Btrfs: Replace the transaction work queue with kthreads

This creates one kthread for commits and one kthread for
deleting old snapshots.  All the work queues are removed.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/ctree.c

@@ -1352,6 +1352,8 @@ again:
 						free_extent_buffer(tmp);
 					goto again;
 				} else {
+					if (tmp)
+						free_extent_buffer(tmp);
 					b = read_node_slot(root, b, slot);
 				}
 			}
@@ -3048,7 +3050,8 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
 		free_extent_buffer(c);
 		path->nodes[level] = next;
 		path->slots[level] = 0;
-		path->locks[level] = 1;
+		if (!path->skip_locking)
+			path->locks[level] = 1;
 		if (!level)
 			break;
 		if (level == 1 && path->locks[1] && path->reada)

fs/btrfs/ctree.h

@@ -23,7 +23,6 @@
 #include <linux/mm.h>
 #include <linux/highmem.h>
 #include <linux/fs.h>
-#include <linux/workqueue.h>
 #include <linux/completion.h>
 #include <linux/backing-dev.h>
 #include <asm/kmap_types.h>
@@ -519,15 +518,14 @@ struct btrfs_fs_info {
 	struct backing_dev_info bdi;
 	spinlock_t hash_lock;
 	struct mutex trans_mutex;
+	struct mutex transaction_kthread_mutex;
+	struct mutex cleaner_mutex;
 	struct mutex alloc_mutex;
 	struct mutex chunk_mutex;
 	struct mutex drop_mutex;
 	struct list_head trans_list;
 	struct list_head hashers;
 	struct list_head dead_roots;
-	struct list_head end_io_work_list;
-	struct work_struct end_io_work;
-	spinlock_t end_io_work_lock;
 	atomic_t nr_async_submits;
 
 	/*
@@ -543,13 +541,10 @@ struct btrfs_fs_info {
 	struct btrfs_workers workers;
 	struct btrfs_workers endio_workers;
 	struct btrfs_workers submit_workers;
+	struct task_struct *transaction_kthread;
+	struct task_struct *cleaner_kthread;
 	int thread_pool_size;
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-	struct work_struct trans_work;
-#else
-	struct delayed_work trans_work;
-#endif
 	struct kobject super_kobj;
 	struct completion kobj_unregister;
 	int do_barriers;

fs/btrfs/disk-io.c

@@ -16,6 +16,7 @@
  * Boston, MA 021110-1307, USA.
  */
 
+#include <linux/version.h>
 #include <linux/fs.h>
 #include <linux/blkdev.h>
 #include <linux/scatterlist.h>
@@ -24,6 +25,12 @@
 #include <linux/writeback.h>
 #include <linux/buffer_head.h> // for block_sync_page
 #include <linux/workqueue.h>
+#include <linux/kthread.h>
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
+# include <linux/freezer.h>
+#else
+# include <linux/sched.h>
+#endif
 #include "crc32c.h"
 #include "ctree.h"
 #include "disk-io.h"
@@ -1100,6 +1107,87 @@ static void end_workqueue_fn(struct btrfs_work *work)
 #endif
 }
 
