Re: [PATCH v2 2/2] wifi: mwifiex: Annotate mwifiex_ie_types_wildcard_ssid_params with __counted_by()

[Brian Norris <briannorris@chromium.org>]

Hi Alper,

On Tue, Oct 08, 2024 at 01:20:55AM +0300, Alper Nebi Yasak wrote:
> Add the __counted_by compiler attribute to the flexible array member
> `ssid` to improve access bounds-checking via CONFIG_UBSAN_BOUNDS and
> CONFIG_FORTIFY_SOURCE.
> 
> Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
> ---
> I've mimicked the commit messages from `git log -S"__counted_by("`.
> Since they refer to UBSAN I tried testing with CONFIG_KASAN (w/
> kasan_multi_shot), CONFIG_UBSAN and CONFIG_FORTIFY_SOURCE. I do not
> get any errors relating to this module with those options, but have
> others -- I even had to test on another board altogether.

It's possible you don't have a new enough compiler to enable the
improved array bounds checks? Supposedly, that's new in GCC 15 and Clang
18. But I'm not too familiar.

> This attribute was suggested in reviews, I don't fully understand it,
> but I am not sure it is correct in the context of this comment from
> scan.c (with irrelelvant parts omitted):
> 
>     ssid_len = user_scan_in->ssid_list[i].ssid_len;
> 
>     wildcard_ssid_tlv =
>             (struct mwifiex_ie_types_wildcard_ssid_params *)
>             tlv_pos;
> 
>     /*
>      * max_ssid_length = 0 tells firmware to perform
>      * specific scan for the SSID filled, whereas
>      * max_ssid_length = IEEE80211_MAX_SSID_LEN is for
>      * wildcard scan.
>      */
>     if (ssid_len)
>             wildcard_ssid_tlv->max_ssid_length = 0;
>     else
>             wildcard_ssid_tlv->max_ssid_length =
>                                     IEEE80211_MAX_SSID_LEN;
> 
>     memcpy(wildcard_ssid_tlv->ssid,
>            user_scan_in->ssid_list[i].ssid, ssid_len);
> 
> I expect we want to use __counted_by(ssid_len) instead, but do not have
> it in the struct. And max_ssid_length = 0 when ssid[] is non-empty, so
> is it really appropriate as the __counted_by()? But then again, I
> couldn't get this to produce a warning.

The correct length would be derived from header.len. But IIUC,
__counted_by neither supports nested structs, nor derived values.

But anyway, like you suspect, __counted_by(max_ssid_length) is actually
wrong.

I'd be happy to be enlighted by other experts on CC, but otherwise,
that's a "NACK" for patch 2. Thanks for looking into this though!

Regards,
Brian

> Changes in v2:
> - Add patch to annotate ssid field with __counted_by(max_ssid_length)
> 
>  drivers/net/wireless/marvell/mwifiex/fw.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/net/wireless/marvell/mwifiex/fw.h b/drivers/net/wireless/marvell/mwifiex/fw.h
> index 4a96281792cc..e4e35ba35454 100644
> --- a/drivers/net/wireless/marvell/mwifiex/fw.h
> +++ b/drivers/net/wireless/marvell/mwifiex/fw.h
> @@ -875,7 +875,7 @@ struct mwifiex_ietypes_chanstats {
>  struct mwifiex_ie_types_wildcard_ssid_params {
>  	struct mwifiex_ie_types_header header;
>  	u8 max_ssid_length;
> -	u8 ssid[];
> +	u8 ssid[] __counted_by(max_ssid_length);
>  } __packed;
>  
>  #define TSF_DATA_SIZE            8
> -- 
> 2.45.2
>