Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable

* git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable:
  Btrfs: fix btrfs fallocate oops and deadlock
  Btrfs: use the right node in reada_for_balance
  Btrfs: fix oops on page->mapping->host during writepage
  Btrfs: add a priority queue to the async thread helpers
  Btrfs: use WRITE_SYNC for synchronous writes

fs/btrfs/async-thread.c

@@ -25,6 +25,7 @@
 #define WORK_QUEUED_BIT 0
 #define WORK_DONE_BIT 1
 #define WORK_ORDER_DONE_BIT 2
+#define WORK_HIGH_PRIO_BIT 3
 
 /*
  * container for the kthread task pointer and the list of pending work
@@ -36,6 +37,7 @@ struct btrfs_worker_thread {
 
 	/* list of struct btrfs_work that are waiting for service */
 	struct list_head pending;
+	struct list_head prio_pending;
 
 	/* list of worker threads from struct btrfs_workers */
 	struct list_head worker_list;
@@ -103,10 +105,16 @@ static noinline int run_ordered_completions(struct btrfs_workers *workers,
 
 	spin_lock_irqsave(&workers->lock, flags);
 
-	while (!list_empty(&workers->order_list)) {
-		work = list_entry(workers->order_list.next,
-				  struct btrfs_work, order_list);
-
+	while (1) {
+		if (!list_empty(&workers->prio_order_list)) {
+			work = list_entry(workers->prio_order_list.next,
+					  struct btrfs_work, order_list);
+		} else if (!list_empty(&workers->order_list)) {
+			work = list_entry(workers->order_list.next,
+					  struct btrfs_work, order_list);
+		} else {
+			break;
+		}
 		if (!test_bit(WORK_DONE_BIT, &work->flags))
 			break;
 
@@ -143,8 +151,14 @@ static int worker_loop(void *arg)
 	do {
 		spin_lock_irq(&worker->lock);
 again_locked:
-		while (!list_empty(&worker->pending)) {
-			cur = worker->pending.next;
+		while (1) {
+			if (!list_empty(&worker->prio_pending))
+				cur = worker->prio_pending.next;
+			else if (!list_empty(&worker->pending))
+				cur = worker->pending.next;
+			else
+				break;
+
 			work = list_entry(cur, struct btrfs_work, list);
 			list_del(&work->list);
 			clear_bit(WORK_QUEUED_BIT, &work->flags);
@@ -163,7 +177,6 @@ again_locked:
 
 			spin_lock_irq(&worker->lock);
 			check_idle_worker(worker);
-
 		}
 		if (freezing(current)) {
 			worker->working = 0;
@@ -178,7 +191,8 @@ again_locked:
 				 * jump_in?
 				 */
 				smp_mb();
-				if (!list_empty(&worker->pending))
+				if (!list_empty(&worker->pending) ||
+				    !list_empty(&worker->prio_pending))
 					continue;
 
 				/*
@@ -191,7 +205,8 @@ again_locked:
 				 */
 				schedule_timeout(1);
 				smp_mb();
-				if (!list_empty(&worker->pending))
+				if (!list_empty(&worker->pending) ||
+				    !list_empty(&worker->prio_pending))
 					continue;
 
 				if (kthread_should_stop())
@@ -200,7 +215,8 @@ again_locked:
 				/* still no more work?, sleep for real */
 				spin_lock_irq(&worker->lock);
 				set_current_state(TASK_INTERRUPTIBLE);
-				if (!list_empty(&worker->pending))
+				if (!list_empty(&worker->pending) ||
+				    !list_empty(&worker->prio_pending))
 					goto again_locked;
 
 				/*
@@ -248,6 +264,7 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
 	INIT_LIST_HEAD(&workers->worker_list);
 	INIT_LIST_HEAD(&workers->idle_list);
 	INIT_LIST_HEAD(&workers->order_list);
+	INIT_LIST_HEAD(&workers->prio_order_list);
 	spin_lock_init(&workers->lock);
 	workers->max_workers = max;
 	workers->idle_thresh = 32;
@@ -273,6 +290,7 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
 		}
 
