Re: [PATCH] wifi: cfg80211: fix a possible memory leak

[Johannes Berg]

On Tue, 2022-11-01 at 15:19 -0500, Dinh Nguyen wrote:
> Klockworks
> 
You probably mean "klocwork" :)

>  reported a possible memory leak when
> cfg80211_inform_single_bss_data() return on an error and ies is left
> allocated.
> 
> Fixes: 0e227084aee3 ("cfg80211: clarify BSS probe response vs. beacon data")
> Signed-off-by: Dinh Nguyen <dinguyen@xxxxxxxxxx>
> ---
>  net/wireless/scan.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/net/wireless/scan.c b/net/wireless/scan.c
> index 806a5f1330ff..3c81dc17e079 100644
> --- a/net/wireless/scan.c
> +++ b/net/wireless/scan.c
> @@ -2015,8 +2015,10 @@ cfg80211_inform_single_bss_data(struct wiphy *wiphy,
>  
>  	signal_valid = data->chan == channel;
>  	res = cfg80211_bss_update(wiphy_to_rdev(wiphy), &tmp, signal_valid, ts);
> -	if (!res)
> +	if (!res) {
> +		kfree(ies);
>  		return NULL;
> +	}
> 

To be honest this makes me a bit nervous - the function will take over
ownership of the tmp BSS in many cases if not all. Not saying it doesn't
have a bug, but at least one case inside of it *does* free it even in
the case of returning NULL and then you have a double-free?

So I think you didn't look at the code closely enough. Please do check
and follow up with a proper fix.

johannes