[PATCH] mwifiex: Fix possible ABBA deadlock

From: Brian Norris <briannorris@chromium.org>
To: Jia-Ju Bai <baijiaju1990@gmail.com>
Cc: amitkarwar@gmail.com, ganapathi017@gmail.com,
	sharvari.harisangam@nxp.com, huxinming820@gmail.com,
	kvalo@codeaurora.org, David Miller <davem@davemloft.net>,
	kuba@kernel.org, linux-wireless@vger.kernel.org,
	netdev@vger.kernel.org,
	linux-kernel <linux-kernel@vger.kernel.org>,
	Doug Anderson <dianders@chromium.org>
Subject: [PATCH] mwifiex: Fix possible ABBA deadlock
Date: Mon, 29 Nov 2021 16:47:34 -0800	[thread overview]
Message-ID: <YaV0pllJ5p/EuUat@google.com> (raw)
In-Reply-To: <0e495b14-efbb-e0da-37bd-af6bd677ee2c@gmail.com>

Quoting Jia-Ju Bai <baijiaju1990@gmail.com>:

  mwifiex_dequeue_tx_packet()
     spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 1432 (Lock A)
     mwifiex_send_addba()
       spin_lock_bh(&priv->sta_list_spinlock); --> Line 608 (Lock B)

  mwifiex_process_sta_tx_pause()
     spin_lock_bh(&priv->sta_list_spinlock); --> Line 398 (Lock B)
     mwifiex_update_ralist_tx_pause()
       spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 941 (Lock A)

Similar report for mwifiex_process_uap_tx_pause().

While the locking expectations in this driver are a bit unclear, the
Fixed commit only intended to protect the sta_ptr, so we can drop the
lock as soon as we're done with it.

IIUC, this deadlock cannot actually happen, because command event
processing (which calls mwifiex_process_sta_tx_pause()) is
sequentialized with TX packet processing (e.g.,
mwifiex_dequeue_tx_packet()) via the main loop (mwifiex_main_process()).
But it's good not to leave this potential issue lurking.

Fixes: ("f0f7c2275fb9 mwifiex: minor cleanups w/ sta_list_spinlock in cfg80211.c")
Cc: Douglas Anderson <dianders@chromium.org>
Reported-by: TOTE Robot <oslab@tsinghua.edu.cn>
Link: https://lore.kernel.org/linux-wireless/0e495b14-efbb-e0da-37bd-af6bd677ee2c@gmail.com/
Signed-off-by: Brian Norris <briannorris@chromium.org>
---

On Tue, Nov 23, 2021 at 11:31:34AM +0800, Jia-Ju Bai wrote:
> I am not quite sure whether these possible deadlocks are real and how to fix
> them if they are real.
> Any feedback would be appreciated, thanks :)

I think these are at least theoretically real, and so we should take
something like the $subject patch probably. But I don't believe we can
actually hit this due to the main-loop structure of this driver.

Anyway, see the surrounding patch.

Thanks,
Brian


 drivers/net/wireless/marvell/mwifiex/sta_event.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/wireless/marvell/mwifiex/sta_event.c b/drivers/net/wireless/marvell/mwifiex/sta_event.c
index 80e5d44bad9d..7d42c5d2dbf6 100644
--- a/drivers/net/wireless/marvell/mwifiex/sta_event.c
+++ b/drivers/net/wireless/marvell/mwifiex/sta_event.c
@@ -365,10 +365,12 @@ static void mwifiex_process_uap_tx_pause(struct mwifiex_private *priv,
 		sta_ptr = mwifiex_get_sta_entry(priv, tp->peermac);
 		if (sta_ptr && sta_ptr->tx_pause != tp->tx_pause) {
 			sta_ptr->tx_pause = tp->tx_pause;
+			spin_unlock_bh(&priv->sta_list_spinlock);
 			mwifiex_update_ralist_tx_pause(priv, tp->peermac,
 						       tp->tx_pause);
+		} else {
+			spin_unlock_bh(&priv->sta_list_spinlock);
 		}
-		spin_unlock_bh(&priv->sta_list_spinlock);
 	}
 }
 
@@ -400,11 +402,13 @@ static void mwifiex_process_sta_tx_pause(struct mwifiex_private *priv,
 			sta_ptr = mwifiex_get_sta_entry(priv, tp->peermac);
 			if (sta_ptr && sta_ptr->tx_pause != tp->tx_pause) {
 				sta_ptr->tx_pause = tp->tx_pause;
+				spin_unlock_bh(&priv->sta_list_spinlock);
 				mwifiex_update_ralist_tx_pause(priv,
 							       tp->peermac,
 							       tp->tx_pause);
+			} else {
+				spin_unlock_bh(&priv->sta_list_spinlock);
 			}
-			spin_unlock_bh(&priv->sta_list_spinlock);
 		}
 	}
 }
-- 
2.34.0.rc2.393.gf8c9666880-goog
