree, state, bits);
-	state->state |= bits;
 	state->start = start;
 	state->end = end;
+	set_state_cb(tree, state, bits);
+	state->state |= bits;
 	node = tree_insert(&tree->state, end, &state->rb_node);
 	if (node) {
 		struct extent_state *found;
@@ -471,10 +471,14 @@ static int clear_state_bit(struct extent_io_tree *tree,
  * bits were already set, or zero if none of the bits were already set.
  */
 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
-		     int bits, int wake, int delete, gfp_t mask)
+		     int bits, int wake, int delete,
+		     struct extent_state **cached_state,
+		     gfp_t mask)
 {
 	struct extent_state *state;
+	struct extent_state *cached;
 	struct extent_state *prealloc = NULL;
+	struct rb_node *next_node;
 	struct rb_node *node;
 	u64 last_end;
 	int err;
@@ -488,6 +492,17 @@ again:
 	}
 
 	spin_lock(&tree->lock);
+	if (cached_state) {
+		cached = *cached_state;
+		*cached_state = NULL;
+		cached_state = NULL;
+		if (cached && cached->tree && cached->start == start) {
+			atomic_dec(&cached->refs);
+			state = cached;
+			goto hit_next;
+		}
+		free_extent_state(cached);
+	}
 	/*
 	 * this search will find the extents that end after
 	 * our range starts
@@ -496,6 +511,7 @@ again:
 	if (!node)
 		goto out;
 	state = rb_entry(node, struct extent_state, rb_node);
+hit_next:
 	if (state->start > end)
 		goto out;
 	WARN_ON(state->end < start);
@@ -531,8 +547,6 @@ again:
 			if (last_end == (u64)-1)
 				goto out;
 			start = last_end + 1;
-		} else {
-			start = state->start;
 		}
 		goto search_again;
 	}
@@ -550,16 +564,28 @@ again:
 
 		if (wake)
 			wake_up(&state->wq);
+
 		set |= clear_state_bit(tree, prealloc, bits,
 				       wake, delete);
 		prealloc = NULL;
 		goto out;
 	}
 
+	if (state->end < end && prealloc && !need_resched())
+		next_node = rb_next(&state->rb_node);
+	else
+		next_node = NULL;
+
 	set |= clear_state_bit(tree, state, bits, wake, delete);
 	if (last_end == (u64)-1)
 		goto out;
 	start = last_end + 1;
+	if (start <= end && next_node) {
+		state = rb_entry(next_node, struct extent_state,
+				 rb_node);
+		if (state->start == start)
+			goto hit_next;
+	}
 	goto search_again;
 
 out:
@@ -653,28 +679,40 @@ static void set_state_bits(struct extent_io_tree *tree,
 	state->state |= bits;
 }
 
+static void cache_state(struct extent_state *state,
+			struct extent_state **cached_ptr)
+{
+	if (cached_ptr && !(*cached_ptr)) {
+		if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
+			*cached_ptr = state;
+			atomic_inc(&state->refs);
+		}
+	}
+}
+
 /*
- * set some bits on a range in the tree.  This may require allocations
- * or sleeping, so the gfp mask is used to indicate what is allowed.
+ * set some bits on a range in the tree.  This may require allocations or
+ * sleeping, so the gfp mask is used to indicate what is allowed.
  *
- * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
- * range already has the desired bits set.  The start of the existing
- * range is returned in failed_start in this case.
+ * If any of the exclusive bits are set, this will fail with -EEXIST if some
+ * part of the range already has the desired bits set.  The start of the
+ * existing range is returned in failed_start in this case.
  *
- * [start, end] is inclusive
- * This takes the tree lock.
+ * [start, end] is inclusive This takes the tree lock.
  */
+
 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
-			  int bits, int exclusive, u64 *failed_start,
+			  int bits, int exclusive_bits, u64 *failed_start,
+			  struct extent_state **cached_state,
 			  gfp_t mask)
 {
 	struct extent_state *state;
 	struct extent_state *prealloc = NULL;
 	struct rb_node *node;
 	int err = 0;
-	int set;
 	u64 last_start;
 	u64 last_end;
+
 again:
 	if (!prealloc && (mask & __GFP_WAIT)) {
 		prealloc = alloc_extent_state(mask);
@@ -683,6 +721,13 @@ again:
 	}
 
 	spin_lock(&tree->lock);
+	if (cached_state && *cached_state) {
+		state = *cached_state;
+		if (state->start == start && state->tree) {
+			node = &state->rb_node;
+			goto hit_next;
+		}
+	}
 	/*
 	 * this search will find all the extents that end after
 	 * our range starts.
@@ -694,8 +739,8 @@ again:
 		BUG_ON(err == -EEXIST);
 		goto out;
 	}
-
 	state = rb_entry(node, struct extent_state, rb_node);
+hit_next:
 	last_start = state->start;
 	last_end = state->end;
 
@@ -706,17 +751,29 @@ again:
 	 * Just lock what we found and keep going
 	 */
 	if (state->start == start && state->end <= end) {
-		set = state->state & bits;
-		if (set && exclusive) {
+		struct rb_node *next_node;
+		if (state->state & exclusive_bits) {
 			*failed_start = state->start;
 			err = -EEXIST;
 			goto out;
 		}
+
 		set_state_bits(tree, state, bits);
+		cache_state(state, cached_state);
 		merge_state(tree, state);
 		if (last_end == (u64)-1)
 			goto out;
+
 		start = last_end + 1;
+		if (start < end && prealloc && !need_resched()) {
+			next_node = rb_next(node);
+			if (next_node) {
+				state = rb_entry(next_node, struct extent_state,
+						 rb_node);
+				if (state->start == start)
+					goto hit_next;
+			}
+		}
 		goto search_again;
 	}
 
@@ -737,8 +794,7 @@ again:
 	 * desired bit on it.
 	 */
 	if (state->start < start) {
-		set = state->state & bits;
-		if (exclusive && set) {
+		if (state->state & exclusive_bits) {
 			*failed_start = start;
 			err = -EEXIST;
 			goto out;
@@ -750,12 +806,11 @@ again:
 			goto out;
 		if (state->end <= end) {
 			set_state_bits(tree, state, bits);
+			cache_state(state, cached_state);
 			merge_state(tree, state);
 			if (last_end == (u64)-1)
 				goto out;
 			start = last_end + 1;
-		} else {
-			start = state->start;
 		}
 		goto search_again;
 	}
@@ -774,6 +829,7 @@ again:
 			this_end = last_start - 1;
 		err = insert_state(tree, prealloc, start, this_end,
 				   bits);
+		cache_state(prealloc, cached_state);
 		prealloc = NULL;
 		BUG_ON(err == -EEXIST);
 		if (err)
@@ -788,8 +844,7 @@ again:
 	 * on the first half
 	 */
 	if (state->start <= end && state->end > end) {
-		set = state->state & bits;
-		if (exclusive && set) {
+		if (state->state & exclusive_bits) {
 			*failed_start = start;
 			err = -EEXIST;
 			goto out;
@@ -798,6 +853,7 @@ again:
 		BUG_ON(err == -EEXIST);
 
 		set_state_bits(tree, prealloc, bits);
+		cache_state(prealloc, cached_state);
 		merge_state(tree, prealloc);
 		prealloc = NULL;
 		goto out;
@@ -826,86 +882,64 @@ int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 		     gfp_t mask)
 {
 	return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
-			      mask);
-}
-
-int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
-		       gfp_t mask)
-{
-	return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
+			      NULL, mask);
 }
 
 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 		    int bits, gfp_t mask)
 {
 	return set_extent_bit(tree, start, end, bits, 0, NULL,
-			      mask);
+			      NULL, mask);
 }
 
 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 		      int bits, gfp_t mask)
 {
-	return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
+	return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
 }
 
 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
 		     gfp_t mask)
 {
 	return set_extent_bit(tree, start, end,
-			      EXTENT_DELALLOC | EXTENT_DIRTY,
-			      0, NULL, mask);
+			      EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
+			      0, NULL, NULL, mask);
 }
 
 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
 		       gfp_t mask)
 {
 	return clear_extent_bit(tree, start, end,
-				EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
-}
-
-int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
-			 gfp_t mask)
-{
-	return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
+				EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
+				NULL, mask);
 }
 
 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 		     gfp_t mask)
 {
 	return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
-			      mask);
+			      NULL, mask);
 }
 
 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
 		       gfp_t mask)
 {
-	return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
+	return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0,
+				NULL, mask);
 }
 
 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 			gfp_t mask)
 {
 	return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
-			      mask);
+			      NULL, mask);
 }
 
 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
 				 u64 end, gfp_t mask)
 {
-	return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
-}
-
-static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
-			 gfp_t mask)
-{
-	return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
-			      0, NULL, mask);
-}
-
-static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
-				  u64 end, gfp_t mask)
-{
-	return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
+	return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
+				NULL, mask);
 }
 
 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
@@ -917,13 +951,15 @@ int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
  * either insert or lock state struct between start and end use mask to tell
  * us if waiting is desired.
  */
-int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
+int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
+		     int bits, struct extent_state **cached_state, gfp_t mask)
 {
 	int err;
 	u64 failed_start;
 	while (1) {
-		err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
-				     &failed_start, mask);
+		err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
+				     EXTENT_LOCKED, &failed_start,
+				     cached_state, mask);
 		if (err == -EEXIST && (mask & __GFP_WAIT)) {
 			wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 			start = failed_start;
@@ -935,27 +971,40 @@ int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
 	return err;
 }
 
+int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
+{
+	return lock_extent_bits(tree, start, end, 0, NULL, mask);
+}
+
 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 		    gfp_t mask)
 {
 	int err;
 	u64 failed_start;
 
-	err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
-			     &failed_start, mask);
+	err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
+			     &failed_start, NULL, mask);
 	if (err == -EEXIST) {
 		if (failed_start > start)
 			clear_extent_bit(tree, start, failed_start - 1,
-					 EXTENT_LOCKED, 1, 0, mask);
+					 EXTENT_LOCKED, 1, 0, NULL, mask);
 		return 0;
 	}
 	return 1;
 }
 
+int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
+			 struct extent_state **cached, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
+				mask);
+}
+
 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 		  gfp_t mask)
 {
-	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
+	return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
+				mask);
 }
 
 /*
@@ -974,7 +1023,6 @@ int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
 		page_cache_release(page);
 		index++;
 	}
-	set_extent_dirty(tree, start, end, GFP_NOFS);
 	return 0;
 }
 
@@ -994,7 +1042,6 @@ static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 		page_cache_release(page);
 		index++;
 	}
-	set_extent_writeback(tree, start, end, GFP_NOFS);
 	return 0;
 }
 
@@ -1232,6 +1279,7 @@ static noinline u64 find_lock_delalloc_range(struct inode *inode,
 	u64 delalloc_start;
 	u64 delalloc_end;
 	u64 found;
+	struct extent_state *cached_state = NULL;
 	int ret;
 	int loops = 0;
 
@@ -1269,6 +1317,7 @@ again:
 		/* some of the pages are gone, lets avoid looping by
 		 * shortening the size of the delalloc range we're searching
 		 */
+		free_extent_state(cached_state);
 		if (!loops) {
 			unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
 			max_bytes = PAGE_CACHE_SIZE - offset;
@@ -1282,18 +1331,21 @@ again:
 	BUG_ON(ret);
 
 	/* step three, lock the state bits for the whole range */
-	lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
+	lock_extent_bits(tree, delalloc_start, delalloc_end,
+			 0, &cached_state, GFP_NOFS);
 
 	/* then test to make sure it is all still delalloc */
 	ret = test_range_bit(tree, delalloc_start, delalloc_end,
-			     EXTENT_DELALLOC, 1);
+			     EXTENT_DELALLOC, 1, cached_state);
 	if (!ret) {
-		unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
+		unlock_extent_cached(tree, delalloc_start, delalloc_end,
+				     &cached_state, GFP_NOFS);
 		__unlock_for_delalloc(inode, locked_page,
 			      delalloc_start, delalloc_end);
 		cond_resched();
 		goto again;
 	}
+	free_extent_state(cached_state);
 	*start = delalloc_start;
 	*end = delalloc_end;
 out_failed:
@@ -1307,7 +1359,8 @@ int extent_clear_unlock_delalloc(struct inode *inode,
 				int clear_unlock,
 				int clear_delalloc, int clear_dirty,
 				int set_writeback,
-				int end_writeback)
+				int end_writeback,
+				int set_private2)
 {
 	int ret;
 	struct page *pages[16];
@@ -1325,8 +1378,9 @@ int extent_clear_unlock_delalloc(struct inode *inode,
 	if (clear_delalloc)
 		clear_bits |= EXTENT_DELALLOC;
 
-	clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
-	if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
+	clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
+	if (!(unlock_pages || clear_dirty || set_writeback || end_writeback ||
+	      set_private2))
 		return 0;
 
 	while (nr_pages > 0) {
@@ -1334,6 +1388,10 @@ int extent_clear_unlock_delalloc(struct inode *inode,
 				     min_t(unsigned long,
 				     nr_pages, ARRAY_SIZE(pages)), pages);
 		for (i = 0; i < ret; i++) {
+
+			if (set_private2)
+				SetPagePrivate2(pages[i]);
+
 			if (pages[i] == locked_page) {
 				page_cache_release(pages[i]);
 				continue;
@@ -1476,14 +1534,17 @@ out:
  * range is found set.
  */
 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
-		   int bits, int filled)
+		   int bits, int filled, struct extent_state *cached)
 {
 	struct extent_state *state = NULL;
 	struct rb_node *node;
 	int bitset = 0;
 
 	spin_lock(&tree->lock);
-	node = tree_search(tree, start);
+	if (cached && cached->tree && cached->start == start)
+		node = &cached->rb_node;
+	else
+		node = tree_search(tree, start);
 	while (node && start <= end) {
 		state = rb_entry(node, struct extent_state, rb_node);
 
@@ -1503,6 +1564,10 @@ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
 			bitset = 0;
 			break;
 		}
+
+		if (state->end == (u64)-1)
+			break;
+
 		start = state->end + 1;
 		if (start > end)
 			break;
@@ -1526,7 +1591,7 @@ static int check_page_uptodate(struct extent_io_tree *tree,
 {
 	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 	u64 end = start + PAGE_CACHE_SIZE - 1;
-	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
+	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
 		SetPageUptodate(page);
 	return 0;
 }
@@ -1540,7 +1605,7 @@ static int check_page_locked(struct extent_io_tree *tree,
 {
 	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 	u64 end = start + PAGE_CACHE_SIZE - 1;
-	if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
+	if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
 		unlock_page(page);
 	return 0;
 }
@@ -1552,10 +1617,7 @@ static int check_page_locked(struct extent_io_tree *tree,
 static int check_page_writeback(struct extent_io_tree *tree,
 			     struct page *page)
 {
-	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
-	u64 end = start + PAGE_CACHE_SIZE - 1;
-	if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
-		end_page_writeback(page);
+	end_page_writeback(page);
 	return 0;
 }
 
@@ -1613,13 +1675,11 @@ static void end_bio_extent_writepage(struct bio *bio, int err)
 		}
 
 		if (!uptodate) {
-			clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
+			clear_extent_uptodate(tree, start, end, GFP_NOFS);
 			ClearPageUptodate(page);
 			SetPageError(page);
 		}
 
-		clear_extent_writeback(tree, start, end, GFP_ATOMIC);
-
 		if (whole_page)
 			end_page_writeback(page);
 		else
@@ -1983,7 +2043,8 @@ static int __extent_read_full_page(struct extent_io_tree *tree,
 			continue;
 		}
 		/* the get_extent function already copied into the page */
-		if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
+		if (test_range_bit(tree, cur, cur_end,
+				   EXTENT_UPTODATE, 1, NULL)) {
 			check_page_uptodate(tree, page);
 			unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
 			cur = cur + iosize;
@@ -2078,6 +2139,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 	u64 iosize;
 	u64 unlock_start;
 	sector_t sector;
+	struct extent_state *cached_state = NULL;
 	struct extent_map *em;
 	struct block_device *bdev;
 	int ret;
@@ -2124,6 +2186,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 	delalloc_end = 0;
 	page_started = 0;
 	if (!epd->extent_locked) {
+		u64 delalloc_to_write = 0;
 		/*
 		 * make sure the wbc mapping index is at least updated
 		 * to this page.
@@ -2143,8 +2206,24 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 			tree->ops->fill_delalloc(inode, page, delalloc_start,
 						 delalloc_end, &page_started,
 						 &nr_written);
+			/*
+			 * delalloc_end is already one less than the total
+			 * length, so we don't subtract one from
+			 * PAGE_CACHE_SIZE
+			 */
+			delalloc_to_write += (delalloc_end - delalloc_start +
+					      PAGE_CACHE_SIZE) >>
+					      PAGE_CACHE_SHIFT;
 			delalloc_start = delalloc_end + 1;
 		}
+		if (wbc->nr_to_write < delalloc_to_write) {
+			int thresh = 8192;
+
+			if (delalloc_to_write < thresh * 2)
+				thresh = delalloc_to_write;
+			wbc->nr_to_write = min_t(u64, delalloc_to_write,
+						 thresh);
+		}
 
