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

[Christophe JAILLET]

Le 05/05/2024 à 16:15, Erick Archer a écrit :
> This is an effort to get rid of all multiplications from allocation
> functions in order to prevent integer overflows [1][2].
>
> As the "rb" variable is a pointer to "struct perf_buffer" and this
> structure ends in a flexible array:
>
> struct perf_buffer {
> 	[...]
> 	void	*data_pages[];
> };
>
> the preferred way in the kernel is to use the struct_size() helper to
> do the arithmetic instead of the calculation "size + count * size" in
> the kzalloc_node() functions.
>
> In the "rb_alloc" function defined in the else branch of the macro
>
>   #ifndef CONFIG_PERF_USE_VMALLOC
>
> the count in the struct_size helper is the literal "1" since only one
> pointer to void is allocated. Also, remove the "size" variable as it
> is no longer needed.
>
> At the same time, prepare for the coming implementation by GCC and Clang
> of the __counted_by attribute. Flexible array members annotated with
> __counted_by can have their accesses bounds-checked at run-time via
> CONFIG_UBSAN_BOUNDS (for array indexing) and CONFIG_FORTIFY_SOURCE (for
> strcpy/memcpy-family functions). In this case, it is important to note
> that the logic needs a little refactoring to ensure that the "nr_pages"
> member is initialized before the first access to the flex array.
>
> In one case, it is only necessary to move the assignment before the
> array-writing loop while in the other case the assignment needs to be
> added.
>
> This way, the code is more safer.
>
> This code was detected with the help of Coccinelle, and audited and
> modified manually.
>
> Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#open-coded-arithmetic-in-allocator-arguments [1]
> Link: https://github.com/KSPP/linux/issues/160 [2]
> Signed-off-by: Erick Archer <erick.archer@xxxxxxxxxxx>
> ---
> Changes in v2:
> - Annotate "struct perf_buffer" with __counted_by() attribute (Kees Cook).
> - Refactor the logic to gain __counted_by() coverage (Kees Cook).
>
> Previous versions:
> v1 -> https://lore.kernel.org/linux-hardening/AS8PR02MB72372AB065EA8340D960CCC48B1B2@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
>
> Hi Peter,
>
> I know that you detest the struct_size() helper, however, as Kees
> explained in v1, this change improves the robustness of the code.
> Also, we will gain __counted_by() coverage.
>
> I hope this patch can be applied this time.
>
> Regards,
> Erick
> ---
>   kernel/events/internal.h    |  2 +-
>   kernel/events/ring_buffer.c | 14 ++++----------
>   2 files changed, 5 insertions(+), 11 deletions(-)
>
> diff --git a/kernel/events/internal.h b/kernel/events/internal.h
> index 5150d5f84c03..dc8d39b01adb 100644
> --- a/kernel/events/internal.h
> +++ b/kernel/events/internal.h
> @@ -55,7 +55,7 @@ struct perf_buffer {
>   	void				*aux_priv;
>   
>   	struct perf_event_mmap_page	*user_page;
> -	void				*data_pages[];
> +	void				*data_pages[] __counted_by(nr_pages);
>   };
>   
>   extern void rb_free(struct perf_buffer *rb);
> 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?

>   	int i, node;
>   
> -	size = sizeof(struct perf_buffer);
> -	size += nr_pages * sizeof(void *);
> -
> +	size = struct_size(rb, data_pages, nr_pages);
>   	if (order_base_2(size) > PAGE_SHIFT+MAX_PAGE_ORDER)
>   		goto fail;
>   
> @@ -833,6 +831,7 @@ struct perf_buffer *rb_alloc(int nr_pages, long watermark, int cpu, int flags)
>   	if (!rb)
>   		goto fail;
>   
> +	rb->nr_pages = nr_pages;
>   	rb->user_page = perf_mmap_alloc_page(cpu);
>   	if (!rb->user_page)
>   		goto fail_user_page;
> @@ -843,8 +842,6 @@ struct perf_buffer *rb_alloc(int nr_pages, long watermark, int cpu, int flags)
>   			goto fail_data_pages;
>   	}
>   
> -	rb->nr_pages = nr_pages;
> -
>   	ring_buffer_init(rb, watermark, flags);
>   
>   	return rb;
> @@ -916,18 +913,15 @@ 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, data_pages, 1), GFP_KERNEL, node);
>   	if (!rb)
>   		goto fail;
>   
> +	rb->nr_pages = nr_pages;

I don't think this is correct.

There is already a logic in place about it a few lines below:

	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 what is needed is to move this block just 2 lines above, 
(before rb->data_pages[0] = ...)


I'm also wondering what should be done if nr_pages = 0.


CJ

>   	INIT_WORK(&rb->work, rb_free_work);
>   
>   	all_buf = vmalloc_user((nr_pages + 1) * PAGE_SIZE);