 		INIT_LIST_HEAD(&worker->pending);
+		INIT_LIST_HEAD(&worker->prio_pending);
 		INIT_LIST_HEAD(&worker->worker_list);
 		spin_lock_init(&worker->lock);
 		atomic_set(&worker->num_pending, 0);
@@ -396,7 +414,10 @@ int btrfs_requeue_work(struct btrfs_work *work)
 		goto out;
 
 	spin_lock_irqsave(&worker->lock, flags);
-	list_add_tail(&work->list, &worker->pending);
+	if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags))
+		list_add_tail(&work->list, &worker->prio_pending);
+	else
+		list_add_tail(&work->list, &worker->pending);
 	atomic_inc(&worker->num_pending);
 
 	/* by definition we're busy, take ourselves off the idle
@@ -422,6 +443,11 @@ out:
 	return 0;
 }
 
+void btrfs_set_work_high_prio(struct btrfs_work *work)
+{
+	set_bit(WORK_HIGH_PRIO_BIT, &work->flags);
+}
+
 /*
  * places a struct btrfs_work into the pending queue of one of the kthreads
  */
@@ -438,7 +464,12 @@ 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);
-		list_add_tail(&work->order_list, &workers->order_list);
+		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);
 	} else {
 		INIT_LIST_HEAD(&work->order_list);
@@ -446,7 +477,10 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
 
 	spin_lock_irqsave(&worker->lock, flags);
 
-	list_add_tail(&work->list, &worker->pending);
+	if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags))
+		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);
 

fs/btrfs/async-thread.h

@@ -85,6 +85,7 @@ struct btrfs_workers {
 	 * of work items waiting for completion
 	 */
 	struct list_head order_list;
+	struct list_head prio_order_list;
 
 	/* lock for finding the next worker thread to queue on */
 	spinlock_t lock;
@@ -98,4 +99,5 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers);
 int btrfs_stop_workers(struct btrfs_workers *workers);
 void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max);
 int btrfs_requeue_work(struct btrfs_work *work);
+void btrfs_set_work_high_prio(struct btrfs_work *work);
 #endif

fs/btrfs/ctree.c

@@ -1325,12 +1325,12 @@ static noinline int reada_for_balance(struct btrfs_root *root,
 	int ret = 0;
 	int blocksize;
 
-	parent = path->nodes[level - 1];
+	parent = path->nodes[level + 1];
 	if (!parent)
 		return 0;
 
 	nritems = btrfs_header_nritems(parent);
-	slot = path->slots[level];
+	slot = path->slots[level + 1];
 	blocksize = btrfs_level_size(root, level);
 
 	if (slot > 0) {
@@ -1341,7 +1341,7 @@ static noinline int reada_for_balance(struct btrfs_root *root,
 			block1 = 0;
 		free_extent_buffer(eb);
 	}
-	if (slot < nritems) {
+	if (slot + 1 < nritems) {
 		block2 = btrfs_node_blockptr(parent, slot + 1);
 		gen = btrfs_node_ptr_generation(parent, slot + 1);
 		eb = btrfs_find_tree_block(root, block2, blocksize);
@@ -1351,7 +1351,11 @@ static noinline int reada_for_balance(struct btrfs_root *root,
 	}
 	if (block1 || block2) {
 		ret = -EAGAIN;
+
+		/* release the whole path */
 		btrfs_release_path(root, path);
+
+		/* read the blocks */
 		if (block1)
 			readahead_tree_block(root, block1, blocksize, 0);
 		if (block2)
@@ -1361,7 +1365,7 @@ static noinline int reada_for_balance(struct btrfs_root *root,
 			eb = read_tree_block(root, block1, blocksize, 0);
 			free_extent_buffer(eb);
 		}
-		if (block1) {
+		if (block2) {
 			eb = read_tree_block(root, block2, blocksize, 0);
 			free_extent_buffer(eb);
 		}
@@ -1481,12 +1485,15 @@ read_block_for_search(struct btrfs_trans_handle *trans,
 	 * of the btree by dropping locks before
 	 * we read.
 	 */
-	btrfs_release_path(NULL, p);
+	btrfs_unlock_up_safe(p, level + 1);
+	btrfs_set_path_blocking(p);
+
 	if (tmp)
 		free_extent_buffer(tmp);
 	if (p->reada)
 		reada_for_search(root, p, level, slot, key->objectid);
 