 		/* did the fill delalloc function already unlock and start
 		 * the IO?
@@ -2160,15 +2239,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 			goto done_unlocked;
 		}
 	}
-	lock_extent(tree, start, page_end, GFP_NOFS);
-
-	unlock_start = start;
-
 	if (tree->ops && tree->ops->writepage_start_hook) {
 		ret = tree->ops->writepage_start_hook(page, start,
 						      page_end);
 		if (ret == -EAGAIN) {
-			unlock_extent(tree, start, page_end, GFP_NOFS);
 			redirty_page_for_writepage(wbc, page);
 			update_nr_written(page, wbc, nr_written);
 			unlock_page(page);
@@ -2184,12 +2258,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 	update_nr_written(page, wbc, nr_written + 1);
 
 	end = page_end;
-	if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
-		printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
-
 	if (last_byte <= start) {
-		clear_extent_dirty(tree, start, page_end, GFP_NOFS);
-		unlock_extent(tree, start, page_end, GFP_NOFS);
 		if (tree->ops && tree->ops->writepage_end_io_hook)
 			tree->ops->writepage_end_io_hook(page, start,
 							 page_end, NULL, 1);
@@ -2197,13 +2266,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 		goto done;
 	}
 
-	set_extent_uptodate(tree, start, page_end, GFP_NOFS);
 	blocksize = inode->i_sb->s_blocksize;
 
 	while (cur <= end) {
 		if (cur >= last_byte) {
-			clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
-			unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
 			if (tree->ops && tree->ops->writepage_end_io_hook)
 				tree->ops->writepage_end_io_hook(page, cur,
 							 page_end, NULL, 1);
@@ -2235,12 +2301,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 		 */
 		if (compressed || block_start == EXTENT_MAP_HOLE ||
 		    block_start == EXTENT_MAP_INLINE) {
-			clear_extent_dirty(tree, cur,
-					   cur + iosize - 1, GFP_NOFS);
-
-			unlock_extent(tree, unlock_start, cur + iosize - 1,
-				      GFP_NOFS);
-
 			/*
 			 * end_io notification does not happen here for
 			 * compressed extents
@@ -2265,13 +2325,12 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 		}
 		/* leave this out until we have a page_mkwrite call */
 		if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
-				   EXTENT_DIRTY, 0)) {
+				   EXTENT_DIRTY, 0, NULL)) {
 			cur = cur + iosize;
 			pg_offset += iosize;
 			continue;
 		}
 
-		clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
 		if (tree->ops && tree->ops->writepage_io_hook) {
 			ret = tree->ops->writepage_io_hook(page, cur,
 						cur + iosize - 1);
@@ -2309,12 +2368,12 @@ done:
 		set_page_writeback(page);
 		end_page_writeback(page);
 	}
-	if (unlock_start <= page_end)
-		unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
 	unlock_page(page);
 
 done_unlocked:
 
+	/* drop our reference on any cached states */
+	free_extent_state(cached_state);
 	return 0;
 }
 
@@ -2339,9 +2398,9 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
 			     writepage_t writepage, void *data,
 			     void (*flush_fn)(void *))
 {
-	struct backing_dev_info *bdi = mapping->backing_dev_info;
 	int ret = 0;
 	int done = 0;
+	int nr_to_write_done = 0;
 	struct pagevec pvec;
 	int nr_pages;
 	pgoff_t index;
@@ -2361,7 +2420,7 @@ static int extent_write_cache_pages(struct extent_io_tree *tree,
 		scanned = 1;
 	}
 retry:
-	while (!done && (index <= end) &&
+	while (!done && !nr_to_write_done && (index <= end) &&
 	       (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 			      PAGECACHE_TAG_DIRTY, min(end - index,
 				  (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
@@ -2412,12 +2471,15 @@ retry:
 				unlock_page(page);
 				ret = 0;
 			}
-			if (ret || wbc->nr_to_write <= 0)
-				done = 1;
-			if (wbc->nonblocking && bdi_write_congested(bdi)) {
-				wbc->encountered_congestion = 1;
+			if (ret)
 				done = 1;
-			}
+
+			/*
+			 * the filesystem may choose to bump up nr_to_write.
+			 * We have to make sure to honor the new nr_to_write
+			 * at any time
+			 */
+			nr_to_write_done = wbc->nr_to_write <= 0;
 		}
 		pagevec_release(&pvec);
 		cond_resched();
@@ -2604,10 +2666,10 @@ int extent_invalidatepage(struct extent_io_tree *tree,
 		return 0;
 
 	lock_extent(tree, start, end, GFP_NOFS);
-	wait_on_extent_writeback(tree, start, end);
+	wait_on_page_writeback(page);
 	clear_extent_bit(tree, start, end,
 			 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
-			 1, 1, GFP_NOFS);
+			 1, 1, NULL, GFP_NOFS);
 	return 0;
 }
 
@@ -2687,7 +2749,7 @@ int extent_prepare_write(struct extent_io_tree *tree,
 		    !isnew && !PageUptodate(page) &&
 		    (block_off_end > to || block_off_start < from) &&
 		    !test_range_bit(tree, block_start, cur_end,
-				    EXTENT_UPTODATE, 1)) {
+				    EXTENT_UPTODATE, 1, NULL)) {
 			u64 sector;
 			u64 extent_offset = block_start - em->start;
 			size_t iosize;
@@ -2701,7 +2763,7 @@ int extent_prepare_write(struct extent_io_tree *tree,
 			 */
 			set_extent_bit(tree, block_start,
 				       block_start + iosize - 1,
-				       EXTENT_LOCKED, 0, NULL, GFP_NOFS);
+				       EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS);
 			ret = submit_extent_page(READ, tree, page,
 					 sector, iosize, page_offset, em->bdev,
 					 NULL, 1,
@@ -2742,13 +2804,18 @@ int try_release_extent_state(struct extent_map_tree *map,
 	int ret = 1;
 
 	if (test_range_bit(tree, start, end,
-			   EXTENT_IOBITS | EXTENT_ORDERED, 0))
+			   EXTENT_IOBITS, 0, NULL))
 		ret = 0;
 	else {
 		if ((mask & GFP_NOFS) == GFP_NOFS)
 			mask = GFP_NOFS;
-		clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
-				 1, 1, mask);
+		/*
+		 * at this point we can safely clear everything except the
+		 * locked bit and the nodatasum bit
+		 */
+		clear_extent_bit(tree, start, end,
+				 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
+				 0, 0, NULL, mask);
 	}
 	return ret;
 }
@@ -2771,29 +2838,28 @@ int try_release_extent_mapping(struct extent_map_tree *map,
 		u64 len;
 		while (start <= end) {
 			len = end - start + 1;
-			spin_lock(&map->lock);
+			write_lock(&map->lock);
 			em = lookup_extent_mapping(map, start, len);
 			if (!em || IS_ERR(em)) {
-				spin_unlock(&map->lock);
+				write_unlock(&map->lock);
 				break;
 			}
 			if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
 			    em->start != start) {
-				spin_unlock(&map->lock);
+				write_unlock(&map->lock);
 				free_extent_map(em);
 				break;
 			}
 			if (!test_range_bit(tree, em->start,
 					    extent_map_end(em) - 1,
-					    EXTENT_LOCKED | EXTENT_WRITEBACK |
-					    EXTENT_ORDERED,
-					    0)) {
+					    EXTENT_LOCKED | EXTENT_WRITEBACK,
+					    0, NULL)) {
 				remove_extent_mapping(map, em);
 				/* once for the rb tree */
 				free_extent_map(em);
 			}
 			start = extent_map_end(em);
-			spin_unlock(&map->lock);
+			write_unlock(&map->lock);
 
 			/* once for us */
 			free_extent_map(em);
@@ -3203,7 +3269,7 @@ int extent_range_uptodate(struct extent_io_tree *tree,
 	int uptodate;
 	unsigned long index;
 
-	ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
+	ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
 	if (ret)
 		return 1;
 	while (start <= end) {
@@ -3233,7 +3299,7 @@ int extent_buffer_uptodate(struct extent_io_tree *tree,
 		return 1;
 
 	ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
-			   EXTENT_UPTODATE, 1);
+			   EXTENT_UPTODATE, 1, NULL);
 	if (ret)
 		return ret;
 
@@ -3269,7 +3335,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 		return 0;
 
 	if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
-			   EXTENT_UPTODATE, 1)) {
+			   EXTENT_UPTODATE, 1, NULL)) {
 		return 0;
 	}
 

fs/btrfs/extent_io.h

@@ -13,10 +13,8 @@
 #define EXTENT_DEFRAG (1 << 6)
 #define EXTENT_DEFRAG_DONE (1 << 7)
 #define EXTENT_BUFFER_FILLED (1 << 8)
-#define EXTENT_ORDERED (1 << 9)
-#define EXTENT_ORDERED_METADATA (1 << 10)
-#define EXTENT_BOUNDARY (1 << 11)
-#define EXTENT_NODATASUM (1 << 12)
+#define EXTENT_BOUNDARY (1 << 9)
+#define EXTENT_NODATASUM (1 << 10)
 #define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
 
 /* flags for bio submission */
@@ -142,6 +140,8 @@ int try_release_extent_state(struct extent_map_tree *map,
 			     struct extent_io_tree *tree, struct page *page,
 			     gfp_t mask);
 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask);
+int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
+		     int bits, struct extent_state **cached, gfp_t mask);
 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask);
 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 		    gfp_t mask);
@@ -155,11 +155,12 @@ u64 count_range_bits(struct extent_io_tree *tree,
 		     u64 max_bytes, unsigned long bits);
 
 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
-		   int bits, int filled);
+		   int bits, int filled, struct extent_state *cached_state);
 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 		      int bits, gfp_t mask);
 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
-		     int bits, int wake, int delete, gfp_t mask);
+		     int bits, int wake, int delete, struct extent_state **cached,
+		     gfp_t mask);
 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 		    int bits, gfp_t mask);
 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
@@ -282,5 +283,6 @@ int extent_clear_unlock_delalloc(struct inode *inode,
 				int clear_unlock,
 				int clear_delalloc, int clear_dirty,
 				int set_writeback,
-				int end_writeback);
+				int end_writeback,
+				int set_private2);
 #endif

fs/btrfs/extent_map.c

@@ -36,7 +36,7 @@ void extent_map_exit(void)
 void extent_map_tree_init(struct extent_map_tree *tree, gfp_t mask)
 {
 	tree->map.rb_node = NULL;
-	spin_lock_init(&tree->lock);
+	rwlock_init(&tree->lock);
 }
 
 /**
@@ -198,6 +198,56 @@ static int mergable_maps(struct extent_map *prev, struct extent_map *next)
 	return 0;
 }
 
+int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len)
+{
+	int ret = 0;
+	struct extent_map *merge = NULL;
+	struct rb_node *rb;
+	struct extent_map *em;
+
+	write_lock(&tree->lock);
+	em = lookup_extent_mapping(tree, start, len);
+
+	WARN_ON(em->start != start || !em);
+
+	if (!em)
+		goto out;
+
+	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+	if (em->start != 0) {
+		rb = rb_prev(&em->rb_node);
+		if (rb)
+			merge = rb_entry(rb, struct extent_map, rb_node);
+		if (rb && mergable_maps(merge, em)) {
+			em->start = merge->start;
+			em->len += merge->len;
+			em->block_len += merge->block_len;
+			em->block_start = merge->block_start;
+			merge->in_tree = 0;
+			rb_erase(&merge->rb_node, &tree->map);
+			free_extent_map(merge);
+		}
+	}
+
+	rb = rb_next(&em->rb_node);
+	if (rb)
+		merge = rb_entry(rb, struct extent_map, rb_node);
+	if (rb && mergable_maps(em, merge)) {
+		em->len += merge->len;
+		em->block_len += merge->len;
+		rb_erase(&merge->rb_node, &tree->map);
+		merge->in_tree = 0;
+		free_extent_map(merge);
+	}
+
+	free_extent_map(em);
+out:
+	write_unlock(&tree->lock);
+	return ret;
+
+}
+
 /**
  * add_extent_mapping - add new extent map to the extent tree
  * @tree:	tree to insert new map in
@@ -222,7 +272,6 @@ int add_extent_mapping(struct extent_map_tree *tree,
 		ret = -EEXIST;
 		goto out;
 	}
-	assert_spin_locked(&tree->lock);
 	rb = tree_insert(&tree->map, em->start, &em->rb_node);
 	if (rb) {
 		ret = -EEXIST;
@@ -285,7 +334,6 @@ struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
 	struct rb_node *next = NULL;
 	u64 end = range_end(start, len);
 
-	assert_spin_locked(&tree->lock);
 	rb_node = __tree_search(&tree->map, start, &prev, &next);
 	if (!rb_node && prev) {
 		em = rb_entry(prev, struct extent_map, rb_node);
@@ -318,6 +366,54 @@ out:
 	return em;
 }
 
+/**
+ * search_extent_mapping - find a nearby extent map
+ * @tree:	tree to lookup in
+ * @start:	byte offset to start the search
+ * @len:	length of the lookup range
+ *
+ * Find and return the first extent_map struct in @tree that intersects the
+ * [start, len] range.
+ *
+ * If one can't be found, any nearby extent may be returned
+ */
+struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
+					 u64 start, u64 len)
+{
+	struct extent_map *em;
+	struct rb_node *rb_node;
+	struct rb_node *prev = NULL;
+	struct rb_node *next = NULL;
+
+	rb_node = __tree_search(&tree->map, start, &prev, &next);
+	if (!rb_node && prev) {
+		em = rb_entry(prev, struct extent_map, rb_node);
+		goto found;
+	}
+	if (!rb_node && next) {
+		em = rb_entry(next, struct extent_map, rb_node);
+		goto found;
+	}
+	if (!rb_node) {
+		em = NULL;
+		goto out;
+	}
+	if (IS_ERR(rb_node)) {
+		em = ERR_PTR(PTR_ERR(rb_node));
+		goto out;
+	}
+	em = rb_entry(rb_node, struct extent_map, rb_node);
+	goto found;
+
+	em = NULL;
+	goto out;
+
+found:
+	atomic_inc(&em->refs);
+out:
+	return em;
+}
+
 /**
  * remove_extent_mapping - removes an extent_map from the extent tree
  * @tree:	extent tree to remove from
@@ -331,7 +427,6 @@ int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
 	int ret = 0;
 
 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
-	assert_spin_locked(&tree->lock);
 	rb_erase(&em->rb_node, &tree->map);
 	em->in_tree = 0;
 	return ret;

fs/btrfs/extent_map.h

@@ -31,7 +31,7 @@ struct extent_map {
 
 struct extent_map_tree {
 	struct rb_root map;
-	spinlock_t lock;
+	rwlock_t lock;
 };
 
 static inline u64 extent_map_end(struct extent_map *em)
@@ -59,4 +59,7 @@ struct extent_map *alloc_extent_map(gfp_t mask);
 void free_extent_map(struct extent_map *em);
 int __init extent_map_init(void);
 void extent_map_exit(void);
+int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len);
+struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
+					 u64 start, u64 len);
 #endif

fs/btrfs/file.c

@@ -112,8 +112,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 	int err = 0;
 	int i;
 	struct inode *inode = fdentry(file)->d_inode;
-	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
-	u64 hint_byte;
 	u64 num_bytes;
 	u64 start_pos;
 	u64 end_of_last_block;
@@ -125,22 +123,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 		    root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 
 	end_of_last_block = start_pos + num_bytes - 1;
-
-	lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
-	trans = btrfs_join_transaction(root, 1);
-	if (!trans) {
-		err = -ENOMEM;
-		goto out_unlock;
-	}
-	btrfs_set_trans_block_group(trans, inode);
-	hint_byte = 0;
-
-	set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
-
-	/* check for reserved extents on each page, we don't want
-	 * to reset the delalloc bit on things that already have
-	 * extents reserved.
-	 */
 	btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 	for (i = 0; i < num_pages; i++) {
 		struct page *p = pages[i];
@@ -155,9 +137,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 		 * at this time.
 		 */
 	}
-	err = btrfs_end_transaction(trans, root);
-out_unlock:
-	unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 	return err;
 }
 
