Re: [PATCH 2/2] s390/iucv: Fix vargs handling in iucv_alloc_device()

[Vasily Gorbik <gor@linux.ibm.com>]

On Tue, Aug 13, 2024 at 02:43:33PM +0200, Greg Kroah-Hartman wrote:
> On Tue, Aug 13, 2024 at 01:50:27PM +0200, Vasily Gorbik wrote:
> > On Tue, Aug 13, 2024 at 12:52:19PM +0200, Greg Kroah-Hartman wrote:
> > > On Tue, Aug 13, 2024 at 12:42:37PM +0200, Vasily Gorbik wrote:
> > > > From: Heiko Carstens <hca@linux.ibm.com>
> > > > 
> > > > iucv_alloc_device() gets a format string and a varying number of
> > > > arguments. This is incorrectly forwarded by calling dev_set_name() with
> > > > the format string and a va_list, while dev_set_name() expects also a
> > > > varying number of arguments.
> > > > 
> > > > Fix this and call kobject_set_name_vargs() instead which expects a
> > > > va_list parameter.
> > > 
> > > I don't understand, why can't dev_set_name() be called here?
> > > 
> > > Calling "raw" kobject functions is almost never the correct thing to be
> > > doing, ESPECIALLY as you have a struct device here.
> > 
> > struct device *iucv_alloc_device(const struct attribute_group **attrs,
> >                                  void *priv, const char *fmt, ...);
> > 
> > va_start(vargs, fmt); initializes vargs to point to the first argument after fmt.
> > 
> > __printf(2, 0) int kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list vargs);
> > 
> > __printf(2, 3) int dev_set_name(struct device *dev, const char *name, ...);
> > 
> > dev_set_name is expecting to receive individual variable arguments
> > directly (...), not a va_list.
> > 
> > The (...) in dev_set_name is meant to be expanded into individual
> > arguments, but when you pass a va_list to it, this expansion doesn't
> > happen. Instead, the va_list is just treated as a pointer or a single
> > argument, leading to undefined or incorrect behavior.
> > 
> > So, would it be okay to reuse kobject_set_name_vargs() here, or would you propose
> > introducing another helper just for this case? e.g.
> > 
> > int dev_set_name_vargs(struct device *dev, const char *fmt, va_list vargs)
> > {
> > ჻·······return kobject_set_name_vargs(&dev->kobj, fmt, vargs);
> > }
> > EXPORT_SYMBOL_GPL(dev_set_name_vargs)
> 
> This function makes more sense if you really want to do this.
> 
> But step back, why is this needed at all anyway?  No other subsystem or
> driver needs/wants this, what makes this api so special?  Why not figure
> out your name beforehand?

That's comming from this patch series:
https://lore.kernel.org/all/20240506194454.1160315-1-hca@linux.ibm.com/#t

"""
Unify IUCV device allocation as suggested by Arnd Bergmann in order
to get rid of code duplication in various device drivers.
"""

It just introduces and utilizes a couple of wrappers to reduce code
duplication, and in this case, introducing this level of indirection
led to the need for forwarding vargs.

And reimplementing parts of kobject_set_name_vargs to format the device
name beforehand is probably not what we want.