net: usb: lan78xx: reorder cleanup operations to avoid UAF bugs
authorDuoming Zhou <duoming@zju.edu.cn>
Wed, 26 Jul 2023 08:14:07 +0000 (16:14 +0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 31 Jul 2023 08:58:29 +0000 (09:58 +0100)
The timer dev->stat_monitor can schedule the delayed work dev->wq and
the delayed work dev->wq can also arm the dev->stat_monitor timer.

When the device is detaching, the net_device will be deallocated. but
the net_device private data could still be dereferenced in delayed work
or timer handler. As a result, the UAF bugs will happen.

One racy situation is shown below:

      (Thread 1)                 |      (Thread 2)
lan78xx_stat_monitor()           |
 ...                             |  lan78xx_disconnect()
 lan78xx_defer_kevent()          |    ...
  ...                            |    cancel_delayed_work_sync(&dev->wq);
  schedule_delayed_work()        |    ...
  (wait some time)               |    free_netdev(net); //free net_device
  lan78xx_delayedwork()          |
  //use net_device private data  |
  dev-> //use                    |

Although we use cancel_delayed_work_sync() to cancel the delayed work
in lan78xx_disconnect(), it could still be scheduled in timer handler
lan78xx_stat_monitor().

Another racy situation is shown below:

      (Thread 1)                |      (Thread 2)
lan78xx_delayedwork             |
 mod_timer()                    |  lan78xx_disconnect()
                                |   cancel_delayed_work_sync()
 (wait some time)               |   if (timer_pending(&dev->stat_monitor))
                              |       del_timer_sync(&dev->stat_monitor);
 lan78xx_stat_monitor()         |   ...
  lan78xx_defer_kevent()        |   free_netdev(net); //free
   //use net_device private data|
   dev-> //use                  |

Although we use del_timer_sync() to delete the timer, the function
timer_pending() returns 0 when the timer is activated. As a result,
the del_timer_sync() will not be executed and the timer could be
re-armed.

In order to mitigate this bug, We use timer_shutdown_sync() to shutdown
the timer and then use cancel_delayed_work_sync() to cancel the delayed
work. As a result, the net_device could be deallocated safely.

What's more, the dev->flags is set to EVENT_DEV_DISCONNECT in
lan78xx_disconnect(). But it could still be set to EVENT_STAT_UPDATE
in lan78xx_stat_monitor(). So this patch put the set_bit() behind
timer_shutdown_sync().

Fixes: 77dfff5bb7e2 ("lan78xx: Fix race condition in disconnect handling")
Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/usb/lan78xx.c

index c458c03..59cde06 100644 (file)
@@ -4224,8 +4224,6 @@ static void lan78xx_disconnect(struct usb_interface *intf)
        if (!dev)
                return;
 
-       set_bit(EVENT_DEV_DISCONNECT, &dev->flags);
-
        netif_napi_del(&dev->napi);
 
        udev = interface_to_usbdev(intf);
@@ -4233,6 +4231,8 @@ static void lan78xx_disconnect(struct usb_interface *intf)
 
        unregister_netdev(net);
 
+       timer_shutdown_sync(&dev->stat_monitor);
+       set_bit(EVENT_DEV_DISCONNECT, &dev->flags);
        cancel_delayed_work_sync(&dev->wq);
 
        phydev = net->phydev;
@@ -4247,9 +4247,6 @@ static void lan78xx_disconnect(struct usb_interface *intf)
 
        usb_scuttle_anchored_urbs(&dev->deferred);
 
-       if (timer_pending(&dev->stat_monitor))
-               del_timer_sync(&dev->stat_monitor);
-
        lan78xx_unbind(dev, intf);
 
        lan78xx_free_tx_resources(dev);