@@ -189,18 +168,18 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 		if (!split2)
 			split2 = alloc_extent_map(GFP_NOFS);
 
-		spin_lock(&em_tree->lock);
+		write_lock(&em_tree->lock);
 		em = lookup_extent_mapping(em_tree, start, len);
 		if (!em) {
-			spin_unlock(&em_tree->lock);
+			write_unlock(&em_tree->lock);
 			break;
 		}
 		flags = em->flags;
 		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
-			spin_unlock(&em_tree->lock);
 			if (em->start <= start &&
 			    (!testend || em->start + em->len >= start + len)) {
 				free_extent_map(em);
+				write_unlock(&em_tree->lock);
 				break;
 			}
 			if (start < em->start) {
@@ -210,6 +189,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 				start = em->start + em->len;
 			}
 			free_extent_map(em);
+			write_unlock(&em_tree->lock);
 			continue;
 		}
 		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
@@ -260,7 +240,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 			free_extent_map(split);
 			split = NULL;
 		}
-		spin_unlock(&em_tree->lock);
+		write_unlock(&em_tree->lock);
 
 		/* once for us */
 		free_extent_map(em);
@@ -289,7 +269,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root, struct inode *inode,
 		       u64 start, u64 end, u64 locked_end,
-		       u64 inline_limit, u64 *hint_byte)
+		       u64 inline_limit, u64 *hint_byte, int drop_cache)
 {
 	u64 extent_end = 0;
 	u64 search_start = start;
@@ -314,7 +294,8 @@ noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 	int ret;
 
 	inline_limit = 0;
-	btrfs_drop_extent_cache(inode, start, end - 1, 0);
+	if (drop_cache)
+		btrfs_drop_extent_cache(inode, start, end - 1, 0);
 
 	path = btrfs_alloc_path();
 	if (!path)

fs/btrfs/free-space-cache.c

@@ -259,7 +259,9 @@ static int link_free_space(struct btrfs_block_group_cache *block_group,
 
 static void recalculate_thresholds(struct btrfs_block_group_cache *block_group)
 {
-	u64 max_bytes, possible_bytes;
+	u64 max_bytes;
+	u64 bitmap_bytes;
+	u64 extent_bytes;
 
 	/*
 	 * The goal is to keep the total amount of memory used per 1gb of space
@@ -269,22 +271,27 @@ static void recalculate_thresholds(struct btrfs_block_group_cache *block_group)
 	max_bytes = MAX_CACHE_BYTES_PER_GIG *
 		(div64_u64(block_group->key.offset, 1024 * 1024 * 1024));
 
-	possible_bytes = (block_group->total_bitmaps * PAGE_CACHE_SIZE) +
-		(sizeof(struct btrfs_free_space) *
-		 block_group->extents_thresh);
+	/*
+	 * we want to account for 1 more bitmap than what we have so we can make
+	 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
+	 * we add more bitmaps.
+	 */
+	bitmap_bytes = (block_group->total_bitmaps + 1) * PAGE_CACHE_SIZE;
 
-	if (possible_bytes > max_bytes) {
-		int extent_bytes = max_bytes -
-			(block_group->total_bitmaps * PAGE_CACHE_SIZE);
+	if (bitmap_bytes >= max_bytes) {
+		block_group->extents_thresh = 0;
+		return;
+	}
 
-		if (extent_bytes <= 0) {
-			block_group->extents_thresh = 0;
-			return;
-		}
+	/*
+	 * we want the extent entry threshold to always be at most 1/2 the maxw
+	 * bytes we can have, or whatever is less than that.
+	 */
+	extent_bytes = max_bytes - bitmap_bytes;
+	extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
 
-		block_group->extents_thresh = extent_bytes /
-			(sizeof(struct btrfs_free_space));
-	}
+	block_group->extents_thresh =
+		div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
 }
 
 static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group,
@@ -403,6 +410,7 @@ static void add_new_bitmap(struct btrfs_block_group_cache *block_group,
 	BUG_ON(block_group->total_bitmaps >= max_bitmaps);
 
 	info->offset = offset_to_bitmap(block_group, offset);
+	info->bytes = 0;
 	link_free_space(block_group, info);
 	block_group->total_bitmaps++;
 

fs/btrfs/inode-item.c

@@ -149,6 +149,8 @@ int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,
 		ptr = (unsigned long)(ref + 1);
 		ret = 0;
 	} else if (ret < 0) {
+		if (ret == -EOVERFLOW)
+			ret = -EMLINK;
 		goto out;
 	} else {
 		ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
@@ -177,8 +179,6 @@ int btrfs_insert_empty_inode(struct btrfs_trans_handle *trans,
 
 	ret = btrfs_insert_empty_item(trans, root, path, &key,
 				      sizeof(struct btrfs_inode_item));
-	if (ret == 0 && objectid > root->highest_inode)
-		root->highest_inode = objectid;
 	return ret;
 }
 

fs/btrfs/inode-map.c

@@ -43,9 +43,10 @@ int btrfs_find_highest_inode(struct btrfs_root *root, u64 *objectid)
 		slot = path->slots[0] - 1;
 		l = path->nodes[0];
 		btrfs_item_key_to_cpu(l, &found_key, slot);
-		*objectid = found_key.objectid;
+		*objectid = max_t(u64, found_key.objectid,
+				  BTRFS_FIRST_FREE_OBJECTID - 1);
 	} else {
-		*objectid = BTRFS_FIRST_FREE_OBJECTID;
+		*objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
 	}
 	ret = 0;
 error:
@@ -53,91 +54,27 @@ error:
 	return ret;
 }
 
-/*
- * walks the btree of allocated inodes and find a hole.
- */
 int btrfs_find_free_objectid(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root,
 			     u64 dirid, u64 *objectid)
 {
-	struct btrfs_path *path;
-	struct btrfs_key key;
 	int ret;
-	int slot = 0;
-	u64 last_ino = 0;
-	int start_found;
-	struct extent_buffer *l;
-	struct btrfs_key search_key;
-	u64 search_start = dirid;
-
 	mutex_lock(&root->objectid_mutex);
-	if (root->last_inode_alloc >= BTRFS_FIRST_FREE_OBJECTID &&
-	    root->last_inode_alloc < BTRFS_LAST_FREE_OBJECTID) {
-		*objectid = ++root->last_inode_alloc;
-		mutex_unlock(&root->objectid_mutex);
-		return 0;
-	}
-	path = btrfs_alloc_path();
-	BUG_ON(!path);
-	search_start = max(search_start, (u64)BTRFS_FIRST_FREE_OBJECTID);
-	search_key.objectid = search_start;
-	search_key.type = 0;
-	search_key.offset = 0;
-
-	start_found = 0;
-	ret = btrfs_search_slot(trans, root, &search_key, path, 0, 0);
-	if (ret < 0)
-		goto error;
 
-	while (1) {
-		l = path->nodes[0];
-		slot = path->slots[0];
-		if (slot >= btrfs_header_nritems(l)) {
-			ret = btrfs_next_leaf(root, path);
-			if (ret == 0)
-				continue;
-			if (ret < 0)
-				goto error;
-			if (!start_found) {
-				*objectid = search_start;
-				start_found = 1;
-				goto found;
-			}
-			*objectid = last_ino > search_start ?
-				last_ino : search_start;
-			goto found;
-		}
-		btrfs_item_key_to_cpu(l, &key, slot);
-		if (key.objectid >= search_start) {
-			if (start_found) {
-				if (last_ino < search_start)
-					last_ino = search_start;
-				if (key.objectid > last_ino) {
-					*objectid = last_ino;
-					goto found;
-				}
-			} else if (key.objectid > search_start) {
-				*objectid = search_start;
-				goto found;
-			}
-		}
-		if (key.objectid >= BTRFS_LAST_FREE_OBJECTID)
-			break;
+	if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
+		ret = btrfs_find_highest_inode(root, &root->highest_objectid);
+		if (ret)
+			goto out;
+	}
 
-		start_found = 1;
-		last_ino = key.objectid + 1;
-		path->slots[0]++;
+	if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
+		ret = -ENOSPC;
+		goto out;
 	}
-	BUG_ON(1);
-found:
-	btrfs_release_path(root, path);
-	btrfs_free_path(path);
-	BUG_ON(*objectid < search_start);
-	mutex_unlock(&root->objectid_mutex);
-	return 0;
-error:
-	btrfs_release_path(root, path);
-	btrfs_free_path(path);
+
+	*objectid = ++root->highest_objectid;
+	ret = 0;
+out:
 	mutex_unlock(&root->objectid_mutex);
 	return ret;
 }

fs/btrfs/inode.c

@@ -231,7 +231,8 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 	}
 
 	ret = btrfs_drop_extents(trans, root, inode, start,
-				 aligned_end, aligned_end, start, &hint_byte);
+				 aligned_end, aligned_end, start,
+				 &hint_byte, 1);
 	BUG_ON(ret);
 
 	if (isize > actual_end)
@@ -240,7 +241,7 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 				   inline_len, compressed_size,
 				   compressed_pages);
 	BUG_ON(ret);
-	btrfs_drop_extent_cache(inode, start, aligned_end, 0);
+	btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 	return 0;
 }
 
@@ -425,7 +426,7 @@ again:
 			extent_clear_unlock_delalloc(inode,
 						     &BTRFS_I(inode)->io_tree,
 						     start, end, NULL, 1, 0,
-						     0, 1, 1, 1);
+						     0, 1, 1, 1, 0);
 			ret = 0;
 			goto free_pages_out;
 		}
@@ -611,9 +612,9 @@ static noinline int submit_compressed_extents(struct inode *inode,
 		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 
 		while (1) {
-			spin_lock(&em_tree->lock);
+			write_lock(&em_tree->lock);
 			ret = add_extent_mapping(em_tree, em);
-			spin_unlock(&em_tree->lock);
+			write_unlock(&em_tree->lock);
 			if (ret != -EEXIST) {
 				free_extent_map(em);
 				break;
@@ -640,7 +641,7 @@ static noinline int submit_compressed_extents(struct inode *inode,
 					     async_extent->start,
 					     async_extent->start +
 					     async_extent->ram_size - 1,
-					     NULL, 1, 1, 0, 1, 1, 0);
+					     NULL, 1, 1, 0, 1, 1, 0, 0);
 
 		ret = btrfs_submit_compressed_write(inode,
 				    async_extent->start,
@@ -713,7 +714,7 @@ static noinline int cow_file_range(struct inode *inode,
 			extent_clear_unlock_delalloc(inode,
 						     &BTRFS_I(inode)->io_tree,
 						     start, end, NULL, 1, 1,
-						     1, 1, 1, 1);
+						     1, 1, 1, 1, 0);
 			*nr_written = *nr_written +
 			     (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 			*page_started = 1;
@@ -725,6 +726,15 @@ static noinline int cow_file_range(struct inode *inode,
 	BUG_ON(disk_num_bytes >
 	       btrfs_super_total_bytes(&root->fs_info->super_copy));
 
+
+	read_lock(&BTRFS_I(inode)->extent_tree.lock);
+	em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
+				   start, num_bytes);
+	if (em) {
+		alloc_hint = em->block_start;
+		free_extent_map(em);
+	}
+	read_unlock(&BTRFS_I(inode)->extent_tree.lock);
 	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 
 	while (disk_num_bytes > 0) {
@@ -737,7 +747,6 @@ static noinline int cow_file_range(struct inode *inode,
 		em = alloc_extent_map(GFP_NOFS);
 		em->start = start;
 		em->orig_start = em->start;
-
 		ram_size = ins.offset;
 		em->len = ins.offset;
 
@@ -747,9 +756,9 @@ static noinline int cow_file_range(struct inode *inode,
 		set_bit(EXTENT_FLAG_PINNED, &em->flags);
 
 		while (1) {
-			spin_lock(&em_tree->lock);
+			write_lock(&em_tree->lock);
 			ret = add_extent_mapping(em_tree, em);
-			spin_unlock(&em_tree->lock);
+			write_unlock(&em_tree->lock);
 			if (ret != -EEXIST) {
 				free_extent_map(em);
 				break;
@@ -776,11 +785,14 @@ static noinline int cow_file_range(struct inode *inode,
 		/* we're not doing compressed IO, don't unlock the first
 		 * page (which the caller expects to stay locked), don't
 		 * clear any dirty bits and don't set any writeback bits
+		 *
+		 * Do set the Private2 bit so we know this page was properly
+		 * setup for writepage
 		 */
 		extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 					     start, start + ram_size - 1,
 					     locked_page, unlock, 1,
-					     1, 0, 0, 0);
+					     1, 0, 0, 0, 1);
 		disk_num_bytes -= cur_alloc_size;
 		num_bytes -= cur_alloc_size;
 		alloc_hint = ins.objectid + ins.offset;
@@ -853,7 +865,7 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 	int limit = 10 * 1024 * 1042;
 
 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
-			 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
+			 EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS);
 	while (start < end) {
 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 		async_cow->inode = inode;
@@ -1080,9 +1092,9 @@ out_check:
 			em->bdev = root->fs_info->fs_devices->latest_bdev;
 			set_bit(EXTENT_FLAG_PINNED, &em->flags);
 			while (1) {
-				spin_lock(&em_tree->lock);
+				write_lock(&em_tree->lock);
 				ret = add_extent_mapping(em_tree, em);
-				spin_unlock(&em_tree->lock);
+				write_unlock(&em_tree->lock);
 				if (ret != -EEXIST) {
 					free_extent_map(em);
 					break;
@@ -1101,7 +1113,7 @@ out_check:
 
 		extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 					cur_offset, cur_offset + num_bytes - 1,
-					locked_page, 1, 1, 1, 0, 0, 0);
+					locked_page, 1, 1, 1, 0, 0, 0, 1);
 		cur_offset = extent_end;
 		if (cur_offset > end)
 			break;
@@ -1374,10 +1386,8 @@ again:
 	lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 
 	/* already ordered? We're done */
-	if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
-			     EXTENT_ORDERED, 0)) {
+	if (PagePrivate2(page))
 		goto out;
-	}
 
 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
 	if (ordered) {
@@ -1413,11 +1423,9 @@ static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
 	struct inode *inode = page->mapping->host;
 	struct btrfs_writepage_fixup *fixup;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
-	int ret;
 
-	ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
-			     EXTENT_ORDERED, 0);
-	if (ret)
+	/* this page is properly in the ordered list */
+	if (TestClearPagePrivate2(page))
 		return 0;
 
 	if (PageChecked(page))
@@ -1455,9 +1463,19 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	BUG_ON(!path);
 
 	path->leave_spinning = 1;
+
+	/*
+	 * we may be replacing one extent in the tree with another.
+	 * The new extent is pinned in the extent map, and we don't want
+	 * to drop it from the cache until it is completely in the btree.
+	 *
+	 * So, tell btrfs_drop_extents to leave this extent in the cache.
+	 * the caller is expected to unpin it and allow it to be merged
+	 * with the others.
+	 */
 	ret = btrfs_drop_extents(trans, root, inode, file_pos,
 				 file_pos + num_bytes, locked_end,
-				 file_pos, &hint);
+				 file_pos, &hint, 0);
 	BUG_ON(ret);
 
 	ins.objectid = inode->i_ino;
@@ -1485,7 +1503,6 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(leaf);
 
 	inode_add_bytes(inode, num_bytes);
-	btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
 
 	ins.objectid = disk_bytenr;
 	ins.offset = disk_num_bytes;
@@ -1596,6 +1613,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 						ordered_extent->len,
 						compressed, 0, 0,
 						BTRFS_FILE_EXTENT_REG);
+		unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
+				   ordered_extent->file_offset,
+				   ordered_extent->len);
 		BUG_ON(ret);
 	}
 	unlock_extent(io_tree, ordered_extent->file_offset,
@@ -1623,6 +1643,7 @@ nocow:
 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 				struct extent_state *state, int uptodate)
 {
+	ClearPagePrivate2(page);
 	return btrfs_finish_ordered_io(page->mapping->host, start, end);
 }
 
@@ -1669,13 +1690,13 @@ static int btrfs_io_failed_hook(struct bio *failed_bio,
 		failrec->last_mirror = 0;
 		failrec->bio_flags = 0;
 
-		spin_lock(&em_tree->lock);
+		read_lock(&em_tree->lock);
 		em = lookup_extent_mapping(em_tree, start, failrec->len);
 		if (em->start > start || em->start + em->len < start) {
 			free_extent_map(em);
 			em = NULL;
 		}
-		spin_unlock(&em_tree->lock);
+		read_unlock(&em_tree->lock);
 
 		if (!em || IS_ERR(em)) {
 			kfree(failrec);
@@ -1794,7 +1815,7 @@ static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 		return 0;
 
