Re: [PATCH] ipmi: code cleanup and prevent potential issue.

[Corey Minyard <minyard@acm.org>]

On Tue, Jun 09, 2020 at 01:04:10AM -0500, wu000273@umn.edu wrote:
> From: Qiushi Wu <wu000273@umn.edu>
> 
> All the previous get/put operations against intf->refcount are
> inside the mutex. Thus, put the last kref_put() also inside mutex
> to make sure get/put functions execute in order and prevent the
> potential race condition.

No, this can result in a crash.  intf and intf->bmc_reg_mutex will
be freed by intf_free.  In fact, every call to kref_put() on intf
better be outside any mutex/lock in intf.  If you saw any, that
is a bug, please report that.  kref_get() is fine inside the
mutex.

Plus, this is not a race condition.  get/put is atomic.

-corey

> 
> Signed-off-by: Qiushi Wu <wu000273@umn.edu>
> ---
>  drivers/char/ipmi/ipmi_msghandler.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/char/ipmi/ipmi_msghandler.c b/drivers/char/ipmi/ipmi_msghandler.c
> index e1b22fe0916c..d34343e34272 100644
> --- a/drivers/char/ipmi/ipmi_msghandler.c
> +++ b/drivers/char/ipmi/ipmi_msghandler.c
> @@ -2583,10 +2583,11 @@ static int __bmc_get_device_id(struct ipmi_smi *intf, struct bmc_device *bmc,
>  			*guid =  bmc->guid;
>  	}
>  
> +	kref_put(&intf->refcount, intf_free);
> +
>  	mutex_unlock(&bmc->dyn_mutex);
>  	mutex_unlock(&intf->bmc_reg_mutex);
>  
> -	kref_put(&intf->refcount, intf_free);
>  	return rv;
>  }
>  
> -- 
> 2.17.1
>