ipmi: Clear smi_info->thread to prevent use-after-free during module unload

During code inspection, I found an use-after-free possibility during unloading
ipmi_si in the polling mode.

If start_new_msg() is called after kthread_stop(), the function will try to
wake up non-existing kthread using the dangling pointer.

Possible scenario is when a new internal message is generated after
ipmi_unregister_smi()[*1] and remains after stop_timer_and_thread()
in clenaup_one_si() [*2].
Use-after-free could occur as follows depending on BMC replies.

  cleanup_one_si
    => ipmi_unregister_smi
       [*1]
    => stop_timer_and_thread
       => kthread_stop(smi_info->thread)
       [*2]
    => poll
       => smi_event_handler
          => start_new_msg
             => if (smi_info->thread)
                    wake_up_process(smi_info->thread) <== use-after-free!!

Although currently it seems no such message is generated in the polling mode,
some changes might introduce that in thefuture. For example in the interrupt
mode, disable_si_irq() does that at [*2].

So let's prevent such a critical issue possibility now.

Signed-off-by: Yamazaki Masamitsu <m-yamazaki@ah.jp.nec.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>

diff --git a/drivers/char/ipmi/ipmi_si_intf.c b/drivers/char/ipmi/ipmi_si_intf.c
index 7499b0cd8326f701a8fd136d70cd9f07c3fc3162..6768cb2dd740e692aeac2480ce8841be9e24ae2b 100644 (file)
--- a/drivers/char/ipmi/ipmi_si_intf.c
+++ b/drivers/char/ipmi/ipmi_si_intf.c
@@ -1938,8 +1938,10 @@ static void check_for_broken_irqs(struct smi_info *smi_info)
 
 static inline void stop_timer_and_thread(struct smi_info *smi_info)
 {
-       if (smi_info->thread != NULL)
+       if (smi_info->thread != NULL) {
                kthread_stop(smi_info->thread);
+               smi_info->thread = NULL;
+       }
 
        smi_info->timer_can_start = false;
        if (smi_info->timer_running)