1 files changed, 160 insertions, 103 deletions

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 1c3a1189c0bd..1f1c25db6f6b 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -162,7 +162,17 @@ static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
 	struct btrfs_root *root, *tmp;
 	struct btrfs_caching_control *caching_ctl, *next;
 
+	/*
+	 * At this point no one can be using this transaction to modify any tree
+	 * and no one can start another transaction to modify any tree either.
+	 */
+	ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING);
+
 	down_write(&fs_info->commit_root_sem);
+
+	if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
+		fs_info->last_reloc_trans = trans->transid;
+
 	list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
 				 dirty_list) {
 		list_del_init(&root->dirty_list);
@@ -413,7 +423,6 @@ static int record_root_in_trans(struct btrfs_trans_handle *trans,
 
 	if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 	    root->last_trans < trans->transid) || force) {
-		WARN_ON(root == fs_info->extent_root);
 		WARN_ON(!force && root->commit_root != root->node);
 
 		/*
@@ -628,7 +637,7 @@ start_transaction(struct btrfs_root *root, unsigned int num_items,
 			reloc_reserved = true;
 		}
 
-		ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
+		ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush);
 		if (ret)
 			goto reserve_fail;
 		if (delayed_refs_bytes) {
@@ -692,7 +701,6 @@ again:
 
 	h->transid = cur_trans->transid;
 	h->transaction = cur_trans;
-	h->root = root;
 	refcount_set(&h->use_count, 1);
 	h->fs_info = root->fs_info;
 
@@ -846,7 +854,37 @@ btrfs_attach_transaction_barrier(struct btrfs_root *root)
 static noinline void wait_for_commit(struct btrfs_transaction *commit,
 				     const enum btrfs_trans_state min_state)
 {
-	wait_event(commit->commit_wait, commit->state >= min_state);
+	struct btrfs_fs_info *fs_info = commit->fs_info;
+	u64 transid = commit->transid;
+	bool put = false;
+
+	while (1) {
+		wait_event(commit->commit_wait, commit->state >= min_state);
+		if (put)
+			btrfs_put_transaction(commit);
+
+		if (min_state < TRANS_STATE_COMPLETED)
+			break;
+
+		/*
+		 * A transaction isn't really completed until all of the
+		 * previous transactions are completed, but with fsync we can
+		 * end up with SUPER_COMMITTED transactions before a COMPLETED
+		 * transaction. Wait for those.
+		 */
+
+		spin_lock(&fs_info->trans_lock);
+		commit = list_first_entry_or_null(&fs_info->trans_list,
+						  struct btrfs_transaction,
+						  list);
+		if (!commit || commit->transid > transid) {
+			spin_unlock(&fs_info->trans_lock);
+			break;
+		}
+		refcount_inc(&commit->use_count);
+		put = true;
+		spin_unlock(&fs_info->trans_lock);
+	}
 }
 
 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
@@ -1236,6 +1274,12 @@ static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
 	struct extent_buffer *eb;
 	int ret;
 
+	/*
+	 * At this point no one can be using this transaction to modify any tree
+	 * and no one can start another transaction to modify any tree either.
+	 */
+	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
+
 	eb = btrfs_lock_root_node(fs_info->tree_root);
 	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
 			      0, &eb, BTRFS_NESTING_COW);
@@ -1267,9 +1311,8 @@ again:
 		root = list_entry(next, struct btrfs_root, dirty_list);
 		clear_bit(BTRFS_ROOT_DIRTY, &root->state);
 
-		if (root != fs_info->extent_root)
-			list_add_tail(&root->dirty_list,
-				      &trans->transaction->switch_commits);
+		list_add_tail(&root->dirty_list,
+			      &trans->transaction->switch_commits);
 		ret = update_cowonly_root(trans, root);
 		if (ret)
 			return ret;
@@ -1299,9 +1342,6 @@ again:
 	if (!list_empty(&fs_info->dirty_cowonly_roots))
 		goto again;
 
-	list_add_tail(&fs_info->extent_root->dirty_list,
-		      &trans->transaction->switch_commits);
-
 	/* Update dev-replace pointer once everything is committed */
 	fs_info->dev_replace.committed_cursor_left =
 		fs_info->dev_replace.cursor_left_last_write_of_item;
@@ -1310,6 +1350,32 @@ again:
 }
 
