Explain why zfs_dirty_inode need TXG_WAIT

module/zfs/zfs_vnops.c

@@ -4633,6 +4633,7 @@ zfs_dirty_inode(struct inode *ip, int flags)
 	sa_bulk_attr_t	bulk[4];
 	int		error = 0;
 	int		cnt = 0;
+	boolean_t       waited = B_FALSE;
 
 	if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
 		return (0);
@@ -4656,15 +4657,20 @@ zfs_dirty_inode(struct inode *ip, int flags)
 
 top:
 	tx = dmu_tx_create(zfsvfs->z_os);
-
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
 
-	boolean_t waited = B_FALSE;
+	/*
+	 * Despite this function allows an error to be returned, it's called
+	 * from zpl_dirty_inode() which is a Linux VFS callback functions
+	 * (.dirty_inode) which must always succeed, so we have to assign a
+	 * txg with TXG_NOTHROTTLE plus TX_WAIT when the first TXG_NOWAIT
+	 * call failed.
+	 */
 	error = dmu_tx_assign(tx,
 	    waited ? (TXG_NOTHROTTLE | TXG_WAIT) : TXG_NOWAIT);
 	if (error) {
-		if (error == ERESTART && waited == B_FALSE) {
+		if (error == ERESTART) {
 			waited = B_TRUE;
 			dmu_tx_wait(tx);
 			dmu_tx_abort(tx);