 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
-	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
+	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
 				  GFP_NOFS);
 		return 0;
@@ -2352,6 +2373,69 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 	return ret;
 }
 
+int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct inode *dir, u64 objectid,
+			const char *name, int name_len)
+{
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_dir_item *di;
+	struct btrfs_key key;
+	u64 index;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
+				   name, name_len, -1);
+	BUG_ON(!di || IS_ERR(di));
+
+	leaf = path->nodes[0];
+	btrfs_dir_item_key_to_cpu(leaf, di, &key);
+	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
+	ret = btrfs_delete_one_dir_name(trans, root, path, di);
+	BUG_ON(ret);
+	btrfs_release_path(root, path);
+
+	ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
+				 objectid, root->root_key.objectid,
+				 dir->i_ino, &index, name, name_len);
+	if (ret < 0) {
+		BUG_ON(ret != -ENOENT);
+		di = btrfs_search_dir_index_item(root, path, dir->i_ino,
+						 name, name_len);
+		BUG_ON(!di || IS_ERR(di));
+
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		btrfs_release_path(root, path);
+		index = key.offset;
+	}
+
+	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
+					 index, name, name_len, -1);
+	BUG_ON(!di || IS_ERR(di));
+
+	leaf = path->nodes[0];
+	btrfs_dir_item_key_to_cpu(leaf, di, &key);
+	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
+	ret = btrfs_delete_one_dir_name(trans, root, path, di);
+	BUG_ON(ret);
+	btrfs_release_path(root, path);
+
+	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
+	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+	ret = btrfs_update_inode(trans, root, dir);
+	BUG_ON(ret);
+	dir->i_sb->s_dirt = 1;
+
+	btrfs_free_path(path);
+	return 0;
+}
+
 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 {
 	struct inode *inode = dentry->d_inode;
@@ -2361,29 +2445,31 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 	struct btrfs_trans_handle *trans;
 	unsigned long nr = 0;
 
-	/*
-	 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
-	 * the root of a subvolume or snapshot
-	 */
 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
-	    inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
+	    inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 		return -ENOTEMPTY;
-	}
 
 	trans = btrfs_start_transaction(root, 1);
 	btrfs_set_trans_block_group(trans, dir);
 
+	if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
+		err = btrfs_unlink_subvol(trans, root, dir,
+					  BTRFS_I(inode)->location.objectid,
+					  dentry->d_name.name,
+					  dentry->d_name.len);
+		goto out;
+	}
+
 	err = btrfs_orphan_add(trans, inode);
 	if (err)
-		goto fail_trans;
+		goto out;
 
 	/* now the directory is empty */
 	err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
 				 dentry->d_name.name, dentry->d_name.len);
 	if (!err)
 		btrfs_i_size_write(inode, 0);
-
-fail_trans:
+out:
 	nr = trans->blocks_used;
 	ret = btrfs_end_transaction_throttle(trans, root);
 	btrfs_btree_balance_dirty(root, nr);
@@ -2935,7 +3021,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
 						 cur_offset,
 						 cur_offset + hole_size,
 						 block_end,
-						 cur_offset, &hint_byte);
+						 cur_offset, &hint_byte, 1);
 			if (err)
 				break;
 			err = btrfs_insert_file_extent(trans, root,
@@ -3003,6 +3089,11 @@ void btrfs_delete_inode(struct inode *inode)
 	}
 	btrfs_wait_ordered_range(inode, 0, (u64)-1);
 
+	if (inode->i_nlink > 0) {
+		BUG_ON(btrfs_root_refs(&root->root_item) != 0);
+		goto no_delete;
+	}
+
 	btrfs_i_size_write(inode, 0);
 	trans = btrfs_join_transaction(root, 1);
 
@@ -3070,29 +3161,67 @@ out_err:
  * is kind of like crossing a mount point.
  */
 static int fixup_tree_root_location(struct btrfs_root *root,
-			     struct btrfs_key *location,
-			     struct btrfs_root **sub_root,
-			     struct dentry *dentry)
+				    struct inode *dir,
+				    struct dentry *dentry,
+				    struct btrfs_key *location,
+				    struct btrfs_root **sub_root)
 {
-	struct btrfs_root_item *ri;
+	struct btrfs_path *path;
+	struct btrfs_root *new_root;
+	struct btrfs_root_ref *ref;
+	struct extent_buffer *leaf;
+	int ret;
+	int err = 0;
 
-	if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
-		return 0;
-	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
-		return 0;
+	path = btrfs_alloc_path();
+	if (!path) {
+		err = -ENOMEM;
+		goto out;
+	}
 
-	*sub_root = btrfs_read_fs_root(root->fs_info, location,
-					dentry->d_name.name,
-					dentry->d_name.len);
-	if (IS_ERR(*sub_root))
-		return PTR_ERR(*sub_root);
+	err = -ENOENT;
+	ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
+				  BTRFS_I(dir)->root->root_key.objectid,
+				  location->objectid);
+	if (ret) {
+		if (ret < 0)
+			err = ret;
+		goto out;
+	}
 
-	ri = &(*sub_root)->root_item;
-	location->objectid = btrfs_root_dirid(ri);
-	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
-	location->offset = 0;
+	leaf = path->nodes[0];
+	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
+	if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
+	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
+		goto out;
 
-	return 0;
+	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
+				   (unsigned long)(ref + 1),
+				   dentry->d_name.len);
+	if (ret)
+		goto out;
+
+	btrfs_release_path(root->fs_info->tree_root, path);
+
+	new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
+	if (IS_ERR(new_root)) {
+		err = PTR_ERR(new_root);
+		goto out;
+	}
+
+	if (btrfs_root_refs(&new_root->root_item) == 0) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	*sub_root = new_root;
+	location->objectid = btrfs_root_dirid(&new_root->root_item);
+	location->type = BTRFS_INODE_ITEM_KEY;
+	location->offset = 0;
+	err = 0;
+out:
+	btrfs_free_path(path);
+	return err;
 }
 
 static void inode_tree_add(struct inode *inode)
@@ -3101,11 +3230,13 @@ static void inode_tree_add(struct inode *inode)
 	struct btrfs_inode *entry;
 	struct rb_node **p;
 	struct rb_node *parent;
-
 again:
 	p = &root->inode_tree.rb_node;
 	parent = NULL;
 
+	if (hlist_unhashed(&inode->i_hash))
+		return;
+
 	spin_lock(&root->inode_lock);
 	while (*p) {
 		parent = *p;
@@ -3132,13 +3263,87 @@ again:
 static void inode_tree_del(struct inode *inode)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
+	int empty = 0;
 
 	spin_lock(&root->inode_lock);
 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
+		empty = RB_EMPTY_ROOT(&root->inode_tree);
 	}
 	spin_unlock(&root->inode_lock);
+
+	if (empty && btrfs_root_refs(&root->root_item) == 0) {
+		synchronize_srcu(&root->fs_info->subvol_srcu);
+		spin_lock(&root->inode_lock);
+		empty = RB_EMPTY_ROOT(&root->inode_tree);
+		spin_unlock(&root->inode_lock);
+		if (empty)
+			btrfs_add_dead_root(root);
+	}
+}
+
+int btrfs_invalidate_inodes(struct btrfs_root *root)
+{
+	struct rb_node *node;
+	struct rb_node *prev;
+	struct btrfs_inode *entry;
+	struct inode *inode;
+	u64 objectid = 0;
+
+	WARN_ON(btrfs_root_refs(&root->root_item) != 0);
+
+	spin_lock(&root->inode_lock);
+again:
+	node = root->inode_tree.rb_node;
+	prev = NULL;
+	while (node) {
+		prev = node;
+		entry = rb_entry(node, struct btrfs_inode, rb_node);
+
+		if (objectid < entry->vfs_inode.i_ino)
+			node = node->rb_left;
+		else if (objectid > entry->vfs_inode.i_ino)
+			node = node->rb_right;
+		else
+			break;
+	}
+	if (!node) {
+		while (prev) {
+			entry = rb_entry(prev, struct btrfs_inode, rb_node);
+			if (objectid <= entry->vfs_inode.i_ino) {
+				node = prev;
+				break;
+			}
+			prev = rb_next(prev);
+		}
+	}
+	while (node) {
+		entry = rb_entry(node, struct btrfs_inode, rb_node);
+		objectid = entry->vfs_inode.i_ino + 1;
+		inode = igrab(&entry->vfs_inode);
+		if (inode) {
+			spin_unlock(&root->inode_lock);
+			if (atomic_read(&inode->i_count) > 1)
+				d_prune_aliases(inode);
+			/*
+			 * btrfs_drop_inode will remove it from
+			 * the inode cache when its usage count
+			 * hits zero.
+			 */
+			iput(inode);
+			cond_resched();
+			spin_lock(&root->inode_lock);
+			goto again;
+		}
+
+		if (cond_resched_lock(&root->inode_lock))
+			goto again;
+
+		node = rb_next(node);
+	}
+	spin_unlock(&root->inode_lock);
+	return 0;
 }
 
 static noinline void init_btrfs_i(struct inode *inode)
@@ -3225,15 +3430,41 @@ struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
 	return inode;
 }
 
+static struct inode *new_simple_dir(struct super_block *s,
+				    struct btrfs_key *key,
+				    struct btrfs_root *root)
+{
+	struct inode *inode = new_inode(s);
+
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+
+	init_btrfs_i(inode);
+
+	BTRFS_I(inode)->root = root;
+	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
+	BTRFS_I(inode)->dummy_inode = 1;
+
+	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
+	inode->i_op = &simple_dir_inode_operations;
+	inode->i_fop = &simple_dir_operations;
+	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
+	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
+
+	return inode;
+}
+
 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
 {
 	struct inode *inode;
-	struct btrfs_inode *bi = BTRFS_I(dir);
-	struct btrfs_root *root = bi->root;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
 	struct btrfs_root *sub_root = root;
 	struct btrfs_key location;
+	int index;
 	int ret;
 
+	dentry->d_op = &btrfs_dentry_operations;
+
 	if (dentry->d_name.len > BTRFS_NAME_LEN)
 		return ERR_PTR(-ENAMETOOLONG);
 
@@ -3242,29 +3473,50 @@ struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
 	if (ret < 0)
 		return ERR_PTR(ret);
 
-	inode = NULL;
-	if (location.objectid) {
-		ret = fixup_tree_root_location(root, &location, &sub_root,
-						dentry);
-		if (ret < 0)
-			return ERR_PTR(ret);
-		if (ret > 0)
-			return ERR_PTR(-ENOENT);
+	if (location.objectid == 0)
+		return NULL;
+
+	if (location.type == BTRFS_INODE_ITEM_KEY) {
+		inode = btrfs_iget(dir->i_sb, &location, root);
+		return inode;
+	}
+
+	BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
+
+	index = srcu_read_lock(&root->fs_info->subvol_srcu);
+	ret = fixup_tree_root_location(root, dir, dentry,
+				       &location, &sub_root);
+	if (ret < 0) {
+		if (ret != -ENOENT)
+			inode = ERR_PTR(ret);
+		else
+			inode = new_simple_dir(dir->i_sb, &location, sub_root);
+	} else {
 		inode = btrfs_iget(dir->i_sb, &location, sub_root);
-		if (IS_ERR(inode))
-			return ERR_CAST(inode);
 	}
+	srcu_read_unlock(&root->fs_info->subvol_srcu, index);
+
 	return inode;
 }
 
+static int btrfs_dentry_delete(struct dentry *dentry)
+{
+	struct btrfs_root *root;
+
+	if (!dentry->d_inode)
+		return 0;
+
+	root = BTRFS_I(dentry->d_inode)->root;
+	if (btrfs_root_refs(&root->root_item) == 0)
+		return 1;
+	return 0;
+}
+
 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 				   struct nameidata *nd)
 {
 	struct inode *inode;
 
-	if (dentry->d_name.len > BTRFS_NAME_LEN)
-		return ERR_PTR(-ENAMETOOLONG);
-
 	inode = btrfs_lookup_dentry(dir, dentry);
 	if (IS_ERR(inode))
 		return ERR_CAST(inode);
@@ -3603,9 +3855,6 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 	if (ret != 0)
 		goto fail;
 
-	if (objectid > root->highest_inode)
-		root->highest_inode = objectid;
-
 	inode->i_uid = current_fsuid();
 
 	if (dir && (dir->i_mode & S_ISGID)) {
@@ -3673,26 +3922,35 @@ int btrfs_add_link(struct btrfs_trans_handle *trans,
 		   struct inode *parent_inode, struct inode *inode,
 		   const char *name, int name_len, int add_backref, u64 index)
 {
-	int ret;
+	int ret = 0;
 	struct btrfs_key key;
 	struct btrfs_root *root = BTRFS_I(parent_inode)->root;
 
-	key.objectid = inode->i_ino;
-	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
-	key.offset = 0;
+	if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
+		memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
+	} else {
+		key.objectid = inode->i_ino;
+		btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
+		key.offset = 0;
+	}
+
+	if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
+		ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
+					 key.objectid, root->root_key.objectid,
+					 parent_inode->i_ino,
+					 index, name, name_len);
+	} else if (add_backref) {
+		ret = btrfs_insert_inode_ref(trans, root,
+					     name, name_len, inode->i_ino,
+					     parent_inode->i_ino, index);
+	}
 
-	ret = btrfs_insert_dir_item(trans, root, name, name_len,
-				    parent_inode->i_ino,
-				    &key, btrfs_inode_type(inode),
-				    index);
 	if (ret == 0) {
-		if (add_backref) {
-			ret = btrfs_insert_inode_ref(trans, root,
-						     name, name_len,
-						     inode->i_ino,
-						     parent_inode->i_ino,
-						     index);
-		}
+		ret = btrfs_insert_dir_item(trans, root, name, name_len,
+					    parent_inode->i_ino, &key,
+					    btrfs_inode_type(inode), index);
+		BUG_ON(ret);
+
 		btrfs_i_size_write(parent_inode, parent_inode->i_size +
 				   name_len * 2);
 		parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
@@ -3875,18 +4133,16 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
 
 	err = btrfs_add_nondir(trans, dentry, inode, 1, index);
 
-	if (err)
-		drop_inode = 1;
-
-	btrfs_update_inode_block_group(trans, dir);
-	err = btrfs_update_inode(trans, root, inode);
-
-	if (err)
+	if (err) {
 		drop_inode = 1;
+	} else {
+		btrfs_update_inode_block_group(trans, dir);
+		err = btrfs_update_inode(trans, root, inode);
+		BUG_ON(err);
+		btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
+	}
 
 	nr = trans->blocks_used;
-
-	btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
 	btrfs_end_transaction_throttle(trans, root);
 fail:
 	if (drop_inode) {
@@ -4064,11 +4320,11 @@ struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
 	int compressed;
 
 again:
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree, start, len);
 	if (em)
 		em->bdev = root->fs_info->fs_devices->latest_bdev;
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 
 	if (em) {
 		if (em->start > start || em->start + em->len <= start)
@@ -4215,6 +4471,11 @@ again:
 				map = kmap(page);
 				read_extent_buffer(leaf, map + pg_offset, ptr,
 						   copy_size);
+				if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
+					memset(map + pg_offset + copy_size, 0,
+					       PAGE_CACHE_SIZE - pg_offset -
+					       copy_size);
+				}
 				kunmap(page);
 			}
 			flush_dcache_page(page);
@@ -4259,7 +4520,7 @@ insert:
 	}
 
 	err = 0;
-	spin_lock(&em_tree->lock);
+	write_lock(&em_tree->lock);
 	ret = add_extent_mapping(em_tree, em);
 	/* it is possible that someone inserted the extent into the tree
 	 * while we had the lock dropped.  It is also possible that
@@ -4299,7 +4560,7 @@ insert:
 			err = 0;
 		}
 	}
-	spin_unlock(&em_tree->lock);
+	write_unlock(&em_tree->lock);
 out:
 	if (path)
 		btrfs_free_path(path);
@@ -4398,13 +4659,21 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
 	u64 page_start = page_offset(page);
 	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 
+
+	/*
+	 * we have the page locked, so new writeback can't start,
+	 * and the dirty bit won't be cleared while we are here.
+	 *
+	 * Wait for IO on this page so that we can safely clear
+	 * the PagePrivate2 bit and do ordered accounting
+	 */
 	wait_on_page_writeback(page);
+
 	tree = &BTRFS_I(page->mapping->host)->io_tree;
 	if (offset) {
 		btrfs_releasepage(page, GFP_NOFS);
 		return;
 	}
-
 	lock_extent(tree, page_start, page_end, GFP_NOFS);
 	ordered = btrfs_lookup_ordered_extent(page->mapping->host,
 					   page_offset(page));
@@ -4415,16 +4684,21 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset)
 		 */
 		clear_extent_bit(tree, page_start, page_end,
 				 EXTENT_DIRTY | EXTENT_DELALLOC |
-				 EXTENT_LOCKED, 1, 0, GFP_NOFS);
-		btrfs_finish_ordered_io(page->mapping->host,
-					page_start, page_end);
+				 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
+		/*
+		 * whoever cleared the private bit is responsible
+		 * for the finish_ordered_io
+		 */
+		if (TestClearPagePrivate2(page)) {
+			btrfs_finish_ordered_io(page->mapping->host,
+						page_start, page_end);
+		}
 		btrfs_put_ordered_extent(ordered);
 		lock_extent(tree, page_start, page_end, GFP_NOFS);
 	}
 	clear_extent_bit(tree, page_start, page_end,
-		 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
-		 EXTENT_ORDERED,
-		 1, 1, GFP_NOFS);
+		 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
+		 1, 1, NULL, GFP_NOFS);
 	__btrfs_releasepage(page, GFP_NOFS);
 
