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
