Re: [RFC] staging: wlan-ng: Driver broken since kernel 5.15

[Dan Carpenter <dan.carpenter@linaro.org>]

On Sun, Mar 17, 2024 at 09:20:34PM +0100, Arnd Bergmann wrote:
> On Sun, Mar 17, 2024, at 21:07, Philipp Hortmann wrote:
> > On 3/11/24 08:04, Dan Carpenter wrote:
> >> On Sat, Mar 09, 2024 at 11:09:24PM +0100, Philipp Hortmann wrote:
> > You are right with the statement that it is this commit.
> > commit ea82ff749587807fa48e3277c977ff3cec266f25 (HEAD)
> > Author: Lee Jones <lee.jones@linaro.org>
> > Date:   Wed Apr 14 19:10:39 2021 +0100
> >
> >      staging: wlan-ng: cfg80211: Move large struct onto the heap
> >
> >      Fixes the following W=1 kernel build warning(s):
> >
> >       drivers/staging/wlan-ng/cfg80211.c: In function ‘prism2_scan’:
> >       drivers/staging/wlan-ng/cfg80211.c:388:1: warning: the frame size 
> > of 1296 bytes is larger than 1024 bytes [-Wframe-larger-than=]
> >
> > But It is not depending on the line you pointed to.
> 
> Right, the kzalloc() already clears the data, so the memset
> is not needed.
> 

No, it's inside a loop so it needs to be cleared on each iteration.

> > I need another week to look into this.
> 
> I'm fairly sure this fixes the bug, the problem here was that
> the cast to (u8 *) hides the incorrect conversion:
> 
> diff --git a/drivers/staging/wlan-ng/cfg80211.c b/drivers/staging/wlan-ng/cfg80211.c
> index 471bb310176f..9d6a2dd35ba9 100644
> --- a/drivers/staging/wlan-ng/cfg80211.c
> +++ b/drivers/staging/wlan-ng/cfg80211.c
> @@ -350,7 +350,7 @@ static int prism2_scan(struct wiphy *wiphy,
>  		msg2->msgcode = DIDMSG_DOT11REQ_SCAN_RESULTS;
>  		msg2->bssindex.data = i;
>  
> -		result = p80211req_dorequest(wlandev, (u8 *)&msg2);
> +		result = p80211req_dorequest(wlandev, (u8 *)msg2);

Ah, well done.

It feels like this is the kind of bug which should be caught with
static analysis.  One of the things that people want from static
analysis is looking at what a patch does.  So if we pass &msg2 and the
patch moved msg from the stack to be kmalloc()ed, then print a warning.
It's not something that Smatch does.

I have my rename_rev.pl script (https://github.com/error27/rename_rev)
which I use to filter out variable renames or see if (1 << foo) is
converted to BIT(foo) correctly.  Maybe I could extend that to check
"move stack to heap" patches...

regards,
dan carpenter