 	ClearPageChecked(page);
@@ -4521,11 +4795,14 @@ again:
 	}
 	ClearPageChecked(page);
 	set_page_dirty(page);
+	SetPageUptodate(page);
 
 	BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
 	unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 
 out_unlock:
+	if (!ret)
+		return VM_FAULT_LOCKED;
 	unlock_page(page);
 out:
 	return ret;
@@ -4594,11 +4871,11 @@ out:
  * create a new subvolume directory/inode (helper for the ioctl).
  */
 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
-			     struct btrfs_root *new_root, struct dentry *dentry,
+			     struct btrfs_root *new_root,
 			     u64 new_dirid, u64 alloc_hint)
 {
 	struct inode *inode;
-	int error;
+	int err;
 	u64 index = 0;
 
 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
@@ -4611,11 +4888,10 @@ int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
 	inode->i_nlink = 1;
 	btrfs_i_size_write(inode, 0);
 
-	error = btrfs_update_inode(trans, new_root, inode);
-	if (error)
-		return error;
+	err = btrfs_update_inode(trans, new_root, inode);
+	BUG_ON(err);
 
-	d_instantiate(dentry, inode);
+	iput(inode);
 	return 0;
 }
 
@@ -4693,6 +4969,16 @@ void btrfs_destroy_inode(struct inode *inode)
 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
 }
 
+void btrfs_drop_inode(struct inode *inode)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+
+	if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
+		generic_delete_inode(inode);
+	else
+		generic_drop_inode(inode);
+}
+
 static void init_once(void *foo)
 {
 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
@@ -4761,31 +5047,32 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 {
 	struct btrfs_trans_handle *trans;
 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
+	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
 	struct inode *new_inode = new_dentry->d_inode;
 	struct inode *old_inode = old_dentry->d_inode;
 	struct timespec ctime = CURRENT_TIME;
 	u64 index = 0;
+	u64 root_objectid;
 	int ret;
 
-	/* we're not allowed to rename between subvolumes */
-	if (BTRFS_I(old_inode)->root->root_key.objectid !=
-	    BTRFS_I(new_dir)->root->root_key.objectid)
+	if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
+		return -EPERM;
+
+	/* we only allow rename subvolume link between subvolumes */
+	if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
 		return -EXDEV;
 
-	if (S_ISDIR(old_inode->i_mode) && new_inode &&
-	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
+	if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
+	    (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
 		return -ENOTEMPTY;
-	}
 
-	/* to rename a snapshot or subvolume, we need to juggle the
-	 * backrefs.  This isn't coded yet
-	 */
-	if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
-		return -EXDEV;
+	if (S_ISDIR(old_inode->i_mode) && new_inode &&
+	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
+		return -ENOTEMPTY;
 
 	ret = btrfs_check_metadata_free_space(root);
 	if (ret)
-		goto out_unlock;
+		return ret;
 
 	/*
 	 * we're using rename to replace one file with another.
@@ -4796,8 +5083,40 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	    old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
 		filemap_flush(old_inode->i_mapping);
 
+	/* close the racy window with snapshot create/destroy ioctl */
+	if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
+		down_read(&root->fs_info->subvol_sem);
+
 	trans = btrfs_start_transaction(root, 1);
+	btrfs_set_trans_block_group(trans, new_dir);
+
+	if (dest != root)
+		btrfs_record_root_in_trans(trans, dest);
 
+	ret = btrfs_set_inode_index(new_dir, &index);
+	if (ret)
+		goto out_fail;
+
+	if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
+		/* force full log commit if subvolume involved. */
+		root->fs_info->last_trans_log_full_commit = trans->transid;
+	} else {
+		ret = btrfs_insert_inode_ref(trans, dest,
+					     new_dentry->d_name.name,
+					     new_dentry->d_name.len,
+					     old_inode->i_ino,
+					     new_dir->i_ino, index);
+		if (ret)
+			goto out_fail;
+		/*
+		 * this is an ugly little race, but the rename is required
+		 * to make sure that if we crash, the inode is either at the
+		 * old name or the new one.  pinning the log transaction lets
+		 * us make sure we don't allow a log commit to come in after
+		 * we unlink the name but before we add the new name back in.
+		 */
+		btrfs_pin_log_trans(root);
+	}
 	/*
 	 * make sure the inode gets flushed if it is replacing
 	 * something.
@@ -4807,18 +5126,6 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 		btrfs_add_ordered_operation(trans, root, old_inode);
 	}
 
-	/*
-	 * this is an ugly little race, but the rename is required to make
-	 * sure that if we crash, the inode is either at the old name
-	 * or the new one.  pinning the log transaction lets us make sure
-	 * we don't allow a log commit to come in after we unlink the
-	 * name but before we add the new name back in.
-	 */
-	btrfs_pin_log_trans(root);
-
-	btrfs_set_trans_block_group(trans, new_dir);
-
-	btrfs_inc_nlink(old_dentry->d_inode);
 	old_dir->i_ctime = old_dir->i_mtime = ctime;
 	new_dir->i_ctime = new_dir->i_mtime = ctime;
 	old_inode->i_ctime = ctime;
@@ -4826,47 +5133,58 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	if (old_dentry->d_parent != new_dentry->d_parent)
 		btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
 
-	ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
-				 old_dentry->d_name.name,
-				 old_dentry->d_name.len);
-	if (ret)
-		goto out_fail;
+	if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
+		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
+		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
+					old_dentry->d_name.name,
+					old_dentry->d_name.len);
+	} else {
+		btrfs_inc_nlink(old_dentry->d_inode);
+		ret = btrfs_unlink_inode(trans, root, old_dir,
+					 old_dentry->d_inode,
+					 old_dentry->d_name.name,
+					 old_dentry->d_name.len);
+	}
+	BUG_ON(ret);
 
 	if (new_inode) {
 		new_inode->i_ctime = CURRENT_TIME;
-		ret = btrfs_unlink_inode(trans, root, new_dir,
-					 new_dentry->d_inode,
-					 new_dentry->d_name.name,
-					 new_dentry->d_name.len);
-		if (ret)
-			goto out_fail;
+		if (unlikely(new_inode->i_ino ==
+			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
+			root_objectid = BTRFS_I(new_inode)->location.objectid;
+			ret = btrfs_unlink_subvol(trans, dest, new_dir,
+						root_objectid,
+						new_dentry->d_name.name,
+						new_dentry->d_name.len);
+			BUG_ON(new_inode->i_nlink == 0);
+		} else {
+			ret = btrfs_unlink_inode(trans, dest, new_dir,
+						 new_dentry->d_inode,
+						 new_dentry->d_name.name,
+						 new_dentry->d_name.len);
+		}
+		BUG_ON(ret);
 		if (new_inode->i_nlink == 0) {
 			ret = btrfs_orphan_add(trans, new_dentry->d_inode);
-			if (ret)
-				goto out_fail;
+			BUG_ON(ret);
 		}
-
 	}
-	ret = btrfs_set_inode_index(new_dir, &index);
-	if (ret)
-		goto out_fail;
 
-	ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
-			     old_inode, new_dentry->d_name.name,
-			     new_dentry->d_name.len, 1, index);
-	if (ret)
-		goto out_fail;
+	ret = btrfs_add_link(trans, new_dir, old_inode,
+			     new_dentry->d_name.name,
+			     new_dentry->d_name.len, 0, index);
+	BUG_ON(ret);
 
-	btrfs_log_new_name(trans, old_inode, old_dir,
-				       new_dentry->d_parent);
+	if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
+		btrfs_log_new_name(trans, old_inode, old_dir,
+				   new_dentry->d_parent);
+		btrfs_end_log_trans(root);
+	}
 out_fail:
-
-	/* this btrfs_end_log_trans just allows the current
-	 * log-sub transaction to complete
-	 */
-	btrfs_end_log_trans(root);
 	btrfs_end_transaction_throttle(trans, root);
-out_unlock:
+
+	if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
+		up_read(&root->fs_info->subvol_sem);
 	return ret;
 }
 
@@ -5058,6 +5376,8 @@ static int prealloc_file_range(struct btrfs_trans_handle *trans,
 						  0, 0, 0,
 						  BTRFS_FILE_EXTENT_PREALLOC);
 		BUG_ON(ret);
+		btrfs_drop_extent_cache(inode, cur_offset,
+					cur_offset + ins.offset -1, 0);
 		num_bytes -= ins.offset;
 		cur_offset += ins.offset;
 		alloc_hint = ins.objectid + ins.offset;
@@ -5223,6 +5543,7 @@ static const struct inode_operations btrfs_dir_ro_inode_operations = {
 	.lookup		= btrfs_lookup,
 	.permission	= btrfs_permission,
 };
+
 static struct file_operations btrfs_dir_file_operations = {
 	.llseek		= generic_file_llseek,
 	.read		= generic_read_dir,
@@ -5309,3 +5630,7 @@ static const struct inode_operations btrfs_symlink_inode_operations = {
 	.listxattr	= btrfs_listxattr,
 	.removexattr	= btrfs_removexattr,
 };
+
+struct dentry_operations btrfs_dentry_operations = {
+	.d_delete	= btrfs_dentry_delete,
+};

fs/btrfs/ioctl.c

@@ -230,8 +230,8 @@ static noinline int create_subvol(struct btrfs_root *root,
 	struct btrfs_root_item root_item;
 	struct btrfs_inode_item *inode_item;
 	struct extent_buffer *leaf;
-	struct btrfs_root *new_root = root;
-	struct inode *dir;
+	struct btrfs_root *new_root;
+	struct inode *dir = dentry->d_parent->d_inode;
 	int ret;
 	int err;
 	u64 objectid;
@@ -241,7 +241,7 @@ static noinline int create_subvol(struct btrfs_root *root,
 
 	ret = btrfs_check_metadata_free_space(root);
 	if (ret)
-		goto fail_commit;
+		return ret;
 
 	trans = btrfs_start_transaction(root, 1);
 	BUG_ON(!trans);
@@ -304,11 +304,17 @@ static noinline int create_subvol(struct btrfs_root *root,
 	if (ret)
 		goto fail;
 
+	key.offset = (u64)-1;
+	new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
+	BUG_ON(IS_ERR(new_root));
+
+	btrfs_record_root_in_trans(trans, new_root);
+
+	ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
+				       BTRFS_I(dir)->block_group);
 	/*
 	 * insert the directory item
 	 */
-	key.offset = (u64)-1;
-	dir = dentry->d_parent->d_inode;
 	ret = btrfs_set_inode_index(dir, &index);
 	BUG_ON(ret);
 
@@ -322,44 +328,18 @@ static noinline int create_subvol(struct btrfs_root *root,
 	ret = btrfs_update_inode(trans, root, dir);
 	BUG_ON(ret);
 
-	/* add the backref first */
 	ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
-				 objectid, BTRFS_ROOT_BACKREF_KEY,
-				 root->root_key.objectid,
+				 objectid, root->root_key.objectid,
 				 dir->i_ino, index, name, namelen);
 
 	BUG_ON(ret);
 
-	/* now add the forward ref */
-	ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
-				 root->root_key.objectid, BTRFS_ROOT_REF_KEY,
-				 objectid,
-				 dir->i_ino, index, name, namelen);
-
-	BUG_ON(ret);
-
-	ret = btrfs_commit_transaction(trans, root);
-	if (ret)
-		goto fail_commit;
-
-	new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
-	BUG_ON(!new_root);
-
-	trans = btrfs_start_transaction(new_root, 1);
-	BUG_ON(!trans);
-
-	ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
-				       BTRFS_I(dir)->block_group);
-	if (ret)
-		goto fail;
-
+	d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 fail:
 	nr = trans->blocks_used;
-	err = btrfs_commit_transaction(trans, new_root);
+	err = btrfs_commit_transaction(trans, root);
 	if (err && !ret)
 		ret = err;
-fail_commit:
-	btrfs_btree_balance_dirty(root, nr);
 	return ret;
 }
 
@@ -420,14 +400,15 @@ static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
  * inside this filesystem so it's quite a bit simpler.
  */
-static noinline int btrfs_mksubvol(struct path *parent, char *name,
-				   int mode, int namelen,
+static noinline int btrfs_mksubvol(struct path *parent,
+				   char *name, int namelen,
 				   struct btrfs_root *snap_src)
 {
+	struct inode *dir  = parent->dentry->d_inode;
 	struct dentry *dentry;
 	int error;
 
-	mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+	mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 
 	dentry = lookup_one_len(name, parent->dentry, namelen);
 	error = PTR_ERR(dentry);
@@ -438,99 +419,39 @@ static noinline int btrfs_mksubvol(struct path *parent, char *name,
 	if (dentry->d_inode)
 		goto out_dput;
 
-	if (!IS_POSIXACL(parent->dentry->d_inode))
-		mode &= ~current_umask();
-
 	error = mnt_want_write(parent->mnt);
 	if (error)
 		goto out_dput;
 
-	error = btrfs_may_create(parent->dentry->d_inode, dentry);
+	error = btrfs_may_create(dir, dentry);
 	if (error)
 		goto out_drop_write;
 
-	/*
-	 * Actually perform the low-level subvolume creation after all
-	 * this VFS fuzz.
-	 *
-	 * Eventually we want to pass in an inode under which we create this
-	 * subvolume, but for now all are under the filesystem root.
-	 *
-	 * Also we should pass on the mode eventually to allow creating new
-	 * subvolume with specific mode bits.
-	 */
+	down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
+
+	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
+		goto out_up_read;
+
 	if (snap_src) {
-		struct dentry *dir = dentry->d_parent;
-		struct dentry *test = dir->d_parent;
-		struct btrfs_path *path = btrfs_alloc_path();
-		int ret;
-		u64 test_oid;
-		u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
-
-		test_oid = snap_src->root_key.objectid;
-
-		ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
-					  path, parent_oid, test_oid);
-		if (ret == 0)
-			goto create;
-		btrfs_release_path(snap_src->fs_info->tree_root, path);
-
-		/* we need to make sure we aren't creating a directory loop
-		 * by taking a snapshot of something that has our current
-		 * subvol in its directory tree.  So, this loops through
-		 * the dentries and checks the forward refs for each subvolume
-		 * to see if is references the subvolume where we are
-		 * placing this new snapshot.
-		 */
-		while (1) {
-			if (!test ||
-			    dir == snap_src->fs_info->sb->s_root ||
-			    test == snap_src->fs_info->sb->s_root ||
-			    test->d_inode->i_sb != snap_src->fs_info->sb) {
-				break;
-			}
-			if (S_ISLNK(test->d_inode->i_mode)) {
-				printk(KERN_INFO "Btrfs symlink in snapshot "
-				       "path, failed\n");
-				error = -EMLINK;
-				btrfs_free_path(path);
-				goto out_drop_write;
-			}
-			test_oid =
-				BTRFS_I(test->d_inode)->root->root_key.objectid;
-			ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
-				  path, test_oid, parent_oid);
-			if (ret == 0) {
-				printk(KERN_INFO "Btrfs snapshot creation "
-				       "failed, looping\n");
-				error = -EMLINK;
-				btrfs_free_path(path);
-				goto out_drop_write;
-			}
-			btrfs_release_path(snap_src->fs_info->tree_root, path);
-			test = test->d_parent;
-		}
-create:
-		btrfs_free_path(path);
-		error = create_snapshot(snap_src, dentry, name, namelen);
+		error = create_snapshot(snap_src, dentry,
+					name, namelen);
 	} else {
-		error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
-				      dentry, name, namelen);
+		error = create_subvol(BTRFS_I(dir)->root, dentry,
+				      name, namelen);
 	}
-	if (error)
-		goto out_drop_write;
-
-	fsnotify_mkdir(parent->dentry->d_inode, dentry);
+	if (!error)
+		fsnotify_mkdir(dir, dentry);
+out_up_read:
+	up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 out_drop_write:
 	mnt_drop_write(parent->mnt);
 out_dput:
 	dput(dentry);
 out_unlock:
-	mutex_unlock(&parent->dentry->d_inode->i_mutex);
+	mutex_unlock(&dir->i_mutex);
 	return error;
 }
 
