diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
index e46e83e876001c..734866ab519413 100644
--- a/fs/btrfs/qgroup.c
+++ b/fs/btrfs/qgroup.c
@@ -2249,6 +2249,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 	int ret = 0;
 	int i;
 	u64 *i_qgroups;
+	bool committing = false;
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_root *quota_root;
 	struct btrfs_qgroup *srcgroup;
@@ -2256,7 +2257,25 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 	u32 level_size = 0;
 	u64 nums;
 
-	mutex_lock(&fs_info->qgroup_ioctl_lock);
+	/*
+	 * There are only two callers of this function.
+	 *
+	 * One in create_subvol() in the ioctl context, which needs to hold
+	 * the qgroup_ioctl_lock.
+	 *
+	 * The other one in create_pending_snapshot() where no other qgroup
+	 * code can modify the fs as they all need to either start a new trans
+	 * or hold a trans handler, thus we don't need to hold
+	 * qgroup_ioctl_lock.
+	 * This would avoid long and complex lock chain and make lockdep happy.
+	 */
+	spin_lock(&fs_info->trans_lock);
+	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
+		committing = true;
+	spin_unlock(&fs_info->trans_lock);
+
+	if (!committing)
+		mutex_lock(&fs_info->qgroup_ioctl_lock);
 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 		goto out;
 
@@ -2420,7 +2439,8 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 unlock:
 	spin_unlock(&fs_info->qgroup_lock);
 out:
-	mutex_unlock(&fs_info->qgroup_ioctl_lock);
+	if (!committing)
+		mutex_unlock(&fs_info->qgroup_ioctl_lock);
 	return ret;
 }