Re: [PATCH] ath9k: Fix possible data races in ath_set_channel()

[Felix Fietkau]

On 2020-01-11 18:15, Jia-Ju Bai wrote:
> The functions ath9k_config() and ath_ani_calibrate() may be concurrently
> executed.
> 
> A variable survey->filled is accessed with holding a spinlock
> common->cc_lock, through:
> ath_ani_calibrate()
>     spin_lock_irqsave(&common->cc_lock, flags);
>     ath_update_survey_stats()
>         ath_update_survey_nf()
>             survey->filled |= SURVEY_INFO_NOISE_DBM;
> 
> The identical variables sc->cur_survey->filled and 
> sc->survey[pos].filled is accessed without holding this lock, through:
> ath9k_config()
>     ath_chanctx_set_channel()
>         ath_set_channel()
>             sc->cur_survey->filled &= ~SURVEY_INFO_IN_USE;
>             sc->cur_survey->filled |= SURVEY_INFO_IN_USE;
>             else if (!(sc->survey[pos].filled & SURVEY_INFO_IN_USE))
>             ath_update_survey_nf
>                 survey->filled |= SURVEY_INFO_NOISE_DBM;
> 
> Thus, possible data races may occur.
> 
> To fix these data races, in ath_set_channel(), these variables are
> accessed with holding the spinlock common->cc_lock.
> 
> These data races are found by the runtime testing of our tool DILP-2.
> 
> Signed-off-by: Jia-Ju Bai <baijiaju1990@xxxxxxxxx>
I think a much better solution would be to stop common->ani.timer
earlier in ath_set_channel to prevent this race from occurring.

- Felix