+	btrfs_release_path(NULL, p);
 	tmp = read_tree_block(root, blocknr, blocksize, gen);
 	if (tmp)
 		free_extent_buffer(tmp);

fs/btrfs/disk-io.c

@@ -579,6 +579,10 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 	async->bio_flags = bio_flags;
 
 	atomic_inc(&fs_info->nr_async_submits);
+
+	if (rw & (1 << BIO_RW_SYNCIO))
+		btrfs_set_work_high_prio(&async->work);
+
 	btrfs_queue_worker(&fs_info->workers, &async->work);
 #if 0
 	int limit = btrfs_async_submit_limit(fs_info);
@@ -656,6 +660,7 @@ static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 		return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 				     mirror_num, 0);
 	}
+
 	/*
 	 * kthread helpers are used to submit writes so that checksumming
 	 * can happen in parallel across all CPUs
@@ -2095,10 +2100,10 @@ static int write_dev_supers(struct btrfs_device *device,
 				device->barriers = 0;
 				get_bh(bh);
 				lock_buffer(bh);
-				ret = submit_bh(WRITE, bh);
+				ret = submit_bh(WRITE_SYNC, bh);
 			}
 		} else {
-			ret = submit_bh(WRITE, bh);
+			ret = submit_bh(WRITE_SYNC, bh);
 		}
 
 		if (!ret && wait) {

fs/btrfs/extent_io.c

@@ -50,7 +50,10 @@ struct extent_page_data {
 	/* tells writepage not to lock the state bits for this range
 	 * it still does the unlocking
 	 */
-	int extent_locked;
+	unsigned int extent_locked:1;
+
+	/* tells the submit_bio code to use a WRITE_SYNC */
+	unsigned int sync_io:1;
 };
 
 int __init extent_io_init(void)
@@ -2101,6 +2104,16 @@ int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
 	return ret;
 }
 
+static noinline void update_nr_written(struct page *page,
+				      struct writeback_control *wbc,
+				      unsigned long nr_written)
+{
+	wbc->nr_to_write -= nr_written;
+	if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
+	    wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
+		page->mapping->writeback_index = page->index + nr_written;
+}
+
 /*
  * the writepage semantics are similar to regular writepage.  extent
  * records are inserted to lock ranges in the tree, and as dirty areas
@@ -2136,8 +2149,14 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 	u64 delalloc_end;
 	int page_started;
 	int compressed;
+	int write_flags;
 	unsigned long nr_written = 0;
 
+	if (wbc->sync_mode == WB_SYNC_ALL)
+		write_flags = WRITE_SYNC_PLUG;
+	else
+		write_flags = WRITE;
+
 	WARN_ON(!PageLocked(page));
 	pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
 	if (page->index > end_index ||
@@ -2164,6 +2183,12 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 	delalloc_end = 0;
 	page_started = 0;
 	if (!epd->extent_locked) {
+		/*
+		 * make sure the wbc mapping index is at least updated
+		 * to this page.
+		 */
+		update_nr_written(page, wbc, 0);
+
 		while (delalloc_end < page_end) {
 			nr_delalloc = find_lock_delalloc_range(inode, tree,
 						       page,
@@ -2185,7 +2210,13 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 		 */
 		if (page_started) {
 			ret = 0;
-			goto update_nr_written;
+			/*
+			 * we've unlocked the page, so we can't update
+			 * the mapping's writeback index, just update
+			 * nr_to_write.
+			 */
+			wbc->nr_to_write -= nr_written;
+			goto done_unlocked;
 		}
 	}
 	lock_extent(tree, start, page_end, GFP_NOFS);
@@ -2198,13 +2229,18 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 		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);
 			ret = 0;
-			goto update_nr_written;
+			goto done_unlocked;
 		}
 	}
 
-	nr_written++;
+	/*
+	 * we don't want to touch the inode after unlocking the page,
+	 * so we update the mapping writeback index now
+	 */
+	update_nr_written(page, wbc, nr_written + 1);
 
 	end = page_end;
 	if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
@@ -2314,9 +2350,9 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 				       (unsigned long long)end);
 			}
 
