Re: [PATCH v2] perf/ring_buffer: Prefer struct_size over open coded arithmetic

[Christophe JAILLET]

Le 06/05/2024 à 18:23, Kees Cook a écrit :
> On Sun, May 05, 2024 at 07:31:24PM +0200, Erick Archer wrote:
> > On Sun, May 05, 2024 at 05:24:55PM +0200, Christophe JAILLET wrote:
> > > Le 05/05/2024 à 16:15, Erick Archer a écrit :
> > > > diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
> > > > index 4013408ce012..080537eff69f 100644
> > > > --- a/kernel/events/ring_buffer.c
> > > > +++ b/kernel/events/ring_buffer.c
> > > > @@ -822,9 +822,7 @@ struct perf_buffer *rb_alloc(int nr_pages, long watermark, int cpu, int flags)
> > > >    	unsigned long size;
> > >
> > > Hi,
> > >
> > > Should size be size_t?
> >
> > I'm sorry, but I don't have enough knowledge to answer this question.
> > The "size" variable is used as a return value by struct_size and as
> > a parameter to the order_base_2() and kzalloc_node() functions.
>
> For Linux, size_t and unsigned long are the same (currently).
> Pedantically, yes, this should be size_t, but it's the same.
>
> > [...]
> > > 	all_buf = vmalloc_user((nr_pages + 1) * PAGE_SIZE);
> > > 	if (!all_buf)
> > > 		goto fail_all_buf;
> > >
> > > 	rb->user_page = all_buf;
> > > 	rb->data_pages[0] = all_buf + PAGE_SIZE;
> > > 	if (nr_pages) {					<--- here
> > > 		rb->nr_pages = 1;			<---
> > > 		rb->page_order = ilog2(nr_pages);
> > > 	}
> > [...]
> > I think that we don't need to deal with the "nr_pages = 0" case
> > since the flex array will always have a length of one.
> >
> > Kees, can you help us with this?
>
> Agh, this code hurt my head for a while.
>
> all_buf contains "nr_pages + 1" pages. all_buf gets attached to
> rb->user_page, and then rb->data_pages[0] points to the second page in
> all_buf... which means, I guess, that rb->data_pages does only have 1
> entry.
>
> However, the nr_pages == 0 case is weird. Currently, data_pages[0] will
> still get set (which points ... off the end of all_buf). If we
> unconditionally set rb->nr_pages to 1, we're changing the behavior. If
> we _don't_ set rb->data_pages[0], we're changing the behavior, but I
> think it's an invalid pointer anyway, so this is the safer change to
> make. I suspect the right replacement is:
>
>
> diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
> index 4013408ce012..7d638ce76799 100644
> --- a/kernel/events/ring_buffer.c
> +++ b/kernel/events/ring_buffer.c
> @@ -916,15 +916,11 @@ void rb_free(struct perf_buffer *rb)
>   struct perf_buffer *rb_alloc(int nr_pages, long watermark, int cpu, int flags)
>   {
>   	struct perf_buffer *rb;
> -	unsigned long size;
>   	void *all_buf;
>   	int node;
>   
> -	size = sizeof(struct perf_buffer);
> -	size += sizeof(void *);
> -
>   	node = (cpu == -1) ? cpu : cpu_to_node(cpu);
> -	rb = kzalloc_node(size, GFP_KERNEL, node);
> +	rb = kzalloc_node(struct_size(rb, nr_pages, 1), GFP_KERNEL, node);
>   	if (!rb)
>   		goto fail;
>   
> @@ -935,9 +931,9 @@ struct perf_buffer *rb_alloc(int nr_pages, long watermark, int cpu, int flags)
>   		goto fail_all_buf;
>   
>   	rb->user_page = all_buf;
> -	rb->data_pages[0] = all_buf + PAGE_SIZE;
>   	if (nr_pages) {
>   		rb->nr_pages = 1;
> +		rb->data_pages[0] = all_buf + PAGE_SIZE;
>   		rb->page_order = ilog2(nr_pages);
>   	}

This is also what make the most sense to me.

CJ

>   
>
>
> Also, why does rb_alloc() take an "int" nr_pages? The only caller has an
> unsigned long argument for nr_pages. Nothing checks for >INT_MAX that I
> can find.