 /*
+ * If we had a pending drop we need to see if there are any others left in our
+ * dead roots list, and if not clear our bit and wake any waiters.
+ */
+void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
+{
+	/*
+	 * We put the drop in progress roots at the front of the list, so if the
+	 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
+	 * up.
+	 */
+	spin_lock(&fs_info->trans_lock);
+	if (!list_empty(&fs_info->dead_roots)) {
+		struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
+							   struct btrfs_root,
+							   root_list);
+		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
+			spin_unlock(&fs_info->trans_lock);
+			return;
+		}
+	}
+	spin_unlock(&fs_info->trans_lock);
+
+	btrfs_wake_unfinished_drop(fs_info);
+}
+
+/*
  * dead roots are old snapshots that need to be deleted.  This allocates
  * a dirty root struct and adds it into the list of dead roots that need to
  * be deleted
@@ -1321,13 +1387,19 @@ void btrfs_add_dead_root(struct btrfs_root *root)
 	spin_lock(&fs_info->trans_lock);
 	if (list_empty(&root->root_list)) {
 		btrfs_grab_root(root);
-		list_add_tail(&root->root_list, &fs_info->dead_roots);
+
+		/* We want to process the partially complete drops first. */
+		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
+			list_add(&root->root_list, &fs_info->dead_roots);
+		else
+			list_add_tail(&root->root_list, &fs_info->dead_roots);
 	}
 	spin_unlock(&fs_info->trans_lock);
 }
 
 /*
- * update all the cowonly tree roots on disk
+ * Update each subvolume root and its relocation root, if it exists, in the tree
+ * of tree roots. Also free log roots if they exist.
  */
 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
 {
@@ -1336,6 +1408,12 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
 	int i;
 	int ret;
 
+	/*
+	 * At this point no one can be using this transaction to modify any tree
+	 * and no one can start another transaction to modify any tree either.
+	 */
+	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
+
 	spin_lock(&fs_info->fs_roots_radix_lock);
 	while (1) {
 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
@@ -1348,6 +1426,14 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
 			struct btrfs_root *root = gang[i];
 			int ret2;
 
+			/*
+			 * At this point we can neither have tasks logging inodes
+			 * from a root nor trying to commit a log tree.
+			 */
+			ASSERT(atomic_read(&root->log_writers) == 0);
+			ASSERT(atomic_read(&root->log_commit[0]) == 0);
+			ASSERT(atomic_read(&root->log_commit[1]) == 0);
+
 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
 					(unsigned long)root->root_key.objectid,
 					BTRFS_ROOT_TRANS_TAG);
@@ -1472,12 +1558,6 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 		return ret;
 	}
 
-	/*
-	 * We are going to commit transaction, see btrfs_commit_transaction()
-	 * comment for reason locking tree_log_mutex
-	 */
-	mutex_lock(&fs_info->tree_log_mutex);
-
 	ret = commit_fs_roots(trans);
 	if (ret)
 		goto out;
@@ -1513,8 +1593,6 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 			"Error while writing out transaction for qgroup");
 
 out:
-	mutex_unlock(&fs_info->tree_log_mutex);
-
 	/*
 	 * Force parent root to be updated, as we recorded it before so its
 	 * last_trans == cur_transid.
@@ -1578,7 +1656,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	btrfs_reloc_pre_snapshot(pending, &to_reserve);
 
 	if (to_reserve > 0) {
-		pending->error = btrfs_block_rsv_add(root,
+		pending->error = btrfs_block_rsv_add(fs_info,
 						     &pending->block_rsv,
 						     to_reserve,
 						     BTRFS_RESERVE_NO_FLUSH);
@@ -1861,50 +1939,14 @@ int btrfs_transaction_blocked(struct btrfs_fs_info *info)
 	return ret;
 }
 