+static int cleaner_kthread(void *arg)
+{
+	struct btrfs_root *root = arg;
+
+	do {
+		smp_mb();
+		if (root->fs_info->closing)
+			break;
+
+		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
+		mutex_lock(&root->fs_info->cleaner_mutex);
+printk("cleaner awake\n");
+		btrfs_clean_old_snapshots(root);
+printk("cleaner done\n");
+		mutex_unlock(&root->fs_info->cleaner_mutex);
+
+		if (freezing(current)) {
+			refrigerator();
+		} else {
+			smp_mb();
+			if (root->fs_info->closing)
+				break;
+			set_current_state(TASK_INTERRUPTIBLE);
+			schedule();
+			__set_current_state(TASK_RUNNING);
+		}
+	} while (!kthread_should_stop());
+	return 0;
+}
+
+static int transaction_kthread(void *arg)
+{
+	struct btrfs_root *root = arg;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_transaction *cur;
+	unsigned long now;
+	unsigned long delay;
+	int ret;
+
+	do {
+		smp_mb();
+		if (root->fs_info->closing)
+			break;
+
+		delay = HZ * 30;
+		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
+		mutex_lock(&root->fs_info->transaction_kthread_mutex);
+
+		mutex_lock(&root->fs_info->trans_mutex);
+		cur = root->fs_info->running_transaction;
+		if (!cur) {
+			mutex_unlock(&root->fs_info->trans_mutex);
+			goto sleep;
+		}
+		now = get_seconds();
+		if (now < cur->start_time || now - cur->start_time < 30) {
+			mutex_unlock(&root->fs_info->trans_mutex);
+			delay = HZ * 5;
+			goto sleep;
+		}
+		mutex_unlock(&root->fs_info->trans_mutex);
+		btrfs_defrag_dirty_roots(root->fs_info);
+		trans = btrfs_start_transaction(root, 1);
+		ret = btrfs_commit_transaction(trans, root);
+sleep:
+		wake_up_process(root->fs_info->cleaner_kthread);
+		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
+
+		if (freezing(current)) {
+			refrigerator();
+		} else {
+			if (root->fs_info->closing)
+				break;
+			set_current_state(TASK_INTERRUPTIBLE);
+			schedule_timeout(delay);
+			__set_current_state(TASK_RUNNING);
+		}
+	} while (!kthread_should_stop());
+	return 0;
+}
+
 struct btrfs_root *open_ctree(struct super_block *sb,
 			      struct btrfs_fs_devices *fs_devices,
 			      char *options)
@@ -1189,11 +1277,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
 	fs_info->do_barriers = 1;
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-	INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
-#else
-	INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
-#endif
 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 	       sizeof(struct btrfs_key));
@@ -1204,6 +1287,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	mutex_init(&fs_info->drop_mutex);
 	mutex_init(&fs_info->alloc_mutex);
 	mutex_init(&fs_info->chunk_mutex);
+	mutex_init(&fs_info->transaction_kthread_mutex);
+	mutex_init(&fs_info->cleaner_mutex);
 
 #if 0
 	ret = add_hasher(fs_info, "crc32c");
@@ -1247,7 +1332,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	btrfs_start_workers(&fs_info->submit_workers, 1);
 	btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
 
-
 	err = -EINVAL;
 	if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
 		printk("Btrfs: wanted %llu devices, but found %llu\n",
@@ -1341,9 +1425,22 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	fs_info->data_alloc_profile = (u64)-1;
 	fs_info->metadata_alloc_profile = (u64)-1;
 	fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
+	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
+					       "btrfs-cleaner");
+	if (!fs_info->cleaner_kthread)
+		goto fail_extent_root;
+
+	fs_info->transaction_kthread = kthread_run(transaction_kthread,
+						   tree_root,
+						   "btrfs-transaction");
+	if (!fs_info->transaction_kthread)
+		goto fail_trans_kthread;
+
 
 	return tree_root;
 
+fail_trans_kthread:
+	kthread_stop(fs_info->cleaner_kthread);
 fail_extent_root:
 	free_extent_buffer(extent_root->node);
 fail_tree_root:
@@ -1562,8 +1659,11 @@ int close_ctree(struct btrfs_root *root)
 	fs_info->closing = 1;
 	smp_mb();
 
-	btrfs_transaction_flush_work(root);
+	kthread_stop(root->fs_info->transaction_kthread);
+	kthread_stop(root->fs_info->cleaner_kthread);
+
 	btrfs_defrag_dirty_roots(root->fs_info);
+	btrfs_clean_old_snapshots(root);
 	trans = btrfs_start_transaction(root, 1);
 	ret = btrfs_commit_transaction(trans, root);
 	/* run commit again to  drop the original snapshot */
@@ -1574,8 +1674,6 @@ int close_ctree(struct btrfs_root *root)
 
 	write_ctree_super(NULL, root);
 
