[PATCH] wifi: mac80211: check the control channel before downgrading the bandwidth

[Johannes Berg]

Hi,

So I've looked at this patch a few times, but it just confuses me ...

On Thu, 2022-12-22 at 17:13 +0800, Michael Lee wrote:
> When the link fails to use the channel, chandef's bandwidth will be 
> downgraded without checking the control channel.
> The issue happens when the STA of an extender with limited channel 
> context associates with a root AP operating on a different channel.
> 
> Below is an example:
> 
>     ______________           ________________
>    | RootAP(ch36) |         | Extender(ch44) |
>    |              | (ASSOC) |       AP       |
>    |      AP <-------------------- STA       |
>    |______________|         |________________|
> 
> - RootAP is operating on channel 36, while Extender is operating
>   on channel 44

What does this matter? The extended is just a STA, no? Or are you saying
it's important that the extender has a concurrent AP interface that's on
channel 44?

And if you say "ch36" or "ch44" that's just the control channel (I
guess), but what's the actual complete channel configuration?

> - When associating with RootAP, Extender-STA downgrades the
>   chandef's bandwidth to be compatible with any channels on the phy

What do you mean by "on the phy" here? That's not mac80211 terminology,
so not sure.

> - Finally, chandef's bandwidth is downgraded to 20MHz and 
>   the association fails
> 
> In this patch, a control channel checking is added to avoid unnecessary
> bandwidth downgrading
> 
> Signed-off-by: Michael Lee <michael-cy.lee at mediatek.com>
> ---
>  net/mac80211/mlme.c | 24 ++++++++++++++++++++++++
>  1 file changed, 24 insertions(+)
> 
> diff --git a/net/mac80211/mlme.c b/net/mac80211/mlme.c
> index 0aee2392dd29..bc435e8508e2 100644
> --- a/net/mac80211/mlme.c
> +++ b/net/mac80211/mlme.c
> @@ -4616,6 +4616,27 @@ ieee80211_verify_sta_he_mcs_support(struct ieee80211_sub_if_data *sdata,
>  	return false;
>  }
>  
> +static bool
> +ieee80211_check_same_ctrl_channel(struct ieee80211_sub_if_data *sdata,
> +				  const struct cfg80211_chan_def *chandef)
> +{
> +	struct ieee80211_local *local = sdata->local;
> +	struct ieee80211_chanctx *ctx;
> +
> +	mutex_lock(&local->chanctx_mtx);
> +	list_for_each_entry(ctx, &local->chanctx_list, list) {
> +		if (ctx->replace_state == IEEE80211_CHANCTX_WILL_BE_REPLACED)
> +			continue;
> +		if (ctx->mode == IEEE80211_CHANCTX_EXCLUSIVE)
> +			continue;
> +		if (chandef->chan == ctx->conf.def.chan)
> +			return true;
> +	}
> +
> +	mutex_unlock(&local->chanctx_mtx);
> +	return false;
> +}
> +
>  static int ieee80211_prep_channel(struct ieee80211_sub_if_data *sdata,
>  				  struct ieee80211_link_data *link,
>  				  struct cfg80211_bss *cbss,
> @@ -4842,6 +4863,9 @@ static int ieee80211_prep_channel(struct ieee80211_sub_if_data *sdata,
>  	    chandef.width == NL80211_CHAN_WIDTH_10)
>  		goto out;
>  
> +	if (!ret || !ieee80211_check_same_ctrl_channel(sdata, &chandef))
> +		goto out;

Not sure I get how this is any different - you're describing a case
where "ret != 0" (because if ret == 0 nothing happens in the while
loop), so then you fail _anyway_? So what's the point?

johannes