-			ret = submit_extent_page(WRITE, tree, page, sector,
-						 iosize, pg_offset, bdev,
-						 &epd->bio, max_nr,
+			ret = submit_extent_page(write_flags, tree, page,
+						 sector, iosize, pg_offset,
+						 bdev, &epd->bio, max_nr,
 						 end_bio_extent_writepage,
 						 0, 0, 0);
 			if (ret)
@@ -2336,11 +2372,8 @@ done:
 		unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
 	unlock_page(page);
 
-update_nr_written:
-	wbc->nr_to_write -= nr_written;
-	if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
-	    wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
-		page->mapping->writeback_index = page->index + nr_written;
+done_unlocked:
+
 	return 0;
 }
 
@@ -2460,15 +2493,23 @@ retry:
 	return ret;
 }
 
-static noinline void flush_write_bio(void *data)
+static void flush_epd_write_bio(struct extent_page_data *epd)
 {
-	struct extent_page_data *epd = data;
 	if (epd->bio) {
-		submit_one_bio(WRITE, epd->bio, 0, 0);
+		if (epd->sync_io)
+			submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
+		else
+			submit_one_bio(WRITE, epd->bio, 0, 0);
 		epd->bio = NULL;
 	}
 }
 
+static noinline void flush_write_bio(void *data)
+{
+	struct extent_page_data *epd = data;
+	flush_epd_write_bio(epd);
+}
+
 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
 			  get_extent_t *get_extent,
 			  struct writeback_control *wbc)
@@ -2480,23 +2521,22 @@ int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
 		.tree = tree,
 		.get_extent = get_extent,
 		.extent_locked = 0,
+		.sync_io = wbc->sync_mode == WB_SYNC_ALL,
 	};
 	struct writeback_control wbc_writepages = {
 		.bdi		= wbc->bdi,
-		.sync_mode	= WB_SYNC_NONE,
+		.sync_mode	= wbc->sync_mode,
 		.older_than_this = NULL,
 		.nr_to_write	= 64,
 		.range_start	= page_offset(page) + PAGE_CACHE_SIZE,
 		.range_end	= (loff_t)-1,
 	};
 
-
 	ret = __extent_writepage(page, wbc, &epd);
 
 	extent_write_cache_pages(tree, mapping, &wbc_writepages,
 				 __extent_writepage, &epd, flush_write_bio);
-	if (epd.bio)
-		submit_one_bio(WRITE, epd.bio, 0, 0);
+	flush_epd_write_bio(&epd);
 	return ret;
 }
 
@@ -2515,6 +2555,7 @@ int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
 		.tree = tree,
 		.get_extent = get_extent,
 		.extent_locked = 1,
+		.sync_io = mode == WB_SYNC_ALL,
 	};
 	struct writeback_control wbc_writepages = {
 		.bdi		= inode->i_mapping->backing_dev_info,
@@ -2540,8 +2581,7 @@ int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
 		start += PAGE_CACHE_SIZE;
 	}
 
-	if (epd.bio)
-		submit_one_bio(WRITE, epd.bio, 0, 0);
+	flush_epd_write_bio(&epd);
 	return ret;
 }
 
@@ -2556,13 +2596,13 @@ int extent_writepages(struct extent_io_tree *tree,
 		.tree = tree,
 		.get_extent = get_extent,
 		.extent_locked = 0,
+		.sync_io = wbc->sync_mode == WB_SYNC_ALL,
 	};
 
 	ret = extent_write_cache_pages(tree, mapping, wbc,
 				       __extent_writepage, &epd,
 				       flush_write_bio);
-	if (epd.bio)
-		submit_one_bio(WRITE, epd.bio, 0, 0);
+	flush_epd_write_bio(&epd);
 	return ret;
 }
 

fs/btrfs/file.c

@@ -830,7 +830,7 @@ again:
 