-
 static int btrfs_defrag_file(struct file *file)
 {
 	struct inode *inode = fdentry(file)->d_inode;
@@ -596,9 +517,8 @@ again:
 		clear_page_dirty_for_io(page);
 
 		btrfs_set_extent_delalloc(inode, page_start, page_end);
-
-		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 		set_page_dirty(page);
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 		unlock_page(page);
 		page_cache_release(page);
 		balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
@@ -609,7 +529,8 @@ out_unlock:
 	return 0;
 }
 
-static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
+static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
+					void __user *arg)
 {
 	u64 new_size;
 	u64 old_size;
@@ -718,10 +639,7 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
 {
 	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 	struct btrfs_ioctl_vol_args *vol_args;
-	struct btrfs_dir_item *di;
-	struct btrfs_path *path;
 	struct file *src_file;
-	u64 root_dirid;
 	int namelen;
 	int ret = 0;
 
@@ -739,32 +657,9 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
 		goto out;
 	}
 
-	path = btrfs_alloc_path();
-	if (!path) {
-		ret = -ENOMEM;
-		goto out;
-	}
-
-	root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
-	di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
-			    path, root_dirid,
-			    vol_args->name, namelen, 0);
-	btrfs_free_path(path);
-
-	if (di && !IS_ERR(di)) {
-		ret = -EEXIST;
-		goto out;
-	}
-
-	if (IS_ERR(di)) {
-		ret = PTR_ERR(di);
-		goto out;
-	}
-
 	if (subvol) {
-		ret = btrfs_mksubvol(&file->f_path, vol_args->name,
-				     file->f_path.dentry->d_inode->i_mode,
-				     namelen, NULL);
+		ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
+				     NULL);
 	} else {
 		struct inode *src_inode;
 		src_file = fget(vol_args->fd);
@@ -781,17 +676,156 @@ static noinline int btrfs_ioctl_snap_create(struct file *file,
 			fput(src_file);
 			goto out;
 		}
-		ret = btrfs_mksubvol(&file->f_path, vol_args->name,
-			     file->f_path.dentry->d_inode->i_mode,
-			     namelen, BTRFS_I(src_inode)->root);
+		ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
+				     BTRFS_I(src_inode)->root);
 		fput(src_file);
 	}
-
 out:
 	kfree(vol_args);
 	return ret;
 }
 
+/*
+ * helper to check if the subvolume references other subvolumes
+ */
+static noinline int may_destroy_subvol(struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = root->root_key.objectid;
+	key.type = BTRFS_ROOT_REF_KEY;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
+				&key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	BUG_ON(ret == 0);
+
+	ret = 0;
+	if (path->slots[0] > 0) {
+		path->slots[0]--;
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+		if (key.objectid == root->root_key.objectid &&
+		    key.type == BTRFS_ROOT_REF_KEY)
+			ret = -ENOTEMPTY;
+	}
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static noinline int btrfs_ioctl_snap_destroy(struct file *file,
+					     void __user *arg)
+{
+	struct dentry *parent = fdentry(file);
+	struct dentry *dentry;
+	struct inode *dir = parent->d_inode;
+	struct inode *inode;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct btrfs_root *dest = NULL;
+	struct btrfs_ioctl_vol_args *vol_args;
+	struct btrfs_trans_handle *trans;
+	int namelen;
+	int ret;
+	int err = 0;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	vol_args = memdup_user(arg, sizeof(*vol_args));
+	if (IS_ERR(vol_args))
+		return PTR_ERR(vol_args);
+
+	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
+	namelen = strlen(vol_args->name);
+	if (strchr(vol_args->name, '/') ||
+	    strncmp(vol_args->name, "..", namelen) == 0) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = mnt_want_write(file->f_path.mnt);
+	if (err)
+		goto out;
+
+	mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
+	dentry = lookup_one_len(vol_args->name, parent, namelen);
+	if (IS_ERR(dentry)) {
+		err = PTR_ERR(dentry);
+		goto out_unlock_dir;
+	}
+
+	if (!dentry->d_inode) {
+		err = -ENOENT;
+		goto out_dput;
+	}
+
+	inode = dentry->d_inode;
+	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
+		err = -EINVAL;
+		goto out_dput;
+	}
+
+	dest = BTRFS_I(inode)->root;
+
+	mutex_lock(&inode->i_mutex);
+	err = d_invalidate(dentry);
+	if (err)
+		goto out_unlock;
+
+	down_write(&root->fs_info->subvol_sem);
+
+	err = may_destroy_subvol(dest);
+	if (err)
+		goto out_up_write;
+
+	trans = btrfs_start_transaction(root, 1);
+	ret = btrfs_unlink_subvol(trans, root, dir,
+				dest->root_key.objectid,
+				dentry->d_name.name,
+				dentry->d_name.len);
+	BUG_ON(ret);
+
+	btrfs_record_root_in_trans(trans, dest);
+
+	memset(&dest->root_item.drop_progress, 0,
+		sizeof(dest->root_item.drop_progress));
+	dest->root_item.drop_level = 0;
+	btrfs_set_root_refs(&dest->root_item, 0);
+
+	ret = btrfs_insert_orphan_item(trans,
+				root->fs_info->tree_root,
+				dest->root_key.objectid);
+	BUG_ON(ret);
+
+	ret = btrfs_commit_transaction(trans, root);
+	BUG_ON(ret);
+	inode->i_flags |= S_DEAD;
+out_up_write:
+	up_write(&root->fs_info->subvol_sem);
+out_unlock:
+	mutex_unlock(&inode->i_mutex);
+	if (!err) {
+		btrfs_invalidate_inodes(dest);
+		d_delete(dentry);
+	}
+out_dput:
+	dput(dentry);
+out_unlock_dir:
+	mutex_unlock(&dir->i_mutex);
+	mnt_drop_write(file->f_path.mnt);
+out:
+	kfree(vol_args);
+	return err;
+}
+
 static int btrfs_ioctl_defrag(struct file *file)
 {
 	struct inode *inode = fdentry(file)->d_inode;
@@ -865,8 +899,8 @@ static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 	return ret;
 }
 
-static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
-		u64 off, u64 olen, u64 destoff)
+static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
+				       u64 off, u64 olen, u64 destoff)
 {
 	struct inode *inode = fdentry(file)->d_inode;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
@@ -976,7 +1010,7 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 
 	/* punch hole in destination first */
 	btrfs_drop_extents(trans, root, inode, off, off + len,
-			   off + len, 0, &hint_byte);
+			   off + len, 0, &hint_byte, 1);
 
 	/* clone data */
 	key.objectid = src->i_ino;
@@ -1071,8 +1105,7 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 					datao += off - key.offset;
 					datal -= off - key.offset;
 				}
-				if (key.offset + datao + datal + key.offset >
-				    off + len)
+				if (key.offset + datao + datal > off + len)
 					datal = off + len - key.offset - datao;
 				/* disko == 0 means it's a hole */
 				if (!disko)
@@ -1258,6 +1291,8 @@ long btrfs_ioctl(struct file *file, unsigned int
 		return btrfs_ioctl_snap_create(file, argp, 0);
 	case BTRFS_IOC_SUBVOL_CREATE:
 		return btrfs_ioctl_snap_create(file, argp, 1);
+	case BTRFS_IOC_SNAP_DESTROY:
+		return btrfs_ioctl_snap_destroy(file, argp);
 	case BTRFS_IOC_DEFRAG:
 		return btrfs_ioctl_defrag(file);
 	case BTRFS_IOC_RESIZE:

fs/btrfs/ioctl.h

@@ -65,5 +65,6 @@ struct btrfs_ioctl_clone_range_args {
 
 #define BTRFS_IOC_SUBVOL_CREATE _IOW(BTRFS_IOCTL_MAGIC, 14, \
 				   struct btrfs_ioctl_vol_args)
-
+#define BTRFS_IOC_SNAP_DESTROY _IOW(BTRFS_IOCTL_MAGIC, 15, \
+				struct btrfs_ioctl_vol_args)
 #endif

fs/btrfs/ordered-data.c

@@ -159,8 +159,6 @@ static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
  *
  * len is the length of the extent
  *
- * This also sets the EXTENT_ORDERED bit on the range in the inode.
- *
  * The tree is given a single reference on the ordered extent that was
  * inserted.
  */
@@ -181,6 +179,7 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 	entry->start = start;
 	entry->len = len;
 	entry->disk_len = disk_len;
+	entry->bytes_left = len;
 	entry->inode = inode;
 	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 		set_bit(type, &entry->flags);
@@ -195,9 +194,6 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 			   &entry->rb_node);
 	BUG_ON(node);
 
-	set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
-			   entry_end(entry) - 1, GFP_NOFS);
-
 	spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 	list_add_tail(&entry->root_extent_list,
 		      &BTRFS_I(inode)->root->fs_info->ordered_extents);
@@ -241,13 +237,10 @@ int btrfs_dec_test_ordered_pending(struct inode *inode,
 	struct btrfs_ordered_inode_tree *tree;
 	struct rb_node *node;
 	struct btrfs_ordered_extent *entry;
-	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 	int ret;
 
 	tree = &BTRFS_I(inode)->ordered_tree;
 	mutex_lock(&tree->mutex);
-	clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
-			     GFP_NOFS);
 	node = tree_search(tree, file_offset);
 	if (!node) {
 		ret = 1;
@@ -260,11 +253,16 @@ int btrfs_dec_test_ordered_pending(struct inode *inode,
 		goto out;
 	}
 
-	ret = test_range_bit(io_tree, entry->file_offset,
-			     entry->file_offset + entry->len - 1,
-			     EXTENT_ORDERED, 0);
-	if (ret == 0)
+	if (io_size > entry->bytes_left) {
+		printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
+		       (unsigned long long)entry->bytes_left,
+		       (unsigned long long)io_size);
+	}
+	entry->bytes_left -= io_size;
+	if (entry->bytes_left == 0)
 		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
+	else
+		ret = 1;
 out:
 	mutex_unlock(&tree->mutex);
 	return ret == 0;
@@ -476,6 +474,7 @@ int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 	u64 orig_end;
 	u64 wait_end;
 	struct btrfs_ordered_extent *ordered;
+	int found;
 
 	if (start + len < start) {
 		orig_end = INT_LIMIT(loff_t);
@@ -502,6 +501,7 @@ again:
 					   orig_end >> PAGE_CACHE_SHIFT);
 
 	end = orig_end;
+	found = 0;
 	while (1) {
 		ordered = btrfs_lookup_first_ordered_extent(inode, end);
 		if (!ordered)
@@ -514,6 +514,7 @@ again:
 			btrfs_put_ordered_extent(ordered);
 			break;
 		}
+		found++;
 		btrfs_start_ordered_extent(inode, ordered, 1);
 		end = ordered->file_offset;
 		btrfs_put_ordered_extent(ordered);
@@ -521,8 +522,8 @@ again:
 			break;
 		end--;
 	}
-	if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
-			   EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
+	if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
+			   EXTENT_DELALLOC, 0, NULL)) {
 		schedule_timeout(1);
 		goto again;
 	}
@@ -613,7 +614,7 @@ int btrfs_ordered_update_i_size(struct inode *inode,
 	 */
 	if (test_range_bit(io_tree, disk_i_size,
 			   ordered->file_offset + ordered->len - 1,
-			   EXTENT_DELALLOC, 0)) {
+			   EXTENT_DELALLOC, 0, NULL)) {
 		goto out;
 	}
 	/*
@@ -664,7 +665,7 @@ int btrfs_ordered_update_i_size(struct inode *inode,
 	 */
 	if (i_size_test > entry_end(ordered) &&
 	    !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
-			   EXTENT_DELALLOC, 0)) {
+			   EXTENT_DELALLOC, 0, NULL)) {
 		new_i_size = min_t(u64, i_size_test, i_size_read(inode));
 	}
 	BTRFS_I(inode)->disk_i_size = new_i_size;

fs/btrfs/ordered-data.h

@@ -85,6 +85,9 @@ struct btrfs_ordered_extent {
 	/* extent length on disk */
 	u64 disk_len;
 
+	/* number of bytes that still need writing */
+	u64 bytes_left;
+
 	/* flags (described above) */
 	unsigned long flags;
 

fs/btrfs/orphan.c

@@ -65,3 +65,23 @@ out:
 	btrfs_free_path(path);
 	return ret;
 }
+
+int btrfs_find_orphan_item(struct btrfs_root *root, u64 offset)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int ret;
+
+	key.objectid = BTRFS_ORPHAN_OBJECTID;
+	key.type = BTRFS_ORPHAN_ITEM_KEY;
+	key.offset = offset;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+
+	btrfs_free_path(path);
+	return ret;
+}

fs/btrfs/relocation.c

@@ -121,6 +121,15 @@ struct inodevec {
 	int nr;
 };
 
+#define MAX_EXTENTS 128
+
+struct file_extent_cluster {
+	u64 start;
+	u64 end;
+	u64 boundary[MAX_EXTENTS];
+	unsigned int nr;
+};
+
 struct reloc_control {
 	/* block group to relocate */
 	struct btrfs_block_group_cache *block_group;
@@ -2180,7 +2189,7 @@ static int tree_block_processed(u64 bytenr, u32 blocksize,
 				struct reloc_control *rc)
 {
 	if (test_range_bit(&rc->processed_blocks, bytenr,
-			   bytenr + blocksize - 1, EXTENT_DIRTY, 1))
+			   bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
 		return 1;
 	return 0;
 }
@@ -2529,56 +2538,94 @@ out:
 }
 
 static noinline_for_stack
-int relocate_inode_pages(struct inode *inode, u64 start, u64 len)
+int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
+			 u64 block_start)
+{
+	struct btrfs_root *root = BTRFS_I(inode)->root;
+	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+	struct extent_map *em;
+	int ret = 0;
+
+	em = alloc_extent_map(GFP_NOFS);
+	if (!em)
+		return -ENOMEM;
+
+	em->start = start;
+	em->len = end + 1 - start;
+	em->block_len = em->len;
+	em->block_start = block_start;
+	em->bdev = root->fs_info->fs_devices->latest_bdev;
+	set_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+	lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+	while (1) {
+		write_lock(&em_tree->lock);
+		ret = add_extent_mapping(em_tree, em);
+		write_unlock(&em_tree->lock);
+		if (ret != -EEXIST) {
+			free_extent_map(em);
+			break;
+		}
+		btrfs_drop_extent_cache(inode, start, end, 0);
+	}
+	unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+	return ret;
+}
+
+static int relocate_file_extent_cluster(struct inode *inode,
+					struct file_extent_cluster *cluster)
 {
 	u64 page_start;
 	u64 page_end;
-	unsigned long i;
-	unsigned long first_index;
+	u64 offset = BTRFS_I(inode)->index_cnt;
+	unsigned long index;
 	unsigned long last_index;
-	unsigned int total_read = 0;
-	unsigned int total_dirty = 0;
+	unsigned int dirty_page = 0;
 	struct page *page;
 	struct file_ra_state *ra;
-	struct btrfs_ordered_extent *ordered;
-	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	int nr = 0;
 	int ret = 0;
 
+	if (!cluster->nr)
+		return 0;
+
 	ra = kzalloc(sizeof(*ra), GFP_NOFS);
 	if (!ra)
 		return -ENOMEM;
 
+	index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
+	last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
+
 	mutex_lock(&inode->i_mutex);
-	first_index = start >> PAGE_CACHE_SHIFT;
-	last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
 
-	/* make sure the dirty trick played by the caller work */
-	while (1) {
-		ret = invalidate_inode_pages2_range(inode->i_mapping,
-						    first_index, last_index);
-		if (ret != -EBUSY)
-			break;
-		schedule_timeout(HZ/10);
-	}
+	i_size_write(inode, cluster->end + 1 - offset);
+	ret = setup_extent_mapping(inode, cluster->start - offset,
+				   cluster->end - offset, cluster->start);
 	if (ret)
 		goto out_unlock;
 
 	file_ra_state_init(ra, inode->i_mapping);
 
