Re: [PATCH v2] net: mac80211: Add NULL checks for sta->sdata

[Simon Horman <simon.horman@xxxxxxxxxxxx>]

On Tue, Mar 21, 2023 at 05:31:22PM +0800, Jia-Ju Bai wrote:
> In a previous commit 69403bad97aa ("wifi: mac80211: sdata can be NULL
> during AMPDU start"), sta->sdata can be NULL, and thus it should be 
> checked before being used.
> 
> However, in the same call stack, sta->sdata is also used in the
> following functions:
> 
> ieee80211_ba_session_work()
>   ___ieee80211_stop_rx_ba_session(sta)
>     ht_dbg(sta->sdata, ...); -> No check
>     sdata_info(sta->sdata, ...); -> No check
>     ieee80211_send_delba(sta->sdata, ...) -> No check
>   ___ieee80211_start_rx_ba_session(sta)
>     ht_dbg(sta->sdata, ...); -> No check
>     ht_dbg_ratelimited(sta->sdata, ...); -> No check
>   ieee80211_tx_ba_session_handle_start(sta)
>     sdata = sta->sdata; if (!sdata) -> Add check by previous commit
>   ___ieee80211_stop_tx_ba_session(sdata)
>     ht_dbg(sta->sdata, ...); -> No check
>   ieee80211_start_tx_ba_cb(sdata)
>     sdata = sta->sdata; local = sdata->local -> No check
>   ieee80211_stop_tx_ba_cb(sdata)
>     ht_dbg(sta->sdata, ...); -> No check
> 
> Thus, to avoid possible null-pointer dereferences, the related checks
> should be added.
> 
> These bugs are reported by a static analysis tool implemented by myself, 
> and they are found by extending a known bug fixed in the previous commit. 
> Thus, they could be theoretical bugs.
> 
> Signed-off-by: Jia-Ju Bai <baijiaju@xxxxxxxxxxx>
> ---
> v2:
> * Fix an error reported by checkpatch.pl, and make the bug finding
>   process more clear in the description. Thanks for Simon's advice.
> ---
>  net/mac80211/agg-rx.c | 68 ++++++++++++++++++++++++++-----------------
>  net/mac80211/agg-tx.c | 16 ++++++++--
>  2 files changed, 55 insertions(+), 29 deletions(-)
> 
> diff --git a/net/mac80211/agg-rx.c b/net/mac80211/agg-rx.c
> index c6fa53230450..6616970785a2 100644
> --- a/net/mac80211/agg-rx.c
> +++ b/net/mac80211/agg-rx.c
> @@ -80,19 +80,21 @@ void ___ieee80211_stop_rx_ba_session(struct sta_info *sta, u16 tid,
>  	RCU_INIT_POINTER(sta->ampdu_mlme.tid_rx[tid], NULL);
>  	__clear_bit(tid, sta->ampdu_mlme.agg_session_valid);
>  
> -	ht_dbg(sta->sdata,
> -	       "Rx BA session stop requested for %pM tid %u %s reason: %d\n", > -	       sta->sta.addr, tid,
> -	       initiator == WLAN_BACK_RECIPIENT ? "recipient" : "initiator",
> -	       (int)reason);
> +	if (sta->sdata) {
> +		ht_dbg(sta->sdata,
> +		       "Rx BA session stop requested for %pM tid %u %s reason: %d\n",
> +		       sta->sta.addr, tid,
> +		       initiator == WLAN_BACK_RECIPIENT ? "recipient" : "initiator",
> +		       (int)reason);
> +	}

The first line of the body of ___ieee80211_stop_rx_ba_session() is:

	struct ieee80211_local *local = sta->sdata->local;

So a NULL pointer dereference will have occurred before
the checks this change adds to that function.


>  
> -	if (drv_ampdu_action(local, sta->sdata, &params))
> +	if (sta->sdata && drv_ampdu_action(local, sta->sdata, &params))
>  		sdata_info(sta->sdata,
>  			   "HW problem - can not stop rx aggregation for %pM tid %d\n",
>  			   sta->sta.addr, tid);
>  

...

> diff --git a/net/mac80211/agg-tx.c b/net/mac80211/agg-tx.c
> index f9514bacbd4a..03b31b6e7ac7 100644
> --- a/net/mac80211/agg-tx.c
> +++ b/net/mac80211/agg-tx.c
> @@ -368,8 +368,10 @@ int ___ieee80211_stop_tx_ba_session(struct sta_info *sta, u16 tid,
>  
>  	spin_unlock_bh(&sta->lock);
>  
> -	ht_dbg(sta->sdata, "Tx BA session stop requested for %pM tid %u\n",
> -	       sta->sta.addr, tid);
> +	if (sta->sdata) {
> +		ht_dbg(sta->sdata, "Tx BA session stop requested for %pM tid %u\n",
> +		       sta->sta.addr, tid);
> +	}

This seems clean :)

>  	del_timer_sync(&tid_tx->addba_resp_timer);
>  	del_timer_sync(&tid_tx->session_timer);
> @@ -776,7 +778,12 @@ void ieee80211_start_tx_ba_cb(struct sta_info *sta, int tid,
>  			      struct tid_ampdu_tx *tid_tx)
>  {
>  	struct ieee80211_sub_if_data *sdata = sta->sdata;
> -	struct ieee80211_local *local = sdata->local;
> +	struct ieee80211_local *local;
> +
> +	if (!sdata)
> +		return;

I'm not sure that silently ignoring non-existent sdata is the right approach.
Perhaps a WARN_ON or WARN_ONCE is appropriate?

> +
> +	local = sdata->local;
>  
>  	if (WARN_ON(test_and_set_bit(HT_AGG_STATE_DRV_READY, &tid_tx->state)))
>  		return;
> @@ -902,6 +909,9 @@ void ieee80211_stop_tx_ba_cb(struct sta_info *sta, int tid,
>  	bool send_delba = false;
>  	bool start_txq = false;
>  
> +	if (!sdata)
> +		return;
> +

Ditto.

>  	ht_dbg(sdata, "Stopping Tx BA session for %pM tid %d\n",
>  	       sta->sta.addr, tid);
>