-/*
- * commit transactions asynchronously. once btrfs_commit_transaction_async
- * returns, any subsequent transaction will not be allowed to join.
- */
-struct btrfs_async_commit {
-	struct btrfs_trans_handle *newtrans;
-	struct work_struct work;
-};
-
-static void do_async_commit(struct work_struct *work)
-{
-	struct btrfs_async_commit *ac =
-		container_of(work, struct btrfs_async_commit, work);
-
-	/*
-	 * We've got freeze protection passed with the transaction.
-	 * Tell lockdep about it.
-	 */
-	if (ac->newtrans->type & __TRANS_FREEZABLE)
-		__sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
-
-	current->journal_info = ac->newtrans;
-
-	btrfs_commit_transaction(ac->newtrans);
-	kfree(ac);
-}
-
-int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
+void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
-	struct btrfs_async_commit *ac;
 	struct btrfs_transaction *cur_trans;
 
-	ac = kmalloc(sizeof(*ac), GFP_NOFS);
-	if (!ac)
-		return -ENOMEM;
-
-	INIT_WORK(&ac->work, do_async_commit);
-	ac->newtrans = btrfs_join_transaction(trans->root);
-	if (IS_ERR(ac->newtrans)) {
-		int err = PTR_ERR(ac->newtrans);
-		kfree(ac);
-		return err;
-	}
+	/* Kick the transaction kthread. */
+	set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
+	wake_up_process(fs_info->transaction_kthread);
 
 	/* take transaction reference */
 	cur_trans = trans->transaction;
@@ -1913,28 +1955,15 @@ int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
 	btrfs_end_transaction(trans);
 
 	/*
-	 * Tell lockdep we've released the freeze rwsem, since the
-	 * async commit thread will be the one to unlock it.
-	 */
-	if (ac->newtrans->type & __TRANS_FREEZABLE)
-		__sb_writers_release(fs_info->sb, SB_FREEZE_FS);
-
-	schedule_work(&ac->work);
-	/*
 	 * Wait for the current transaction commit to start and block
 	 * subsequent transaction joins
 	 */
 	wait_event(fs_info->transaction_blocked_wait,
 		   cur_trans->state >= TRANS_STATE_COMMIT_START ||
 		   TRANS_ABORTED(cur_trans));
-	if (current->journal_info == trans)
-		current->journal_info = NULL;
-
 	btrfs_put_transaction(cur_trans);
-	return 0;
 }
 
-
 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
@@ -1986,7 +2015,7 @@ static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
 	btrfs_put_transaction(cur_trans);
 	btrfs_put_transaction(cur_trans);
 
-	trace_btrfs_transaction_commit(trans->root);
+	trace_btrfs_transaction_commit(fs_info);
 
 	if (current->journal_info == trans)
 		current->journal_info = NULL;
@@ -2013,16 +2042,24 @@ static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
 {
 	/*
-	 * We use writeback_inodes_sb here because if we used
+	 * We use try_to_writeback_inodes_sb() here because if we used
 	 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
 	 * Currently are holding the fs freeze lock, if we do an async flush
 	 * we'll do btrfs_join_transaction() and deadlock because we need to
 	 * wait for the fs freeze lock.  Using the direct flushing we benefit
 	 * from already being in a transaction and our join_transaction doesn't
 	 * have to re-take the fs freeze lock.
+	 *
+	 * Note that try_to_writeback_inodes_sb() will only trigger writeback
+	 * if it can read lock sb->s_umount. It will always be able to lock it,
+	 * except when the filesystem is being unmounted or being frozen, but in
+	 * those cases sync_filesystem() is called, which results in calling
+	 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
+	 * Note that we don't call writeback_inodes_sb() directly, because it
+	 * will emit a warning if sb->s_umount is not locked.
 	 */
 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
-		writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
+		try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
 	return 0;
 }
 
@@ -2032,6 +2069,27 @@ static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
 		btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 }
 
+/*
+ * Add a pending snapshot associated with the given transaction handle to the
+ * respective handle. This must be called after the transaction commit started
+ * and while holding fs_info->trans_lock.
+ * This serves to guarantee a caller of btrfs_commit_transaction() that it can
+ * safely free the pending snapshot pointer in case btrfs_commit_transaction()
+ * returns an error.
+ */
+static void add_pending_snapshot(struct btrfs_trans_handle *trans)
+{
+	struct btrfs_transaction *cur_trans = trans->transaction;
+
+	if (!trans->pending_snapshot)
+		return;
+
+	lockdep_assert_held(&trans->fs_info->trans_lock);
+	ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START);
+
+	list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
+}
+
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
@@ -2105,6 +2163,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
 		enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
 