-	for (i = first_index ; i <= last_index; i++) {
-		if (total_read % ra->ra_pages == 0) {
-			btrfs_force_ra(inode->i_mapping, ra, NULL, i,
-				min(last_index, ra->ra_pages + i - 1));
-		}
-		total_read++;
-again:
-		if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
-			BUG_ON(1);
-		page = grab_cache_page(inode->i_mapping, i);
+	WARN_ON(cluster->start != cluster->boundary[0]);
+	while (index <= last_index) {
+		page = find_lock_page(inode->i_mapping, index);
 		if (!page) {
-			ret = -ENOMEM;
-			goto out_unlock;
+			page_cache_sync_readahead(inode->i_mapping,
+						  ra, NULL, index,
+						  last_index + 1 - index);
+			page = grab_cache_page(inode->i_mapping, index);
+			if (!page) {
+				ret = -ENOMEM;
+				goto out_unlock;
+			}
+		}
+
+		if (PageReadahead(page)) {
+			page_cache_async_readahead(inode->i_mapping,
+						   ra, NULL, page, index,
+						   last_index + 1 - index);
 		}
+
 		if (!PageUptodate(page)) {
 			btrfs_readpage(NULL, page);
 			lock_page(page);
@@ -2589,75 +2636,79 @@ again:
 				goto out_unlock;
 			}
 		}
-		wait_on_page_writeback(page);
 
 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 		page_end = page_start + PAGE_CACHE_SIZE - 1;
-		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
-
-		ordered = btrfs_lookup_ordered_extent(inode, page_start);
-		if (ordered) {
-			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
-			unlock_page(page);
-			page_cache_release(page);
-			btrfs_start_ordered_extent(inode, ordered, 1);
-			btrfs_put_ordered_extent(ordered);
-			goto again;
-		}
+
+		lock_extent(&BTRFS_I(inode)->io_tree,
+			    page_start, page_end, GFP_NOFS);
+
 		set_page_extent_mapped(page);
 
-		if (i == first_index)
-			set_extent_bits(io_tree, page_start, page_end,
+		if (nr < cluster->nr &&
+		    page_start + offset == cluster->boundary[nr]) {
+			set_extent_bits(&BTRFS_I(inode)->io_tree,
+					page_start, page_end,
 					EXTENT_BOUNDARY, GFP_NOFS);
+			nr++;
+		}
 		btrfs_set_extent_delalloc(inode, page_start, page_end);
 
 		set_page_dirty(page);
-		total_dirty++;
+		dirty_page++;
 
-		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+		unlock_extent(&BTRFS_I(inode)->io_tree,
+			      page_start, page_end, GFP_NOFS);
 		unlock_page(page);
 		page_cache_release(page);
+
+		index++;
+		if (nr < cluster->nr &&
+		    page_end + 1 + offset == cluster->boundary[nr]) {
+			balance_dirty_pages_ratelimited_nr(inode->i_mapping,
+							   dirty_page);
+			dirty_page = 0;
+		}
+	}
+	if (dirty_page) {
+		balance_dirty_pages_ratelimited_nr(inode->i_mapping,
+						   dirty_page);
 	}
+	WARN_ON(nr != cluster->nr);
 out_unlock:
 	mutex_unlock(&inode->i_mutex);
 	kfree(ra);
-	balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
 	return ret;
 }
 
 static noinline_for_stack
-int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key)
+int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
+			 struct file_extent_cluster *cluster)
 {
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
-	struct extent_map *em;
-	u64 start = extent_key->objectid - BTRFS_I(inode)->index_cnt;
-	u64 end = start + extent_key->offset - 1;
-
-	em = alloc_extent_map(GFP_NOFS);
-	em->start = start;
-	em->len = extent_key->offset;
-	em->block_len = extent_key->offset;
-	em->block_start = extent_key->objectid;
-	em->bdev = root->fs_info->fs_devices->latest_bdev;
-	set_bit(EXTENT_FLAG_PINNED, &em->flags);
+	int ret;
 
-	/* setup extent map to cheat btrfs_readpage */
-	lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
-	while (1) {
-		int ret;
-		spin_lock(&em_tree->lock);
-		ret = add_extent_mapping(em_tree, em);
-		spin_unlock(&em_tree->lock);
-		if (ret != -EEXIST) {
-			free_extent_map(em);
-			break;
-		}
-		btrfs_drop_extent_cache(inode, start, end, 0);
+	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
+		ret = relocate_file_extent_cluster(inode, cluster);
+		if (ret)
+			return ret;
+		cluster->nr = 0;
 	}
-	unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
 
-	return relocate_inode_pages(inode, start, extent_key->offset);
+	if (!cluster->nr)
+		cluster->start = extent_key->objectid;
+	else
+		BUG_ON(cluster->nr >= MAX_EXTENTS);
+	cluster->end = extent_key->objectid + extent_key->offset - 1;
+	cluster->boundary[cluster->nr] = extent_key->objectid;
+	cluster->nr++;
+
+	if (cluster->nr >= MAX_EXTENTS) {
+		ret = relocate_file_extent_cluster(inode, cluster);
+		if (ret)
+			return ret;
+		cluster->nr = 0;
+	}
+	return 0;
 }
 
 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
@@ -3203,10 +3254,12 @@ static int check_extent_flags(u64 flags)
 	return 0;
 }
 
+
 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 {
 	struct rb_root blocks = RB_ROOT;
 	struct btrfs_key key;
+	struct file_extent_cluster *cluster;
 	struct btrfs_trans_handle *trans = NULL;
 	struct btrfs_path *path;
 	struct btrfs_extent_item *ei;
@@ -3216,10 +3269,17 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 	int ret;
 	int err = 0;
 
+	cluster = kzalloc(sizeof(*cluster), GFP_NOFS);
+	if (!cluster)
+		return -ENOMEM;
+
 	path = btrfs_alloc_path();
 	if (!path)
 		return -ENOMEM;
 
+	rc->extents_found = 0;
+	rc->extents_skipped = 0;
+
 	rc->search_start = rc->block_group->key.objectid;
 	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
 			  GFP_NOFS);
@@ -3306,14 +3366,15 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 		}
 
 		nr = trans->blocks_used;
-		btrfs_end_transaction_throttle(trans, rc->extent_root);
+		btrfs_end_transaction(trans, rc->extent_root);
 		trans = NULL;
 		btrfs_btree_balance_dirty(rc->extent_root, nr);
 
 		if (rc->stage == MOVE_DATA_EXTENTS &&
 		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
 			rc->found_file_extent = 1;
-			ret = relocate_data_extent(rc->data_inode, &key);
+			ret = relocate_data_extent(rc->data_inode,
+						   &key, cluster);
 			if (ret < 0) {
 				err = ret;
 				break;
@@ -3328,6 +3389,14 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 		btrfs_btree_balance_dirty(rc->extent_root, nr);
 	}
 
+	if (!err) {
+		ret = relocate_file_extent_cluster(rc->data_inode, cluster);
+		if (ret < 0)
+			err = ret;
+	}
+
+	kfree(cluster);
+
 	rc->create_reloc_root = 0;
 	smp_mb();
 
@@ -3348,8 +3417,7 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 }
 
 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
-				 struct btrfs_root *root,
-				 u64 objectid, u64 size)
+				 struct btrfs_root *root, u64 objectid)
 {
 	struct btrfs_path *path;
 	struct btrfs_inode_item *item;
@@ -3368,7 +3436,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
 	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
 	btrfs_set_inode_generation(leaf, item, 1);
-	btrfs_set_inode_size(leaf, item, size);
+	btrfs_set_inode_size(leaf, item, 0);
 	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
 	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
 	btrfs_mark_buffer_dirty(leaf);
@@ -3404,12 +3472,7 @@ static struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
 	if (err)
 		goto out;
 
-	err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
-	BUG_ON(err);
-
-	err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
-				       group->key.offset, 0, group->key.offset,
-				       0, 0, 0);
+	err = __insert_orphan_inode(trans, root, objectid);
 	BUG_ON(err);
 
 	key.objectid = objectid;
@@ -3475,14 +3538,15 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 	btrfs_wait_ordered_extents(fs_info->tree_root, 0);
 
 	while (1) {
-		mutex_lock(&fs_info->cleaner_mutex);
-		btrfs_clean_old_snapshots(fs_info->tree_root);
-		mutex_unlock(&fs_info->cleaner_mutex);
-
 		rc->extents_found = 0;
 		rc->extents_skipped = 0;
 
+		mutex_lock(&fs_info->cleaner_mutex);
+
+		btrfs_clean_old_snapshots(fs_info->tree_root);
 		ret = relocate_block_group(rc);
+
+		mutex_unlock(&fs_info->cleaner_mutex);
 		if (ret < 0) {
 			err = ret;
 			break;
@@ -3514,10 +3578,10 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 		}
 	}
 
-	filemap_fdatawrite_range(fs_info->btree_inode->i_mapping,
-				 rc->block_group->key.objectid,
-				 rc->block_group->key.objectid +
-				 rc->block_group->key.offset - 1);
+	filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
+				     rc->block_group->key.objectid,
+				     rc->block_group->key.objectid +
+				     rc->block_group->key.offset - 1);
 
 	WARN_ON(rc->block_group->pinned > 0);
 	WARN_ON(rc->block_group->reserved > 0);
@@ -3530,6 +3594,26 @@ out:
 	return err;
 }
 
+static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
+{
+	struct btrfs_trans_handle *trans;
+	int ret;
+
+	trans = btrfs_start_transaction(root->fs_info->tree_root, 1);
+
+	memset(&root->root_item.drop_progress, 0,
+		sizeof(root->root_item.drop_progress));
+	root->root_item.drop_level = 0;
+	btrfs_set_root_refs(&root->root_item, 0);
+	ret = btrfs_update_root(trans, root->fs_info->tree_root,
+				&root->root_key, &root->root_item);
+	BUG_ON(ret);
+
+	ret = btrfs_end_transaction(trans, root->fs_info->tree_root);
+	BUG_ON(ret);
+	return 0;
+}
+
 /*
  * recover relocation interrupted by system crash.
  *
@@ -3589,8 +3673,12 @@ int btrfs_recover_relocation(struct btrfs_root *root)
 			fs_root = read_fs_root(root->fs_info,
 					       reloc_root->root_key.offset);
 			if (IS_ERR(fs_root)) {
-				err = PTR_ERR(fs_root);
-				goto out;
+				ret = PTR_ERR(fs_root);
+				if (ret != -ENOENT) {
+					err = ret;
+					goto out;
+				}
+				mark_garbage_root(reloc_root);
 			}
 		}
 

fs/btrfs/root-tree.c

@@ -94,17 +94,23 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid,
 		goto out;
 
 	BUG_ON(ret == 0);
+	if (path->slots[0] == 0) {
+		ret = 1;
+		goto out;
+	}
 	l = path->nodes[0];
-	BUG_ON(path->slots[0] == 0);
 	slot = path->slots[0] - 1;
 	btrfs_item_key_to_cpu(l, &found_key, slot);
-	if (found_key.objectid != objectid) {
+	if (found_key.objectid != objectid ||
+	    found_key.type != BTRFS_ROOT_ITEM_KEY) {
 		ret = 1;
 		goto out;
 	}
-	read_extent_buffer(l, item, btrfs_item_ptr_offset(l, slot),
-			   sizeof(*item));
-	memcpy(key, &found_key, sizeof(found_key));
+	if (item)
+		read_extent_buffer(l, item, btrfs_item_ptr_offset(l, slot),
+				   sizeof(*item));
+	if (key)
+		memcpy(key, &found_key, sizeof(found_key));
 	ret = 0;
 out:
 	btrfs_free_path(path);
@@ -249,6 +255,59 @@ err:
 	return ret;
 }
 
+int btrfs_find_orphan_roots(struct btrfs_root *tree_root)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int err = 0;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	key.objectid = BTRFS_ORPHAN_OBJECTID;
+	key.type = BTRFS_ORPHAN_ITEM_KEY;
+	key.offset = 0;
+
+	while (1) {
+		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
+		if (ret < 0) {
+			err = ret;
+			break;
+		}
+
+		leaf = path->nodes[0];
+		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
+			ret = btrfs_next_leaf(tree_root, path);
+			if (ret < 0)
+				err = ret;
+			if (ret != 0)
+				break;
+			leaf = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+		btrfs_release_path(tree_root, path);
+
+		if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
+		    key.type != BTRFS_ORPHAN_ITEM_KEY)
+			break;
+
+		ret = btrfs_find_dead_roots(tree_root, key.offset);
+		if (ret) {
+			err = ret;
+			break;
+		}
+
+		key.offset++;
+	}
+
+	btrfs_free_path(path);
+	return err;
+}
+
 /* drop the root item for 'key' from 'root' */
 int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		   struct btrfs_key *key)
@@ -278,31 +337,57 @@ out:
 	return ret;
 }
 
-#if 0 /* this will get used when snapshot deletion is implemented */
 int btrfs_del_root_ref(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *tree_root,
-		       u64 root_id, u8 type, u64 ref_id)
+		       u64 root_id, u64 ref_id, u64 dirid, u64 *sequence,
+		       const char *name, int name_len)
+
 {
+	struct btrfs_path *path;
+	struct btrfs_root_ref *ref;
+	struct extent_buffer *leaf;
 	struct btrfs_key key;
+	unsigned long ptr;
+	int err = 0;
 	int ret;
-	struct btrfs_path *path;
 
 	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
 
 	key.objectid = root_id;
-	key.type = type;
+	key.type = BTRFS_ROOT_BACKREF_KEY;
 	key.offset = ref_id;
-
+again:
 	ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
-	BUG_ON(ret);
-
-	ret = btrfs_del_item(trans, tree_root, path);
-	BUG_ON(ret);
+	BUG_ON(ret < 0);
+	if (ret == 0) {
+		leaf = path->nodes[0];
+		ref = btrfs_item_ptr(leaf, path->slots[0],
+				     struct btrfs_root_ref);
+
+		WARN_ON(btrfs_root_ref_dirid(leaf, ref) != dirid);
+		WARN_ON(btrfs_root_ref_name_len(leaf, ref) != name_len);
+		ptr = (unsigned long)(ref + 1);
+		WARN_ON(memcmp_extent_buffer(leaf, name, ptr, name_len));
+		*sequence = btrfs_root_ref_sequence(leaf, ref);
+
+		ret = btrfs_del_item(trans, tree_root, path);
+		BUG_ON(ret);
+	} else
+		err = -ENOENT;
+
+	if (key.type == BTRFS_ROOT_BACKREF_KEY) {
+		btrfs_release_path(tree_root, path);
+		key.objectid = ref_id;
+		key.type = BTRFS_ROOT_REF_KEY;
+		key.offset = root_id;
+		goto again;
+	}
 
 	btrfs_free_path(path);
-	return ret;
+	return err;
 }
-#endif
 
 int btrfs_find_root_ref(struct btrfs_root *tree_root,
 		   struct btrfs_path *path,
@@ -319,7 +404,6 @@ int btrfs_find_root_ref(struct btrfs_root *tree_root,
 	return ret;
 }
 
-
 /*
  * add a btrfs_root_ref item.  type is either BTRFS_ROOT_REF_KEY
  * or BTRFS_ROOT_BACKREF_KEY.
@@ -335,8 +419,7 @@ int btrfs_find_root_ref(struct btrfs_root *tree_root,
  */
 int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *tree_root,
-		       u64 root_id, u8 type, u64 ref_id,
-		       u64 dirid, u64 sequence,
+		       u64 root_id, u64 ref_id, u64 dirid, u64 sequence,
 		       const char *name, int name_len)
 {
 	struct btrfs_key key;
@@ -346,13 +429,14 @@ int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
 	struct extent_buffer *leaf;
 	unsigned long ptr;
 
-
 	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
 
 	key.objectid = root_id;
-	key.type = type;
+	key.type = BTRFS_ROOT_BACKREF_KEY;
 	key.offset = ref_id;
-
+again:
 	ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
 				      sizeof(*ref) + name_len);
 	BUG_ON(ret);
@@ -366,6 +450,14 @@ int btrfs_add_root_ref(struct btrfs_trans_handle *trans,
 	write_extent_buffer(leaf, name, ptr, name_len);
 	btrfs_mark_buffer_dirty(leaf);
 
+	if (key.type == BTRFS_ROOT_BACKREF_KEY) {
+		btrfs_release_path(tree_root, path);
+		key.objectid = ref_id;
+		key.type = BTRFS_ROOT_REF_KEY;
+		key.offset = root_id;
+		goto again;
+	}
+
 	btrfs_free_path(path);
-	return ret;
+	return 0;
 }

fs/btrfs/super.c

@@ -676,6 +676,7 @@ static int btrfs_unfreeze(struct super_block *sb)
 }
 
 static const struct super_operations btrfs_super_ops = {
+	.drop_inode	= btrfs_drop_inode,
 	.delete_inode	= btrfs_delete_inode,
 	.put_super	= btrfs_put_super,
 	.sync_fs	= btrfs_sync_fs,

fs/btrfs/transaction.c

@@ -104,7 +104,6 @@ static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
 {
 	if (root->ref_cows && root->last_trans < trans->transid) {
 		WARN_ON(root == root->fs_info->extent_root);
-		WARN_ON(root->root_item.refs == 0);
 		WARN_ON(root->commit_root != root->node);
 
 		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
@@ -720,7 +719,8 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 
 	key.objectid = objectid;
-	key.offset = 0;
+	/* record when the snapshot was created in key.offset */
+	key.offset = trans->transid;
 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 
 	old = btrfs_lock_root_node(root);
@@ -778,24 +778,14 @@ static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
 	ret = btrfs_update_inode(trans, parent_root, parent_inode);
 	BUG_ON(ret);
 
-	/* add the backref first */
 	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
 				 pending->root_key.objectid,
-				 BTRFS_ROOT_BACKREF_KEY,
 				 parent_root->root_key.objectid,
 				 parent_inode->i_ino, index, pending->name,
 				 namelen);
 
 	BUG_ON(ret);
 
-	/* now add the forward ref */
-	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
-				 parent_root->root_key.objectid,
-				 BTRFS_ROOT_REF_KEY,
-				 pending->root_key.objectid,
-				 parent_inode->i_ino, index, pending->name,
-				 namelen);
-
 	inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
 	d_instantiate(pending->dentry, inode);
 fail:
@@ -874,7 +864,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	unsigned long timeout = 1;
 	struct btrfs_transaction *cur_trans;
 	struct btrfs_transaction *prev_trans = NULL;
-	struct extent_io_tree *pinned_copy;
 	DEFINE_WAIT(wait);
 	int ret;
 	int should_grow = 0;
@@ -915,13 +904,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 		return 0;
 	}
 
-	pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
-	if (!pinned_copy)
-		return -ENOMEM;
-
-	extent_io_tree_init(pinned_copy,
-			     root->fs_info->btree_inode->i_mapping, GFP_NOFS);
-
 	trans->transaction->in_commit = 1;
 	trans->transaction->blocked = 1;
 	if (cur_trans->list.prev != &root->fs_info->trans_list) {
@@ -1019,6 +1001,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	ret = commit_cowonly_roots(trans, root);
 	BUG_ON(ret);
 
+	btrfs_prepare_extent_commit(trans, root);
+
 	cur_trans = root->fs_info->running_transaction;
 	spin_lock(&root->fs_info->new_trans_lock);
 	root->fs_info->running_transaction = NULL;
@@ -1042,8 +1026,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 	       sizeof(root->fs_info->super_copy));
 
-	btrfs_copy_pinned(root, pinned_copy);
-
 	trans->transaction->blocked = 0;
 
 	wake_up(&root->fs_info->transaction_wait);
@@ -1059,8 +1041,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	 */
 	mutex_unlock(&root->fs_info->tree_log_mutex);
 
-	btrfs_finish_extent_commit(trans, root, pinned_copy);
-	kfree(pinned_copy);
+	btrfs_finish_extent_commit(trans, root);
 
 	/* do the directory inserts of any pending snapshot creations */
 	finish_pending_snapshots(trans, root->fs_info);
@@ -1096,8 +1077,13 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
 
 	while (!list_empty(&list)) {
 		root = list_entry(list.next, struct btrfs_root, root_list);
-		list_del_init(&root->root_list);
-		btrfs_drop_snapshot(root, 0);
+		list_del(&root->root_list);
+
+		if (btrfs_header_backref_rev(root->node) <
+		    BTRFS_MIXED_BACKREF_REV)
+			btrfs_drop_snapshot(root, 0);
+		else
+			btrfs_drop_snapshot(root, 1);
 	}
 	return 0;
 }

fs/btrfs/tree-log.c

@@ -263,8 +263,8 @@ static int process_one_buffer(struct btrfs_root *log,
 			      struct walk_control *wc, u64 gen)
 {
 	if (wc->pin)
-		btrfs_update_pinned_extents(log->fs_info->extent_root,
-					    eb->start, eb->len, 1);
+		btrfs_pin_extent(log->fs_info->extent_root,
+				 eb->start, eb->len, 0);
 
 	if (btrfs_buffer_uptodate(eb, gen)) {
 		if (wc->write)
@@ -534,7 +534,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 	saved_nbytes = inode_get_bytes(inode);
 	/* drop any overlapping extents */
 	ret = btrfs_drop_extents(trans, root, inode,
-			 start, extent_end, extent_end, start, &alloc_hint);
+			 start, extent_end, extent_end, start, &alloc_hint, 1);
 	BUG_ON(ret);
 
 	if (found_type == BTRFS_FILE_EXTENT_REG ||
@@ -2841,7 +2841,7 @@ static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
 		if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
 			break;
 
-		if (parent == sb->s_root)
+		if (IS_ROOT(parent))
 			break;
 
 		parent = parent->d_parent;
@@ -2880,6 +2880,12 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 		goto end_no_trans;
 	}
 
+	if (root != BTRFS_I(inode)->root ||
+	    btrfs_root_refs(&root->root_item) == 0) {
+		ret = 1;
+		goto end_no_trans;
+	}
+
 	ret = check_parent_dirs_for_sync(trans, inode, parent,
 					 sb, last_committed);
 	if (ret)
@@ -2907,12 +2913,15 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 			break;
 
 		inode = parent->d_inode;
+		if (root != BTRFS_I(inode)->root)
+			break;
+
 		if (BTRFS_I(inode)->generation >
 		    root->fs_info->last_trans_committed) {
 			ret = btrfs_log_inode(trans, root, inode, inode_only);
 			BUG_ON(ret);
 		}
-		if (parent == sb->s_root)
+		if (IS_ROOT(parent))
 			break;
 
 		parent = parent->d_parent;
@@ -2951,7 +2960,6 @@ int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
 	struct btrfs_key tmp_key;
 	struct btrfs_root *log;
 	struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
-	u64 highest_inode;
 	struct walk_control wc = {
 		.process_func = process_one_buffer,
 		.stage = 0,
@@ -3010,11 +3018,6 @@ again:
 						      path);
 			BUG_ON(ret);
 		}
-		ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
-		if (ret == 0) {
-			wc.replay_dest->highest_inode = highest_inode;
-			wc.replay_dest->last_inode_alloc = highest_inode;
-		}
 
 		key.offset = found_key.offset - 1;
 		wc.replay_dest->log_root = NULL;

fs/btrfs/volumes.c

@@ -276,7 +276,7 @@ loop_lock:
 		 * is now congested.  Back off and let other work structs
 		 * run instead
 		 */
-		if (pending && bdi_write_congested(bdi) && batch_run > 32 &&
+		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 		    fs_info->fs_devices->open_devices > 1) {
 			struct io_context *ioc;
 
@@ -719,10 +719,9 @@ error:
  * called very infrequently and that a given device has a small number
  * of extents
  */
-static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
-					 struct btrfs_device *device,
-					 u64 num_bytes, u64 *start,
-					 u64 *max_avail)
+int find_free_dev_extent(struct btrfs_trans_handle *trans,
+			 struct btrfs_device *device, u64 num_bytes,
+			 u64 *start, u64 *max_avail)
 {
 	struct btrfs_key key;
 	struct btrfs_root *root = device->dev_root;
@@ -1736,6 +1735,10 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
 	extent_root = root->fs_info->extent_root;
 	em_tree = &root->fs_info->mapping_tree.map_tree;
 
+	ret = btrfs_can_relocate(extent_root, chunk_offset);
+	if (ret)
+		return -ENOSPC;
+
 	/* step one, relocate all the extents inside this chunk */
 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
 	BUG_ON(ret);
@@ -1749,9 +1752,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
 	 * step two, delete the device extents and the
 	 * chunk tree entries
 	 */
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 
 	BUG_ON(em->start > chunk_offset ||
 	       em->start + em->len < chunk_offset);
@@ -1780,9 +1783,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root,
 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
 	BUG_ON(ret);
 
-	spin_lock(&em_tree->lock);
+	write_lock(&em_tree->lock);
 	remove_extent_mapping(em_tree, em);
-	spin_unlock(&em_tree->lock);
+	write_unlock(&em_tree->lock);
 
 	kfree(map);
 	em->bdev = NULL;
@@ -1807,12 +1810,15 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
 	struct btrfs_key found_key;
 	u64 chunk_tree = chunk_root->root_key.objectid;
 	u64 chunk_type;
+	bool retried = false;
+	int failed = 0;
 	int ret;
 
 	path = btrfs_alloc_path();
 	if (!path)
 		return -ENOMEM;
 
+again:
 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 	key.offset = (u64)-1;
 	key.type = BTRFS_CHUNK_ITEM_KEY;
@@ -1842,7 +1848,10 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
 						   found_key.objectid,
 						   found_key.offset);
-			BUG_ON(ret);
+			if (ret == -ENOSPC)
+				failed++;
+			else if (ret)
+				BUG();
 		}
 
 		if (found_key.offset == 0)
@@ -1850,6 +1859,14 @@ static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
 		key.offset = found_key.offset - 1;
 	}
 	ret = 0;
+	if (failed && !retried) {
+		failed = 0;
+		retried = true;
+		goto again;
+	} else if (failed && retried) {
+		WARN_ON(1);
+		ret = -ENOSPC;
+	}
 error:
 	btrfs_free_path(path);
 	return ret;
@@ -1894,6 +1911,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
 			continue;
 
 		ret = btrfs_shrink_device(device, old_size - size_to_free);
+		if (ret == -ENOSPC)
+			break;
 		BUG_ON(ret);
 
 		trans = btrfs_start_transaction(dev_root, 1);
@@ -1938,9 +1957,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
 		chunk = btrfs_item_ptr(path->nodes[0],
 				       path->slots[0],
 				       struct btrfs_chunk);
-		key.offset = found_key.offset;
 		/* chunk zero is special */
-		if (key.offset == 0)
+		if (found_key.offset == 0)
 			break;
 
 		btrfs_release_path(chunk_root, path);
@@ -1948,7 +1966,8 @@ int btrfs_balance(struct btrfs_root *dev_root)
 					   chunk_root->root_key.objectid,
 					   found_key.objectid,
 					   found_key.offset);
-		BUG_ON(ret);
+		BUG_ON(ret && ret != -ENOSPC);
+		key.offset = found_key.offset - 1;
 	}
 	ret = 0;
 error:
@@ -1974,10 +1993,13 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 	u64 chunk_offset;
 	int ret;
 	int slot;
+	int failed = 0;
+	bool retried = false;
 	struct extent_buffer *l;
 	struct btrfs_key key;
 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 	u64 old_total = btrfs_super_total_bytes(super_copy);
+	u64 old_size = device->total_bytes;
 	u64 diff = device->total_bytes - new_size;
 
 	if (new_size >= device->total_bytes)
@@ -1987,12 +2009,6 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 	if (!path)
 		return -ENOMEM;
 
-	trans = btrfs_start_transaction(root, 1);
-	if (!trans) {
-		ret = -ENOMEM;
-		goto done;
-	}
-
 	path->reada = 2;
 
 	lock_chunks(root);
@@ -2001,8 +2017,8 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 	if (device->writeable)
 		device->fs_devices->total_rw_bytes -= diff;
 	unlock_chunks(root);
-	btrfs_end_transaction(trans, root);
 
+again:
 	key.objectid = device->devid;
 	key.offset = (u64)-1;
 	key.type = BTRFS_DEV_EXTENT_KEY;
@@ -2017,6 +2033,7 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 			goto done;
 		if (ret) {
 			ret = 0;
+			btrfs_release_path(root, path);
 			break;
 		}
 
@@ -2024,14 +2041,18 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 		slot = path->slots[0];
 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 
-		if (key.objectid != device->devid)
+		if (key.objectid != device->devid) {
+			btrfs_release_path(root, path);
 			break;
+		}
 
 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 		length = btrfs_dev_extent_length(l, dev_extent);
 
-		if (key.offset + length <= new_size)
+		if (key.offset + length <= new_size) {
+			btrfs_release_path(root, path);
 			break;
+		}
 
 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
@@ -2040,8 +2061,26 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 
 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
 					   chunk_offset);
-		if (ret)
+		if (ret && ret != -ENOSPC)
 			goto done;
+		if (ret == -ENOSPC)
+			failed++;
+		key.offset -= 1;
+	}
+
+	if (failed && !retried) {
+		failed = 0;
+		retried = true;
+		goto again;
+	} else if (failed && retried) {
+		ret = -ENOSPC;
+		lock_chunks(root);
+
+		device->total_bytes = old_size;
+		if (device->writeable)
+			device->fs_devices->total_rw_bytes += diff;
+		unlock_chunks(root);
+		goto done;
 	}
 
 	/* Shrinking succeeded, else we would be at "done". */
@@ -2294,9 +2333,9 @@ again:
 	em->block_len = em->len;
 
 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
-	spin_lock(&em_tree->lock);
+	write_lock(&em_tree->lock);
 	ret = add_extent_mapping(em_tree, em);
-	spin_unlock(&em_tree->lock);
+	write_unlock(&em_tree->lock);
 	BUG_ON(ret);
 	free_extent_map(em);
 
@@ -2491,9 +2530,9 @@ int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
 	int readonly = 0;
 	int i;
 
-	spin_lock(&map_tree->map_tree.lock);
+	read_lock(&map_tree->map_tree.lock);
 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
-	spin_unlock(&map_tree->map_tree.lock);
+	read_unlock(&map_tree->map_tree.lock);
 	if (!em)
 		return 1;
 
@@ -2518,11 +2557,11 @@ void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
 	struct extent_map *em;
 
 	while (1) {
-		spin_lock(&tree->map_tree.lock);
+		write_lock(&tree->map_tree.lock);
 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
 		if (em)
 			remove_extent_mapping(&tree->map_tree, em);
-		spin_unlock(&tree->map_tree.lock);
+		write_unlock(&tree->map_tree.lock);
 		if (!em)
 			break;
 		kfree(em->bdev);
@@ -2540,9 +2579,9 @@ int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
 	struct extent_map_tree *em_tree = &map_tree->map_tree;
 	int ret;
 
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree, logical, len);
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 	BUG_ON(!em);
 
 	BUG_ON(em->start > logical || em->start + em->len < logical);
@@ -2604,9 +2643,9 @@ again:
 		atomic_set(&multi->error, 0);
 	}
 
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree, logical, *length);
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 
 	if (!em && unplug_page)
 		return 0;
@@ -2763,9 +2802,9 @@ int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
 	u64 stripe_nr;
 	int i, j, nr = 0;
 
-	spin_lock(&em_tree->lock);
+	read_lock(&em_tree->lock);
 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
-	spin_unlock(&em_tree->lock);
+	read_unlock(&em_tree->lock);
 
 	BUG_ON(!em || em->start != chunk_start);
 	map = (struct map_lookup *)em->bdev;
@@ -3053,9 +3092,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
 	logical = key->offset;
 	length = btrfs_chunk_length(leaf, chunk);
 
-	spin_lock(&map_tree->map_tree.lock);
+	read_lock(&map_tree->map_tree.lock);
 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
-	spin_unlock(&map_tree->map_tree.lock);
+	read_unlock(&map_tree->map_tree.lock);
 
 	/* already mapped? */
 	if (em && em->start <= logical && em->start + em->len > logical) {
@@ -3114,9 +3153,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
 		map->stripes[i].dev->in_fs_metadata = 1;
 	}
 
-	spin_lock(&map_tree->map_tree.lock);
+	write_lock(&map_tree->map_tree.lock);
 	ret = add_extent_mapping(&map_tree->map_tree, em);
-	spin_unlock(&map_tree->map_tree.lock);
+	write_unlock(&map_tree->map_tree.lock);
 	BUG_ON(ret);
 	free_extent_map(em);
 

fs/btrfs/volumes.h

@@ -181,4 +181,7 @@ int btrfs_balance(struct btrfs_root *dev_root);
 void btrfs_unlock_volumes(void);
 void btrfs_lock_volumes(void);
 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset);
+int find_free_dev_extent(struct btrfs_trans_handle *trans,
+			 struct btrfs_device *device, u64 num_bytes,
+			 u64 *start, u64 *max_avail);
 #endif