Re: [PATCH] perf/ring_buffer: Prefer struct_size over open coded arithmetic

[Erick Archer]

On Thu, May 02, 2024 at 03:55:36PM -0700, Kees Cook wrote:
> On Thu, May 02, 2024 at 11:18:37AM +0200, Peter Zijlstra wrote:
> > On Wed, May 01, 2024 at 01:21:42PM -0700, Kees Cook wrote:
> > > On Tue, Apr 30, 2024 at 11:15:04AM +0200, Peter Zijlstra wrote:
> > > > On Mon, Apr 29, 2024 at 07:40:58PM +0200, Erick Archer wrote:
> > > > > This is an effort to get rid of all multiplications from allocation
> > > > > functions in order to prevent integer overflows [1][2].
> > > > 
> > > > So personally I detest struct_size() because I can never remember wtf it
> > > > does, whereas the code it replaces is simple and straight forward :/
> > > 
> > > Sure, new APIs can involved a learning curve. If we can all handle
> > > container_of(), we can deal with struct_size(). :)
> > 
> > containre_of() is actually *much* shorter than typing it all out. Which
> > is a benefit.
> > 
> > struct_size() not so much. That's just obfuscation for obfuscation's
> > sake.

I do not agree with this.
> 
> It's really not -- it's making sure that the calculation is semantically
> sane: all the right things are being used for the struct size calculation
> and things can't "drift", if types change, flex array changes, etc. It's
> both a code robustness improvement and a wrap-around stopping improvement.
> 

Also, in the "Deprecated Interfaces, Language Features, Attributes, and
Conventions" [1] it says verbatim:

   Another common case to avoid is calculating the size of a structure
   with a trailing array of others structures, as in:

   header = kzalloc(sizeof(*header) + count * sizeof(*header->item),
                    GFP_KERNEL);

   Instead, use the helper:

   header = kzalloc(struct_size(header, item, count), GFP_KERNEL);

Therefore, if there is a convention to follow, we should not make an
exception. Moreover, struct_size is widely used in the kernel and
widely accepted. Also makes the code safer.

So, I will send a new patch with the changes Kees proposed and I
hope that it will be the first step in the adoption of struct_size
in the perf and sched subsystems ;)

Regards,
Erick

[1] https://docs.kernel.org/process/deprecated.html

> -- 
> Kees Cook