 		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 		BUG_ON(ret);
-		goto done;
+		goto release;
 	} else if (split == start) {
 		if (locked_end < extent_end) {
 			ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
@@ -926,6 +926,8 @@ again:
 	}
 done:
 	btrfs_mark_buffer_dirty(leaf);
+
+release:
 	btrfs_release_path(root, path);
 	if (split_end && split == start) {
 		split = end;
@@ -1131,7 +1133,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		if (will_write) {
 			btrfs_fdatawrite_range(inode->i_mapping, pos,
 					       pos + write_bytes - 1,
-					       WB_SYNC_NONE);
+					       WB_SYNC_ALL);
 		} else {
 			balance_dirty_pages_ratelimited_nr(inode->i_mapping,
 							   num_pages);

fs/btrfs/inode.c

@@ -4970,10 +4970,10 @@ out_fail:
 	return err;
 }
 
-static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
+static int prealloc_file_range(struct btrfs_trans_handle *trans,
+			       struct inode *inode, u64 start, u64 end,
 			       u64 alloc_hint, int mode)
 {
-	struct btrfs_trans_handle *trans;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_key ins;
 	u64 alloc_size;
@@ -4981,10 +4981,6 @@ static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
 	u64 num_bytes = end - start;
 	int ret = 0;
 
-	trans = btrfs_join_transaction(root, 1);
-	BUG_ON(!trans);
-	btrfs_set_trans_block_group(trans, inode);
-
 	while (num_bytes > 0) {
 		alloc_size = min(num_bytes, root->fs_info->max_extent);
 		ret = btrfs_reserve_extent(trans, root, alloc_size,
@@ -5015,7 +5011,6 @@ out:
 		BUG_ON(ret);
 	}
 
-	btrfs_end_transaction(trans, root);
 	return ret;
 }
 
@@ -5029,11 +5024,18 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 	u64 alloc_hint = 0;
 	u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
 	struct extent_map *em;
+	struct btrfs_trans_handle *trans;
 	int ret;
 
 	alloc_start = offset & ~mask;
 	alloc_end =  (offset + len + mask) & ~mask;
 
+	/*
+	 * wait for ordered IO before we have any locks.  We'll loop again
+	 * below with the locks held.
+	 */
+	btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
+
 	mutex_lock(&inode->i_mutex);
 	if (alloc_start > inode->i_size) {
 		ret = btrfs_cont_expand(inode, alloc_start);
@@ -5043,6 +5045,16 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 
 	while (1) {
 		struct btrfs_ordered_extent *ordered;
+
+		trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
+		if (!trans) {
+			ret = -EIO;
+			goto out;
+		}
+
+		/* the extent lock is ordered inside the running
+		 * transaction
+		 */
 		lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
 			    alloc_end - 1, GFP_NOFS);
 		ordered = btrfs_lookup_first_ordered_extent(inode,
@@ -5053,6 +5065,12 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 			btrfs_put_ordered_extent(ordered);
 			unlock_extent(&BTRFS_I(inode)->io_tree,
 				      alloc_start, alloc_end - 1, GFP_NOFS);
+			btrfs_end_transaction(trans, BTRFS_I(inode)->root);
+
+			/*
+			 * we can't wait on the range with the transaction
+			 * running or with the extent lock held
+			 */
 			btrfs_wait_ordered_range(inode, alloc_start,
 						 alloc_end - alloc_start);
 		} else {
@@ -5070,7 +5088,7 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 		last_byte = min(extent_map_end(em), alloc_end);
 		last_byte = (last_byte + mask) & ~mask;
 		if (em->block_start == EXTENT_MAP_HOLE) {
-			ret = prealloc_file_range(inode, cur_offset,
+			ret = prealloc_file_range(trans, inode, cur_offset,
 					last_byte, alloc_hint, mode);
 			if (ret < 0) {
 				free_extent_map(em);
@@ -5089,6 +5107,8 @@ static long btrfs_fallocate(struct inode *inode, int mode,
 	}
 	unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
 		      GFP_NOFS);
+
+	btrfs_end_transaction(trans, BTRFS_I(inode)->root);
 out:
 	mutex_unlock(&inode->i_mutex);
 	return ret;

fs/btrfs/ordered-data.c

@@ -489,7 +489,7 @@ again:
 	/* start IO across the range first to instantiate any delalloc
 	 * extents
 	 */
-	btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
+	btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
 
 	/* The compression code will leave pages locked but return from
 	 * writepage without setting the page writeback.  Starting again

fs/btrfs/volumes.c

@@ -125,6 +125,20 @@ static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 	return NULL;
 }
 
+static void requeue_list(struct btrfs_pending_bios *pending_bios,
+			struct bio *head, struct bio *tail)
+{
+
+	struct bio *old_head;
+
+	old_head = pending_bios->head;
+	pending_bios->head = head;
+	if (pending_bios->tail)
+		tail->bi_next = old_head;
+	else
+		pending_bios->tail = tail;
+}
+
 /*
  * we try to collect pending bios for a device so we don't get a large
  * number of procs sending bios down to the same device.  This greatly
@@ -141,10 +155,12 @@ static noinline int run_scheduled_bios(struct btrfs_device *device)
 	struct bio *pending;
 	struct backing_dev_info *bdi;
 	struct btrfs_fs_info *fs_info;
+	struct btrfs_pending_bios *pending_bios;
 	struct bio *tail;
 	struct bio *cur;
 	int again = 0;
-	unsigned long num_run = 0;
+	unsigned long num_run;
+	unsigned long num_sync_run;
 	unsigned long limit;
 	unsigned long last_waited = 0;
 
@@ -153,20 +169,30 @@ static noinline int run_scheduled_bios(struct btrfs_device *device)
 	limit = btrfs_async_submit_limit(fs_info);
 	limit = limit * 2 / 3;
 
+	/* we want to make sure that every time we switch from the sync
+	 * list to the normal list, we unplug
+	 */
+	num_sync_run = 0;
+
 loop:
 	spin_lock(&device->io_lock);
+	num_run = 0;
 
 loop_lock:
+
 	/* take all the bios off the list at once and process them
 	 * later on (without the lock held).  But, remember the
 	 * tail and other pointers so the bios can be properly reinserted
 	 * into the list if we hit congestion
 	 */
-	pending = device->pending_bios;
-	tail = device->pending_bio_tail;
+	if (device->pending_sync_bios.head)
+		pending_bios = &device->pending_sync_bios;
+	else
+		pending_bios = &device->pending_bios;
+
+	pending = pending_bios->head;
+	tail = pending_bios->tail;
 	WARN_ON(pending && !tail);
-	device->pending_bios = NULL;
-	device->pending_bio_tail = NULL;
 
 	/*
 	 * if pending was null this time around, no bios need processing
@@ -176,16 +202,41 @@ loop_lock:
 	 * device->running_pending is used to synchronize with the
 	 * schedule_bio code.
 	 */
-	if (pending) {
-		again = 1;
-		device->running_pending = 1;
-	} else {
+	if (device->pending_sync_bios.head == NULL &&
+	    device->pending_bios.head == NULL) {
 		again = 0;
 		device->running_pending = 0;
+	} else {
+		again = 1;
+		device->running_pending = 1;
 	}
+
+	pending_bios->head = NULL;
+	pending_bios->tail = NULL;
+
 	spin_unlock(&device->io_lock);
 
+	/*
+	 * if we're doing the regular priority list, make sure we unplug
+	 * for any high prio bios we've sent down
+	 */
+	if (pending_bios == &device->pending_bios && num_sync_run > 0) {
+		num_sync_run = 0;
+		blk_run_backing_dev(bdi, NULL);
+	}
+
 	while (pending) {
+
+		rmb();
+		if (pending_bios != &device->pending_sync_bios &&
+		    device->pending_sync_bios.head &&
+		    num_run > 16) {
+			cond_resched();
+			spin_lock(&device->io_lock);
+			requeue_list(pending_bios, pending, tail);
+			goto loop_lock;
+		}
+
 		cur = pending;
 		pending = pending->bi_next;
 		cur->bi_next = NULL;
@@ -196,10 +247,18 @@ loop_lock:
 			wake_up(&fs_info->async_submit_wait);
 
 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
-		bio_get(cur);
 		submit_bio(cur->bi_rw, cur);
-		bio_put(cur);
 		num_run++;
+		if (bio_sync(cur))
+			num_sync_run++;
+
+		if (need_resched()) {
+			if (num_sync_run) {
+				blk_run_backing_dev(bdi, NULL);
+				num_sync_run = 0;
+			}
+			cond_resched();
+		}
 
 		/*
 		 * we made progress, there is more work to do and the bdi
@@ -208,7 +267,6 @@ loop_lock:
 		 */
 		if (pending && bdi_write_congested(bdi) && num_run > 16 &&
 		    fs_info->fs_devices->open_devices > 1) {
-			struct bio *old_head;
 			struct io_context *ioc;
 
 			ioc = current->io_context;
@@ -233,17 +291,17 @@ loop_lock:
 				 * against it before looping
 				 */
 				last_waited = ioc->last_waited;
+				if (need_resched()) {
+					if (num_sync_run) {
+						blk_run_backing_dev(bdi, NULL);
+						num_sync_run = 0;
+					}
+					cond_resched();
+				}
 				continue;
 			}
 			spin_lock(&device->io_lock);
-
-			old_head = device->pending_bios;
-			device->pending_bios = pending;
-			if (device->pending_bio_tail)
-				tail->bi_next = old_head;
-			else
-				device->pending_bio_tail = tail;
-
+			requeue_list(pending_bios, pending, tail);
 			device->running_pending = 1;
 
 			spin_unlock(&device->io_lock);
@@ -251,11 +309,18 @@ loop_lock:
 			goto done;
 		}
 	}
+
+	if (num_sync_run) {
+		num_sync_run = 0;
+		blk_run_backing_dev(bdi, NULL);
+	}
+
+	cond_resched();
 	if (again)
 		goto loop;
 
 	spin_lock(&device->io_lock);
-	if (device->pending_bios)
+	if (device->pending_bios.head || device->pending_sync_bios.head)
 		goto loop_lock;
 	spin_unlock(&device->io_lock);
 
@@ -2497,7 +2562,7 @@ again:
 			max_errors = 1;
 		}
 	}
-	if (multi_ret && rw == WRITE &&
+	if (multi_ret && (rw & (1 << BIO_RW)) &&
 	    stripes_allocated < stripes_required) {
 		stripes_allocated = map->num_stripes;
 		free_extent_map(em);
@@ -2762,6 +2827,7 @@ static noinline int schedule_bio(struct btrfs_root *root,
 				 int rw, struct bio *bio)
 {
 	int should_queue = 1;
+	struct btrfs_pending_bios *pending_bios;
 
 	/* don't bother with additional async steps for reads, right now */
 	if (!(rw & (1 << BIO_RW))) {
@@ -2783,13 +2849,17 @@ static noinline int schedule_bio(struct btrfs_root *root,
 	bio->bi_rw |= rw;
 
 	spin_lock(&device->io_lock);
+	if (bio_sync(bio))
+		pending_bios = &device->pending_sync_bios;
+	else
+		pending_bios = &device->pending_bios;
 
-	if (device->pending_bio_tail)
-		device->pending_bio_tail->bi_next = bio;
+	if (pending_bios->tail)
+		pending_bios->tail->bi_next = bio;
 
-	device->pending_bio_tail = bio;
-	if (!device->pending_bios)
-		device->pending_bios = bio;
+	pending_bios->tail = bio;
+	if (!pending_bios->head)
+		pending_bios->head = bio;
 	if (device->running_pending)
 		should_queue = 0;
 

fs/btrfs/volumes.h

@@ -23,13 +23,22 @@
 #include "async-thread.h"
 
 struct buffer_head;
+struct btrfs_pending_bios {
+	struct bio *head;
+	struct bio *tail;
+};
+
 struct btrfs_device {
 	struct list_head dev_list;
 	struct list_head dev_alloc_list;
 	struct btrfs_fs_devices *fs_devices;
 	struct btrfs_root *dev_root;
-	struct bio *pending_bios;
-	struct bio *pending_bio_tail;
+
+	/* regular prio bios */
+	struct btrfs_pending_bios pending_bios;
+	/* WRITE_SYNC bios */
+	struct btrfs_pending_bios pending_sync_bios;
+
 	int running_pending;
 	u64 generation;