Re: [PATCH v1 1/5] devres: Introduce devm_kmemdup_array()

[Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>]

On Mon, Nov 25, 2024 at 08:29:10AM -0800, Dmitry Torokhov wrote:
> On Mon, Nov 25, 2024 at 05:08:13PM +0200, Raag Jadav wrote:
> > On Mon, Nov 25, 2024 at 09:49:22AM +0200, Andy Shevchenko wrote:
> > > On Sun, Nov 24, 2024 at 07:03:36AM +0000, Dmitry Torokhov wrote:
> > > > On Sun, Nov 24, 2024 at 01:35:23AM +0530, Raag Jadav wrote:

...

> > > > > Introduce '_array' variant of devm_kmemdup() for the users which lack
> > > > > multiplication overflow check.
> > > > 
> > > > I am not sure that this new helper is needed. Unlike allocators for
> > > > brand new objects, such as kmalloc_array(), devm_kmemdup() makes a copy
> > > > of already existing object, which is supposed to be a valid object and
> > > > therefore will have a reasonable size. So there should be no chance for
> > > > hitting this overflow unless the caller is completely confused and calls
> > > > devm_kmemdup() with random arguments (in which case all bets are off).
> > > 
> > > Don't we want to have a code more robust even if all what you say applies?
> > > Also this makes the call consistent with zillions of others from the alloc
> > > family of calls in the Linux kernel.
> 
> Having a clean API is fine, just do not bill it as something that is
> "safer". As I mentioned, unlike other allocators this one is supposed to
> operate with a valid source object and size passed to devm_kmemdup()
> should not exceed the size of the source object. There is no chance of
> overflowing.

Agree.

> > Agree. Although shooting in the foot is never the expectation, it is
> > atleast better than having to debug such unexpected cases.
> 
> Then maybe have a BUG() there instead of returning NULL? I know BUG()s
> are frowned upon, but I think in this case overflow is really an
> indicator of a hard error by the caller which is passing garbage
> arguments to this function.
> 
> Hm, I see we have kmemdup_array() already. Ok. How about making your
> devm_kmemdup_array() be similar to kmemdup_array()?
> 
> static inline void *devm_kmemdup_array(struct device *dev, const void *src,
> 				       size_t n, size_t size, gfp_t flags)
> {
> 	return devm_kmemdup(dev, src, size_mul(size, n), flags);
> }
> 
> This will trigger a warning on a too large order of allocation in
> mm/page_alloc.c::__alloc_pages_noprof().

This is nice! I have overlooked that kmemdup_array() uses size_mul()
instead of a check. Raag, can you rebuild your series on this?

-- 
With Best Regards,
Andy Shevchenko