Re: [PATCH v2 1/2] media: staging/intel-ipu3: css: Fix wrong size comparison

[Sakari Ailus <sakari.ailus@linux.intel.com>]

Hi Gustavo,

Apologies for the delay.

On Mon, Aug 02, 2021 at 08:46:20AM -0500, Gustavo A. R. Silva wrote:
> Hi Sakari,
> 
> On 8/2/21 01:05, Sakari Ailus wrote:
> > Hi Gustavo,
> > 
> > I missed you already had sent v2...
> > 
> > On Fri, Jul 30, 2021 at 07:08:13AM -0500, Gustavo A. R. Silva wrote:
> >> There is a wrong comparison of the total size of the loaded firmware
> >> css->fw->size with the size of a pointer to struct imgu_fw_header.
> >>
> >> Fix this by using the right operand 'struct imgu_fw_header' for
> >> sizeof, instead of 'struct imgu_fw_header *' and turn binary_header
> >> into a flexible-array member. Also, adjust the relational operator
> >> to be '<=' instead of '<', as it seems that the intention of the
> >> comparison is to determine if the loaded firmware contains any
> >> 'struct imgu_fw_info' items in the binary_header[] array than merely
> >> the file_header (struct imgu_fw_bi_file_h).
> >>
> >> The replacement of the one-element array with a flexible-array member
> >> also help with the ongoing efforts to globally enable -Warray-bounds
> >> and get us closer to being able to tighten the FORTIFY_SOURCE routines
> >> on memcpy().
> >>
> >> Link: https://github.com/KSPP/linux/issues/79
> >> Link: https://github.com/KSPP/linux/issues/109
> >> Fixes: 09d290f0ba21 ("media: staging/intel-ipu3: css: Add support for firmware management")
> >> Cc: stable@vger.kernel.org
> >> Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
> >> ---
> >>
> >> It'd be just great if someone that knows this code better can confirm
> >> these changes are correct. In particular the adjustment of the
> >> relational operator. Thanks!
> >>
> >> Changes in v2:
> >>  - Use flexible array and adjust relational operator, accordingly.
> > 
> > The operator was just correct. The check is just there to see the firmware
> > is at least as large as the struct as which it is being accessed.
> 
> I'm a bit confused, so based on your reply to v1 of this series, this patch
> is now correct, right?
> 
> The operator in v1 _was_ correct as long as the one-element array wasn't
> transformed into a flexible array, right?
> 
> Notice that generally speaking flexible-array members don't occupy space in the
> containing structure:
> 
> $ pahole -C imgu_fw_header drivers/staging/media/ipu3/ipu3-css-fw.o
> struct imgu_fw_header {
> 	struct imgu_fw_bi_file_h   file_header;          /*     0    72 */
> 	/* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
> 	struct imgu_fw_info        binary_header[] __attribute__((__aligned__(8))); /*    72     0 */
> 
> 	/* size: 72, cachelines: 2, members: 2 */
> 	/* forced alignments: 1 */
> 	/* last cacheline: 8 bytes */
> } __attribute__((__aligned__(8)));
> 
> $ pahole -C imgu_fw_header drivers/staging/media/ipu3/ipu3-css-fw.o
> struct imgu_fw_header {
> 	struct imgu_fw_bi_file_h   file_header;          /*     0    72 */
> 	/* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
> 	struct imgu_fw_info        binary_header[1] __attribute__((__aligned__(8))); /*    72  1200 */
> 
> 	/* size: 1272, cachelines: 20, members: 2 */
> 	/* forced alignments: 1 */
> 	/* last cacheline: 56 bytes */
> } __attribute__((__aligned__(8)));
> 
> So, now that the flexible array transformation is included in the same patch as the
> bugfix, the operator is changed from '<' to '<='

'<' is correct since you only need as much data as the struct you're about
to access is large, not a byte more than that. As Dan noted.

I think you could add a check for binary_nr is at least one.

-- 
Kind regards,

Sakari Ailus