This patch (as1607) fixes a race that can occur if a USB host
controller is removed while a process is reading the
/sys/kernel/debug/usb/devices file.
The usb_device_read() routine uses the bus->root_hub pointer to
determine whether or not the root hub is registered. The is not a
valid test, because the pointer is set before the root hub gets
registered and remains set even after the root hub is unregistered and
deallocated. As a result, usb_device_read() or usb_device_dump() can
access freed memory, causing an oops.
The patch changes the test to use the hcd->rh_registered flag, which
does get set and cleared at the appropriate times. It also makes sure
to hold the usb_bus_list_lock mutex while setting the flag, so that
usb_device_read() will become aware of new root hubs as soon as they
are registered.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: Don Zickus <dzickus@redhat.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
/* print devices for all busses */
list_for_each_entry(bus, &usb_bus_list, bus_list) {
/* recurse through all children of the root hub */
- if (!bus->root_hub)
+ if (!bus_to_hcd(bus)->rh_registered)
continue;
usb_lock_device(bus->root_hub);
ret = usb_device_dump(&buf, &nbytes, &skip_bytes, ppos,
if (retval) {
dev_err (parent_dev, "can't register root hub for %s, %d\n",
dev_name(&usb_dev->dev), retval);
- }
- mutex_unlock(&usb_bus_list_lock);
-
- if (retval == 0) {
+ } else {
spin_lock_irq (&hcd_root_hub_lock);
hcd->rh_registered = 1;
spin_unlock_irq (&hcd_root_hub_lock);
if (HCD_DEAD(hcd))
usb_hc_died (hcd); /* This time clean up */
}
+ mutex_unlock(&usb_bus_list_lock);
return retval;
}