[PATCH] wifi: rtl8xxxu: fix use-after-free from rx_urb_wq on stop

[Fan Wu]

rtl8xxxu arms rx_urb_wq from the RX completion path:
rtl8xxxu_rx_complete() hands the URB to rtl8xxxu_queue_rx_urb(), which
queues it on rx_urb_pending_list and, once the list grows past
RTL8XXXU_RX_URB_PENDING_WATER, schedules rx_urb_wq.  The worker
rtl8xxxu_rx_urb_work() drains rx_urb_pending_list, recovers priv through
container_of, and resubmits each URB through rtl8xxxu_submit_rx_urb(),
which anchors it on rx_anchor and dereferences priv->udev.

rtl8xxxu_stop() cancels the sibling work items (c2hcmd_work, ra_watchdog,
update_beacon_work) but never cancels rx_urb_wq, so a worker armed during
the last burst of RX traffic can run rtl8xxxu_rx_urb_work() after
rtl8xxxu_disconnect() has called ieee80211_free_hw(), which frees priv,
producing a use-after-free.  The window opens under active RX traffic
(pending count above the watermark) followed by a disconnect.

There are two teardown races to close:

  * rtl8xxxu_queue_rx_urb() decided whether to enqueue under rx_urb_lock
    but called schedule_work() after dropping the lock.  A completion
    that observed shutdown == false and released the lock could then call
    schedule_work() after rtl8xxxu_stop() had set shutdown and
    cancel_work_sync() had already returned, arming the worker to run
    after the teardown.  Move schedule_work() under the same !shutdown
    branch so the arming decision is atomic with the shutdown check.

  * rtl8xxxu_rx_urb_work() anchors every URB it drained back onto
    rx_anchor through rtl8xxxu_submit_rx_urb().  A worker still running
    when usb_kill_anchored_urbs(&priv->rx_anchor) returned would submit a
    URB that escaped the kill.  In rtl8xxxu_stop(), call
    cancel_work_sync(&priv->rx_urb_wq) before the kill so the worker is
    drained first.

After priv->shutdown is set under rx_urb_lock, completions can no longer
queue rx_urb_wq. cancel_work_sync() then drains the last queued or running
worker, and the following usb_kill_anchored_urbs() kills the URBs it may
have submitted.

rtl8xxxu_disconnect() is covered because ieee80211_unregister_hw()
guarantees .stop() runs for a live interface before ieee80211_free_hw()
frees priv.  The probe error path needs no cancel: rx_urb_wq is
INIT_WORK()'d there but cannot have been scheduled, since no URB is
submitted before ieee80211_register_hw() succeeds.

This bug was found by static analysis.

Fixes: 26f1fad29ad9 ("New driver: rtl8xxxu (mac80211)")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Fan Wu <fanwu01@xxxxxxxxxx>
---
 drivers/net/wireless/realtek/rtl8xxxu/core.c | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/drivers/net/wireless/realtek/rtl8xxxu/core.c b/drivers/net/wireless/realtek/rtl8xxxu/core.c
index c06ad064f37c..b447ce78ff05 100644
--- a/drivers/net/wireless/realtek/rtl8xxxu/core.c
+++ b/drivers/net/wireless/realtek/rtl8xxxu/core.c
@@ -5792,14 +5792,19 @@ static void rtl8xxxu_queue_rx_urb(struct rtl8xxxu_priv *priv,
 {
 	struct sk_buff *skb;
 	unsigned long flags;
-	int pending = 0;
 
 	spin_lock_irqsave(&priv->rx_urb_lock, flags);
 
 	if (!priv->shutdown) {
 		list_add_tail(&rx_urb->list, &priv->rx_urb_pending_list);
 		priv->rx_urb_pending_count++;
-		pending = priv->rx_urb_pending_count;
+		/*
+		 * Arm the worker under rx_urb_lock so this is atomic with the
+		 * shutdown check: moving it out of the lock would let a
+		 * completion arm the work after rtl8xxxu_stop() canceled it.
+		 */
+		if (priv->rx_urb_pending_count > RTL8XXXU_RX_URB_PENDING_WATER)
+			schedule_work(&priv->rx_urb_wq);
 	} else {
 		skb = (struct sk_buff *)rx_urb->urb.context;
 		dev_kfree_skb_irq(skb);
@@ -5807,9 +5812,6 @@ static void rtl8xxxu_queue_rx_urb(struct rtl8xxxu_priv *priv,
 	}
 
 	spin_unlock_irqrestore(&priv->rx_urb_lock, flags);
-
-	if (pending > RTL8XXXU_RX_URB_PENDING_WATER)
-		schedule_work(&priv->rx_urb_wq);
 }
 
 static void rtl8xxxu_rx_urb_work(struct work_struct *work)
@@ -7461,6 +7463,13 @@ static void rtl8xxxu_stop(struct ieee80211_hw *hw, bool suspend)
 	priv->shutdown = true;
 	spin_unlock_irqrestore(&priv->rx_urb_lock, flags);
 
+	/*
+	 * Cancel before killing rx_anchor: the worker re-anchors every URB
+	 * it drained via rtl8xxxu_submit_rx_urb(), so a worker still running
+	 * after the kill could submit a URB that escapes it.
+	 */
+	cancel_work_sync(&priv->rx_urb_wq);
+
 	usb_kill_anchored_urbs(&priv->rx_anchor);
 	usb_kill_anchored_urbs(&priv->tx_anchor);
 	if (priv->usb_interrupts)
-- 
2.34.1