Re: [PATCH] ASoC: Intel: Skylake: Replace 1-element array with flex-array

From: Amadeusz Sławiński
Date: Mon Feb 13 2023 - 07:52:58 EST


On 2/10/2023 8:06 PM, Kees Cook wrote:
On Fri, Feb 10, 2023 at 02:10:56PM +0100, Amadeusz Sławiński wrote:
On 2/10/2023 6:14 AM, Kees Cook wrote:
The kernel is globally removing the ambiguous 0-length and 1-element
arrays in favor of flexible arrays, so that we can gain both compile-time
and run-time array bounds checking[1]. In this instance, struct
skl_cpr_cfg contains struct skl_cpr_gtw_cfg, which defined "config_data"
as a 1-element array.

Normally when switching from a 1-element array to a flex-array, any
related size calculations must be adjusted too. However, it seems the
original code was over-allocating space, since 1 extra u32 would be
included by the sizeof():

param_size = sizeof(struct skl_cpr_cfg);
param_size += mconfig->formats_config[SKL_PARAM_INIT].caps_size;

But the copy uses caps_size bytes, and cap_size / 4 (i.e. sizeof(u32))
for the length tracking:

memcpy(cpr_mconfig->gtw_cfg.config_data,
mconfig->formats_config[SKL_PARAM_INIT].caps,
mconfig->formats_config[SKL_PARAM_INIT].caps_size);

cpr_mconfig->gtw_cfg.config_length =
(mconfig->formats_config[SKL_PARAM_INIT].caps_size) / 4;

Therefore, no size calculations need adjusting. Change the struct
skl_cpr_gtw_cfg config_data member to be a true flexible array, which
also fixes the over-allocation, and silences this memcpy run-time false
positive:

memcpy: detected field-spanning write (size 100) of single field "cpr_mconfig->gtw_cfg.config_data" at sound/soc/intel/skylake/skl-messages.c:554 (size 4)

[1] For lots of details, see both:
https://docs.kernel.org/process/deprecated.html#zero-length-and-one-element-arrays
https://people.kernel.org/kees/bounded-flexible-arrays-in-c

Reported-by: Sasa Ostrouska <casaxa@xxxxxxxxx>
Link: https://lore.kernel.org/all/CALFERdwvq5day_sbDfiUsMSZCQu9HG8-SBpOZDNPeMdZGog6XA@xxxxxxxxxxxxxx/
Cc: Cezary Rojewski <cezary.rojewski@xxxxxxxxx>
Cc: Pierre-Louis Bossart <pierre-louis.bossart@xxxxxxxxxxxxxxx>
Cc: Liam Girdwood <liam.r.girdwood@xxxxxxxxxxxxxxx>
Cc: Peter Ujfalusi <peter.ujfalusi@xxxxxxxxxxxxxxx>
Cc: Bard Liao <yung-chuan.liao@xxxxxxxxxxxxxxx>
Cc: Ranjani Sridharan <ranjani.sridharan@xxxxxxxxxxxxxxx>
Cc: Kai Vehmanen <kai.vehmanen@xxxxxxxxxxxxxxx>
Cc: Mark Brown <broonie@xxxxxxxxxx>
Cc: Jaroslav Kysela <perex@xxxxxxxx>
Cc: Takashi Iwai <tiwai@xxxxxxxx>
Cc: "Gustavo A. R. Silva" <gustavoars@xxxxxxxxxx>
Cc: "Amadeusz Sławiński" <amadeuszx.slawinski@xxxxxxxxxxxxxxx>
Cc: alsa-devel@xxxxxxxxxxxxxxxx
Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
---
sound/soc/intel/skylake/skl-topology.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sound/soc/intel/skylake/skl-topology.h b/sound/soc/intel/skylake/skl-topology.h
index 6db0fd7bad49..ad94f8020c27 100644
--- a/sound/soc/intel/skylake/skl-topology.h
+++ b/sound/soc/intel/skylake/skl-topology.h
@@ -115,7 +115,7 @@ struct skl_cpr_gtw_cfg {
u32 dma_buffer_size;
u32 config_length;
/* not mandatory; required only for DMIC/I2S */
- u32 config_data[1];
+ u32 config_data[];
} __packed;
struct skl_dma_control {

This fails in our validation.

Ah, okay. Thanks for checking!

Maybe we can use the union workaround, to
leave the size as is?

Following seems to work in manual test:
diff --git a/sound/soc/intel/skylake/skl-topology.h
b/sound/soc/intel/skylake/skl-topology.h
index 6db0fd7bad49..ffbd2e60fede 100644
--- a/sound/soc/intel/skylake/skl-topology.h
+++ b/sound/soc/intel/skylake/skl-topology.h
@@ -115,7 +115,10 @@ struct skl_cpr_gtw_cfg {
u32 dma_buffer_size;
u32 config_length;
/* not mandatory; required only for DMIC/I2S */
- u32 config_data[1];
+ union {
+ u32 x;
+ u32 config_data[0];
+ };

Yeah, that could work, though the last member would be:
DECLARE_FLEX_ARRAY(u32, config_data);
otherwise the array is 0 length (rather than a proper flex array).

But before that, let me see if I can track down where the size is being
used, in case we can avoid adding the padding.


I did spend some more time on this, apparently struct was missing one field according to IPC protocol. I've send fix for this.