Re: INFO: task hung in __get_super

[Al Viro]

On Tue, Apr 30, 2019 at 03:07:39PM +0200, Jan Kara wrote:
> On Tue 30-04-19 04:11:44, Al Viro wrote:
> > On Tue, Apr 30, 2019 at 04:55:01AM +0200, Jan Kara wrote:
> > 
> > > Yeah, you're right. And if we push the patch a bit further to not take
> > > loop_ctl_mutex for invalid ioctl number, that would fix the problem. I
> > > can send a fix.
> > 
> > Huh?  We don't take it until in lo_simple_ioctl(), and that patch doesn't
> > get to its call on invalid ioctl numbers.  What am I missing here?
> 
> Doesn't it? blkdev_ioctl() calls into __blkdev_driver_ioctl() for
> unrecognized ioctl numbers. That calls into lo_ioctl() in case of a loop
> device. lo_ioctl() calls into lo_simple_ioctl() for ioctl numbers it
> doesn't recognize and lo_simple_ioctl() will lock loop_ctl_mutex as you
> say.

Not with the patch upthread.  lo_ioctl() part was

@@ -1567,10 +1564,9 @@ static int lo_ioctl(struct block_device *bdev, fmode_t mode,
 	case LOOP_SET_BLOCK_SIZE:
 		if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
 			return -EPERM;
-		/* Fall through */
+		return lo_simple_ioctl(lo, cmd, arg);
 	default:
-		err = lo_simple_ioctl(lo, cmd, arg);
-		break;
+		return -EINVAL;
 	}
 
 	return err;

so anything unrecognized doesn't make it to lo_simple_ioctl() at all.