Received-SPF: pass (google.com: domain of linux-wireless-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Jean-Pierre TOSONI <jp.tosoni@acksys.fr>
To:     "Johannes Berg (johannes@sipsolutions.net)" 
        <johannes@sipsolutions.net>
CC:     "linux-wireless@vger.kernel.org" <linux-wireless@vger.kernel.org>
Subject: [RFC] mac80211: fix rx blockack session race condition
Thread-Topic: [RFC] mac80211: fix rx blockack session race condition
Thread-Index: AdfKa2Vnf+51pzw3S2ywVKfXICdlgQ==
Date:   Tue, 26 Oct 2021 14:04:41 +0000
Message-ID: <DB9PR01MB735472EB045E560573F9C369E4849@DB9PR01MB7354.eurprd01.prod.exchangelabs.com>
Accept-Language: fr-FR, en-US
Content-Language: fr-FR
x-ms-exchange-antispam-messagedata-chunkcount: 1
x-ms-exchange-antispam-messagedata-0: bs2CZ+9SY/aE0RzK39DU83fLAE2i0NDnsScOS6cerVbKURbXrCT/JP3c6QmygiBF3v4uRDwgUyJV+am+g4cWTuss+lE3VHfqxBKf/zmoZFTAQpcN5oSkWeyUHSxIh/6lRu0yOviI/DwDWxIPGu8M6RQ3LXpRTvg4l0YS5XPg5W2IH9WQexnsiM0N9NbzAv5qOfaOkamfaWYcyra7lNwQm2t5qw2rDIkYndoIMIsqm7IC/xogGnEqIEdWlX2NDWBEan6m40j1YbxSjvtJHDAUlcfTmIwncFbfFn/pTA6A/xTRrWb6pgaaVcwyHPgX3oC9xPo/FINdqgn/IIwDrLHMWyUdHIwmqssa9KHrOzk2mQqn8lWSdA6ozKrFyhFvC+xQX+EwEyg69xK1hC4XPkvw/NtmcRprx23rHJpQ1IaJQPk5oXd6ewYLWVyK/rnFJI36
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
X-OriginatorOrg: acksys.fr
X-MS-Exchange-CrossTenant-AuthAs: Internal
X-MS-Exchange-CrossTenant-AuthSource: DB9PR01MB7354.eurprd01.prod.exchangelabs.com
X-MS-Exchange-CrossTenant-Network-Message-Id: 6bfb4262-131e-46e3-352a-08d998898e22
X-MS-Exchange-CrossTenant-originalarrivaltime: 26 Oct 2021 14:04:41.0909
 (UTC)
X-MS-Exchange-CrossTenant-fromentityheader: Hosted
X-MS-Exchange-CrossTenant-id: f18a6414-d5f3-4b5c-9345-f30c01d87e32
X-MS-Exchange-CrossTenant-mailboxtype: HOSTED
X-MS-Exchange-CrossTenant-userprincipalname: GNBPQp7ahwpiUArHagRo0WteAP+YnfJXAts7QgyCh1BKZ/+7oENi6r08eHKXfbaJcwSFkJTVCYuKdUXIrGTe1A==
X-MS-Exchange-Transport-CrossTenantHeadersStamped: DB8PR01MB6390
Precedence: bulk
List-ID: <linux-wireless.vger.kernel.org>
X-Mailing-List: linux-wireless@vger.kernel.org

When used with ath10k, the following may happen:
a) radio card firmware receives ADDBA-REQ from peer
b) radio sends back ADDBA-RESP to peer and signals the ath10k driver
c) ath10k calls back ieee80211_manage_rx_ba_offl() in mac80211
   to signal rx ba offloading
d) mac80211::agg-rx.c::ieee80211_manage_rx_ba_offl()
  d1) sets a flag: sta->ampdu_mlme.tid_rx_manage_offl
  d2) queues a call to ht.c::ieee80211_ba_session_work()
e) ...scheduler runs...
f) ht.c::ieee80211_ba_session_work() checks the flag, clears it
   and sets up the rx ba session.

During (e), a fast peer may already have sent a BAREQ which is
propagated to rx.c::ieee80211_rx_h_ctrl(). Since the session is not
yet established, mac80211 sends back a DELBA to the peer, which can
hang the BA session.

The phenomenon can be observed between two QCA988X fw 10.2.4 radios,
using a loop of associate/arping from client to AP/disconnect. After
a few thousand loops, arping does not get a response and a sniffer
detects a DELBA action frame from the client, following an ADDBA.

Fix:
1) check the offload flag in addition to the check for a valid
   aggregation session
2) surround the checks with the existing dedicated mutex, to avoid
   interference from ieee80211_ba_session_work() during the check.

Note 1: there is another dubious DELBA generation in
ieee80211_rx_reorder_ampdu(), where the same kind of fix should fit,
but I did not fix it since I knew no easy way to test.

Note 2: this fix applies to wireless backports from 5.4-rc8.
---
Index: b/net/mac80211/rx.c
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
--- a/net/mac80211/rx.c
+++ b/net/mac80211/rx.c
@@ -2992,6 +2992,7 @@ ieee80211_rx_h_ctrl(struct ieee80211_rx_
 		return RX_CONTINUE;
=20
 	if (ieee80211_is_back_req(bar->frame_control)) {
+int ii =3D 99;
 		struct {
 			__le16 control, start_seq_num;
 		} __packed bar_data;
@@ -3008,11 +3009,19 @@ ieee80211_rx_h_ctrl(struct ieee80211_rx_
=20
 		tid =3D le16_to_cpu(bar_data.control) >> 12;
=20
+		mutex_lock(&rx->sta->ampdu_mlme.mtx);
 		if (!test_bit(tid, rx->sta->ampdu_mlme.agg_session_valid) &&
-		    !test_and_set_bit(tid, rx->sta->ampdu_mlme.unexpected_agg))
+		    /* back_req is allowed if the fw just received addba */
+		    !(ii=3Dtest_bit(tid, rx->sta->ampdu_mlme.tid_rx_manage_offl)) &&
+		    !test_and_set_bit(tid, rx->sta->ampdu_mlme.unexpected_agg)) {
+			mutex_unlock(&rx->sta->ampdu_mlme.mtx);
 			ieee80211_send_delba(rx->sdata, rx->sta->sta.addr, tid,
 					     WLAN_BACK_RECIPIENT,
 					     WLAN_REASON_QSTA_REQUIRE_SETUP);
+		} else {
+			mutex_unlock(&rx->sta->ampdu_mlme.mtx);
+if (ii !=3D 99) printk(KERN_ERR "JPT delba avoided, rx ba offload=3D%d", i=
i);
+		}
=20
 		tid_agg_rx =3D rcu_dereference(rx->sta->ampdu_mlme.tid_rx[tid]);
 		if (!tid_agg_rx)
--
quilt 0.63