Driver core: fix race in __device_release_driver

This patch (as1013) was suggested by David Woodhouse; it fixes a race
in the driver core.  If a device is unregistered at the same time as
its driver is unloaded, the driver's code pages may be unmapped while
the remove method is still running.  The calls to get_driver() and
put_driver() were intended to prevent this, but they don't work if the
driver's module count has already dropped to 0.

Instead, the patch keeps the device on the driver's list until after
the remove method has returned.  This forces the necessary
synchronization to occur.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

1 files changed, 2 insertions, 3 deletions

diff --git a/drivers/base/dd.c b/drivers/base/dd.c
index 54922647522d..b0726eb6405e 100644
--- a/drivers/base/dd.c
+++ b/drivers/base/dd.c
@@ -289,11 +289,10 @@ static void __device_release_driver(struct device * dev)
 {
         struct device_driver * drv;
 
-        drv = get_driver(dev->driver);
+        drv = dev->driver;
         if (drv) {
                 driver_sysfs_remove(dev);
                 sysfs_remove_link(&dev->kobj, "driver");
-                klist_remove(&dev->knode_driver);
 
                 if (dev->bus)
                         blocking_notifier_call_chain(&dev->bus->p->bus_notifier,
@@ -306,7 +305,7 @@ static void __device_release_driver(struct device * dev)
                         drv->remove(dev);
                 devres_release_all(dev);
                 dev->driver = NULL;
-                put_driver(drv);
+                klist_remove(&dev->knode_driver);
         }
 }