+		add_pending_snapshot(trans);
+
 		spin_unlock(&fs_info->trans_lock);
 		refcount_inc(&cur_trans->use_count);
 
@@ -2195,11 +2255,19 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
 	 */
 	spin_lock(&fs_info->trans_lock);
+	add_pending_snapshot(trans);
 	cur_trans->state = TRANS_STATE_COMMIT_DOING;
 	spin_unlock(&fs_info->trans_lock);
 	wait_event(cur_trans->writer_wait,
 		   atomic_read(&cur_trans->num_writers) == 1);
 
+	/*
+	 * We've started the commit, clear the flag in case we were triggered to
+	 * do an async commit but somebody else started before the transaction
+	 * kthread could do the work.
+	 */
+	clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
+
 	if (TRANS_ABORTED(cur_trans)) {
 		ret = cur_trans->aborted;
 		goto scrub_continue;
@@ -2246,24 +2314,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 
 	WARN_ON(cur_trans != trans->transaction);
 
-	/* btrfs_commit_tree_roots is responsible for getting the
-	 * various roots consistent with each other.  Every pointer
-	 * in the tree of tree roots has to point to the most up to date
-	 * root for every subvolume and other tree.  So, we have to keep
-	 * the tree logging code from jumping in and changing any
-	 * of the trees.
-	 *
-	 * At this point in the commit, there can't be any tree-log
-	 * writers, but a little lower down we drop the trans mutex
-	 * and let new people in.  By holding the tree_log_mutex
-	 * from now until after the super is written, we avoid races
-	 * with the tree-log code.
-	 */
-	mutex_lock(&fs_info->tree_log_mutex);
-
 	ret = commit_fs_roots(trans);
 	if (ret)
-		goto unlock_tree_log;
+		goto unlock_reloc;
 
 	/*
 	 * Since the transaction is done, we can apply the pending changes
@@ -2282,11 +2335,11 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	 */
 	ret = btrfs_qgroup_account_extents(trans);
 	if (ret < 0)
-		goto unlock_tree_log;
+		goto unlock_reloc;
 
 	ret = commit_cowonly_roots(trans);
 	if (ret)
-		goto unlock_tree_log;
+		goto unlock_reloc;
 
 	/*
 	 * The tasks which save the space cache and inode cache may also
@@ -2294,7 +2347,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	 */
 	if (TRANS_ABORTED(cur_trans)) {
 		ret = cur_trans->aborted;
-		goto unlock_tree_log;
+		goto unlock_reloc;
 	}
 
 	cur_trans = fs_info->running_transaction;
@@ -2327,6 +2380,16 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 
 	btrfs_trans_release_chunk_metadata(trans);
 
+	/*
+	 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
+	 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
+	 * make sure that before we commit our superblock, no other task can
+	 * start a new transaction and commit a log tree before we commit our
+	 * superblock. Anyone trying to commit a log tree locks this mutex before
+	 * writing its superblock.
+	 */
+	mutex_lock(&fs_info->tree_log_mutex);
+
 	spin_lock(&fs_info->trans_lock);
 	cur_trans->state = TRANS_STATE_UNBLOCKED;
 	fs_info->running_transaction = NULL;
@@ -2339,10 +2402,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	if (ret) {
 		btrfs_handle_fs_error(fs_info, ret,
 				      "Error while writing out transaction");
-		/*
-		 * reloc_mutex has been unlocked, tree_log_mutex is still held
-		 * but we can't jump to unlock_tree_log causing double unlock
-		 */
 		mutex_unlock(&fs_info->tree_log_mutex);
 		goto scrub_continue;
 	}
@@ -2393,7 +2452,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	if (trans->type & __TRANS_FREEZABLE)
 		sb_end_intwrite(fs_info->sb);
 
-	trace_btrfs_transaction_commit(trans->root);
+	trace_btrfs_transaction_commit(fs_info);
 
 	btrfs_scrub_continue(fs_info);
 
@@ -2404,8 +2463,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 
 	return ret;
 
-unlock_tree_log:
-	mutex_unlock(&fs_info->tree_log_mutex);
 unlock_reloc:
 	mutex_unlock(&fs_info->reloc_mutex);
 scrub_continue: