Re: Proper error-handling of cdev_add() after cdev_alloc() ?

[Jan Kara]

Hello!

On Wed 27-05-26 16:01:31, Philipp Matthias Hahn wrote:
> > How should errors of `cdev_add()` after `cdev_alloc()` be handled?
> 
> sorry for ping-ing you directly, but so far got no answer and I would
> like to get a definitive answer on this before I spend more time on this
> to create and send patches. Thank you.

So I don't really feel like chrdev specialist but based on reading the
code:

> Am Tue, May 12, 2026 at 10:20:13AM +0200 schrieb Philipp Matthias Hahn:
> > The basic schema seems to be like this:
> > 
> > 	cdev = cdev_alloc();
> > 	if (!cdev)
> > 		return -ENOMEM;
> > 	...
> > 	ret = cdev_add(cdev, ...);
> > 	if (ret < 0) {
> > 		/* WHAT HERE ??? */
> > 		return ret;
> > 	}
> > 	...
> > 	return 0;  // success
> > 
> > I found 5 case of how errors are handled:
> > 
> > 1.  kfree(cdev);
> > 	drivers/media/v4l2-core/v4l2-dev.c:1060
> > 
> > 2.  kobject_put(&cdev->kobj);
> > 	drivers/char/xillybus/xillybus_class.c:107
> > 	fs/char_dev.c:293
> > 	drivers/s390/char/vmlogrdr.c:780
> > 	drivers/tty/tty_io.c:3171
> > 	drivers/uio/uio.c:925
> > 
> > 3.  cdev_del(cdev);
> > 	drivers/char/virtio_console.c:1435
> > 	fs/fuse/cuse.c:419
> > 	drivers/misc/mei/main.c:1318
> > 	drivers/s390/char/tape_class.c:92
> > 	drivers/s390/char/vmur.c:928
> > 	drivers/scsi/sg.c:1560
> > 	drivers/scsi/st.c:4307
> > 	drivers/staging/vme_user/vme_user.c:694
> > 
> > 4.  No error handling at all.
> > 	arch/sh/boards/mach-landisk/gio.c:145
> > 
> > 5.  panic()
> > 
> > Currently I'm unsure of what's the correct way.
> > My current understanding is, that `cdev_del()` should only be called if
> > `cdev_add()` succeeded, so those findings above should be fixed. But
> > to what?
> 
> I guess __register_chrdev() from fs/char_dev.c:293 is correct and
> 	kobject_put(&cdev->kobj);
> should be used.

This indeed looks like proper handling of the failure.

								Honza
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR