Re: [PATCH 4.19 63/92] libertas: Fix two buffer overflows at parsing bss descriptor

[Pavel Machek]

Hi!

> From: Wen Huang <huangwenabc@xxxxxxxxx>
> 
> commit e5e884b42639c74b5b57dc277909915c0aefc8bb upstream.

> --- a/drivers/net/wireless/marvell/libertas/cfg.c
> +++ b/drivers/net/wireless/marvell/libertas/cfg.c
> @@ -1717,6 +1721,9 @@ static int lbs_ibss_join_existing(struct
>  	struct cmd_ds_802_11_ad_hoc_join cmd;
>  	u8 preamble = RADIO_PREAMBLE_SHORT;
>  	int ret = 0;
> +	int hw, i;
> +	u8 rates_max;
> +	u8 *rates;
>  
>  	/* TODO: set preamble based on scan result */
>  	ret = lbs_set_radio(priv, preamble, 1);
> @@ -1775,9 +1782,12 @@ static int lbs_ibss_join_existing(struct
>  	if (!rates_eid) {
>  		lbs_add_rates(cmd.bss.rates);
>  	} else {
> -		int hw, i;
> -		u8 rates_max = rates_eid[1];
> -		u8 *rates = cmd.bss.rates;
> +		rates_max = rates_eid[1];

I believe original version (with variables being local to the else)
was better.

> +		if (rates_max > MAX_RATES) {
> +			lbs_deb_join("invalid rates");
> +			goto out;
> +		}
> +		rates = cmd.bss.rates;

"goto out" goes to "return ret". ret will be 0 at this point, so this
will return success. I don't think that's right.

Best regards,
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Attachment: signature.asc
Description: Digital signature