writeback: synchronize sync(2) against cgroup writeback membership switches

sync_inodes_sb() can race against cgwb (cgroup writeback) membership
switches and fail to writeback some inodes.  For example, if an inode
switches to another wb while sync_inodes_sb() is in progress, the new
wb might not be visible to bdi_split_work_to_wbs() at all or the inode
might jump from a wb which hasn't issued writebacks yet to one which
already has.

This patch adds backing_dev_info->wb_switch_rwsem to synchronize cgwb
switch path against sync_inodes_sb() so that sync_inodes_sb() is
guaranteed to see all the target wbs and inodes can't jump wbs to
escape syncing.

v2: Fixed misplaced rwsem init.  Spotted by Jiufei.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Jiufei Xue <xuejiufei@gmail.com>
Link: http://lkml.kernel.org/r/dc694ae2-f07f-61e1-7097-7c8411cee12d@gmail.com
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs/fs-writeback.c

@@ -331,11 +331,22 @@ struct inode_switch_wbs_context {
 	struct work_struct	work;
 };
 
+static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi)
+{
+	down_write(&bdi->wb_switch_rwsem);
+}
+
+static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi)
+{
+	up_write(&bdi->wb_switch_rwsem);
+}
+
 static void inode_switch_wbs_work_fn(struct work_struct *work)
 {
 	struct inode_switch_wbs_context *isw =
 		container_of(work, struct inode_switch_wbs_context, work);
 	struct inode *inode = isw->inode;
+	struct backing_dev_info *bdi = inode_to_bdi(inode);
 	struct address_space *mapping = inode->i_mapping;
 	struct bdi_writeback *old_wb = inode->i_wb;
 	struct bdi_writeback *new_wb = isw->new_wb;
@@ -343,6 +354,12 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
 	struct page *page;
 	bool switched = false;
 
+	/*
+	 * If @inode switches cgwb membership while sync_inodes_sb() is
+	 * being issued, sync_inodes_sb() might miss it.  Synchronize.
+	 */
+	down_read(&bdi->wb_switch_rwsem);
+
 	/*
 	 * By the time control reaches here, RCU grace period has passed
 	 * since I_WB_SWITCH assertion and all wb stat update transactions
@@ -428,6 +445,8 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
 	spin_unlock(&new_wb->list_lock);
 	spin_unlock(&old_wb->list_lock);
 
+	up_read(&bdi->wb_switch_rwsem);
+
 	if (switched) {
 		wb_wakeup(new_wb);
 		wb_put(old_wb);
@@ -468,9 +487,18 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
 	if (inode->i_state & I_WB_SWITCH)
 		return;
 
+	/*
+	 * Avoid starting new switches while sync_inodes_sb() is in
+	 * progress.  Otherwise, if the down_write protected issue path
+	 * blocks heavily, we might end up starting a large number of
+	 * switches which will block on the rwsem.
+	 */
+	if (!down_read_trylock(&bdi->wb_switch_rwsem))
+		return;
+
 	isw = kzalloc(sizeof(*isw), GFP_ATOMIC);
 	if (!isw)
-		return;
+		goto out_unlock;
 
 	/* find and pin the new wb */
 	rcu_read_lock();
@@ -504,12 +532,14 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
 	 * Let's continue after I_WB_SWITCH is guaranteed to be visible.
 	 */
 	call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn);
-	return;
+	goto out_unlock;
 
 out_free:
 	if (isw->new_wb)
 		wb_put(isw->new_wb);
 	kfree(isw);
+out_unlock:
+	up_read(&bdi->wb_switch_rwsem);
 }
 
 /**
@@ -887,6 +917,9 @@ fs_initcall(cgroup_writeback_init);
 
 #else	/* CONFIG_CGROUP_WRITEBACK */
 
+static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) { }
+static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) { }
+
 static struct bdi_writeback *
 locked_inode_to_wb_and_lock_list(struct inode *inode)
 	__releases(&inode->i_lock)
@@ -2413,8 +2446,11 @@ void sync_inodes_sb(struct super_block *sb)
 		return;
 	WARN_ON(!rwsem_is_locked(&sb->s_umount));
 
+	/* protect against inode wb switch, see inode_switch_wbs_work_fn() */
+	bdi_down_write_wb_switch_rwsem(bdi);
 	bdi_split_work_to_wbs(bdi, &work, false);
 	wb_wait_for_completion(bdi, &done);
+	bdi_up_write_wb_switch_rwsem(bdi);
 
 	wait_sb_inodes(sb);
 }

include/linux/backing-dev-defs.h

@@ -190,6 +190,7 @@ struct backing_dev_info {
 	struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
 	struct rb_root cgwb_congested_tree; /* their congested states */
 	struct mutex cgwb_release_mutex;  /* protect shutdown of wb structs */
+	struct rw_semaphore wb_switch_rwsem; /* no cgwb switch while syncing */
 #else
 	struct bdi_writeback_congested *wb_congested;
 #endif

mm/backing-dev.c

@@ -689,6 +689,7 @@ static int cgwb_bdi_init(struct backing_dev_info *bdi)
 	INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
 	bdi->cgwb_congested_tree = RB_ROOT;
 	mutex_init(&bdi->cgwb_release_mutex);
+	init_rwsem(&bdi->wb_switch_rwsem);
 
 	ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
 	if (!ret) {