-	btrfs_transaction_flush_work(root);
-
 	if (fs_info->delalloc_bytes) {
 		printk("btrfs: at unmount delalloc count %Lu\n",
 		       fs_info->delalloc_bytes);

fs/btrfs/extent-tree.c

@@ -1216,15 +1216,16 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 	if (ret == -ENOSPC) {
 printk("space info full %Lu\n", flags);
 		space_info->full = 1;
-		goto out;
+		goto out_unlock;
 	}
 	BUG_ON(ret);
 
 	ret = btrfs_make_block_group(trans, extent_root, 0, flags,
 		     BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
 	BUG_ON(ret);
-out:
+out_unlock:
 	mutex_unlock(&extent_root->fs_info->chunk_mutex);
+out:
 	return 0;
 }
 
@@ -2274,7 +2275,8 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
 			free_extent_buffer(next);
 			mutex_unlock(&root->fs_info->alloc_mutex);
 
-			reada_walk_down(root, cur, path->slots[*level]);
+			if (path->slots[*level] == 0)
+				reada_walk_down(root, cur, path->slots[*level]);
 
 			next = read_tree_block(root, bytenr, blocksize,
 					       ptr_gen);
@@ -2446,8 +2448,6 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 			break;
 		if (wret < 0)
 			ret = wret;
-		ret = -EAGAIN;
-		break;
 	}
 	for (i = 0; i <= orig_level; i++) {
 		if (path->nodes[i]) {

fs/btrfs/super.c

@@ -340,7 +340,6 @@ static int btrfs_fill_super(struct super_block * sb,
 		goto fail_close;
 
 	sb->s_root = root_dentry;
-	btrfs_transaction_queue_work(tree_root, HZ * 30);
 
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,25)
 	save_mount_options(sb, data);
@@ -416,9 +415,7 @@ static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
 		goto error_free_subvol_name;
 
 	bdev = fs_devices->latest_bdev;
-	btrfs_lock_volumes();
 	s = sget(fs_type, btrfs_test_super, set_anon_super, fs_devices);
-	btrfs_unlock_volumes();
 	if (IS_ERR(s))
 		goto error_s;
 
@@ -530,13 +527,15 @@ out:
 static void btrfs_write_super_lockfs(struct super_block *sb)
 {
 	struct btrfs_root *root = btrfs_sb(sb);
-	btrfs_transaction_flush_work(root);
+	mutex_lock(&root->fs_info->transaction_kthread_mutex);
+	mutex_lock(&root->fs_info->cleaner_mutex);
 }
 
 static void btrfs_unlockfs(struct super_block *sb)
 {
 	struct btrfs_root *root = btrfs_sb(sb);
-	btrfs_transaction_queue_work(root, HZ * 30);
+	mutex_unlock(&root->fs_info->cleaner_mutex);
+	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
 }
 
 static struct super_operations btrfs_super_ops = {
@@ -589,10 +588,9 @@ static int __init init_btrfs_fs(void)
 	if (err)
 		return err;
 
-	btrfs_init_transaction_sys();
 	err = btrfs_init_cachep();
 	if (err)
-		goto free_transaction_sys;
+		goto free_sysfs;
 
 	err = extent_io_init();
 	if (err)
@@ -618,15 +616,13 @@ free_extent_io:
 	extent_io_exit();
 free_cachep:
 	btrfs_destroy_cachep();
-free_transaction_sys:
-	btrfs_exit_transaction_sys();
+free_sysfs:
 	btrfs_exit_sysfs();
 	return err;
 }
 
 static void __exit exit_btrfs_fs(void)
 {
-	btrfs_exit_transaction_sys();
 	btrfs_destroy_cachep();
 	extent_map_exit();
 	extent_io_exit();

fs/btrfs/transaction.c

@@ -29,8 +29,6 @@ static int total_trans = 0;
 extern struct kmem_cache *btrfs_trans_handle_cachep;
 extern struct kmem_cache *btrfs_transaction_cachep;
 
-static struct workqueue_struct *trans_wq;
-
 #define BTRFS_ROOT_TRANS_TAG 0
 #define BTRFS_ROOT_DEFRAG_TAG 1
 
@@ -807,81 +805,15 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
 {
 	struct list_head dirty_roots;
 	INIT_LIST_HEAD(&dirty_roots);
-
+again:
 	mutex_lock(&root->fs_info->trans_mutex);
 	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 	mutex_unlock(&root->fs_info->trans_mutex);
 
 	if (!list_empty(&dirty_roots)) {
 		drop_dirty_roots(root, &dirty_roots);
+		goto again;
 	}
 	return 0;
 }
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-void btrfs_transaction_cleaner(void *p)
-#else
-void btrfs_transaction_cleaner(struct work_struct *work)
-#endif
-{
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-	struct btrfs_fs_info *fs_info = p;
-#else
-	struct btrfs_fs_info *fs_info = container_of(work,
-						     struct btrfs_fs_info,
-						     trans_work.work);
-
-#endif
-	struct btrfs_root *root = fs_info->tree_root;
-	struct btrfs_transaction *cur;
-	struct btrfs_trans_handle *trans;
-	unsigned long now;
-	unsigned long delay = HZ * 30;
-	int ret;
-
-	smp_mb();
-	if (root->fs_info->closing)
-		goto out;
-
-	mutex_lock(&root->fs_info->trans_mutex);
-	cur = root->fs_info->running_transaction;
-	if (!cur) {
-		mutex_unlock(&root->fs_info->trans_mutex);
-		goto out;
-	}
-	now = get_seconds();
-	if (now < cur->start_time || now - cur->start_time < 30) {
-		mutex_unlock(&root->fs_info->trans_mutex);
-		delay = HZ * 5;
-		goto out;
-	}
-	mutex_unlock(&root->fs_info->trans_mutex);
-	btrfs_defrag_dirty_roots(root->fs_info);
-	trans = btrfs_start_transaction(root, 1);
-	ret = btrfs_commit_transaction(trans, root);
-out:
-	btrfs_clean_old_snapshots(root);
-	btrfs_transaction_queue_work(root, delay);
-}
-
-void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
-{
-	if (!root->fs_info->closing)
-		queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
-}
-
-void btrfs_transaction_flush_work(struct btrfs_root *root)
-{
-	cancel_delayed_work(&root->fs_info->trans_work);
-	flush_workqueue(trans_wq);
-}
-
-void __init btrfs_init_transaction_sys(void)
-{
-	trans_wq = create_workqueue("btrfs-transaction");
-}
-
-void btrfs_exit_transaction_sys(void)
-{
-	destroy_workqueue(trans_wq);
-}
 

fs/btrfs/transaction.h

@@ -82,16 +82,6 @@ int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 			    struct btrfs_root *root);
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-void btrfs_transaction_cleaner(void *p);
-#else
-void btrfs_transaction_cleaner(struct work_struct *work);
-#endif
-
-void btrfs_transaction_flush_work(struct btrfs_root *root);
-void btrfs_transaction_queue_work(struct btrfs_root *root, int delay);
-void btrfs_init_transaction_sys(void);
-void btrfs_exit_transaction_sys(void);
 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest,
 			struct list_head *dead_list);
 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info);

fs/btrfs/volumes.c

@@ -271,13 +271,17 @@ again:
 	list_for_each(cur, head) {
 		device = list_entry(cur, struct btrfs_device, dev_list);
 		if (!device->in_fs_metadata) {
-			if (device->bdev) {
-				close_bdev_excl(device->bdev);
-				fs_devices->open_devices--;
-			}
+			struct block_device *bdev;
 			list_del(&device->dev_list);
 			list_del(&device->dev_alloc_list);
 			fs_devices->num_devices--;
+			if (device->bdev) {
+				bdev = device->bdev;
+				fs_devices->open_devices--;
+				mutex_unlock(&uuid_mutex);
+				close_bdev_excl(bdev);
+				mutex_lock(&uuid_mutex);
+			}
 			kfree(device->name);
 			kfree(device);
 			goto again;