Re: [PATCH] ftrace: Fix use-after-free for dynamic ftrace_ops

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 1 Nov 2022 14:41:46 +0800
Li Huafei <lihuafei1@huawei.com> wrote:


> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index fbf2543111c0..4219cc2a04a6 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -3030,13 +3030,16 @@ int ftrace_shutdown(struct ftrace_ops *ops, int command)
>  
>  	if (!command || !ftrace_enabled) {

ftrace_enabled is seldom not set. I don't think we even need to check
it. It's just the value of /proc/sys/kernel/ftrace_enabled, where most
people don't even know that file exists. I do want to get rid of it one
day too. So let's not optimize for it.

>  		/*
> -		 * If these are dynamic or per_cpu ops, they still
> -		 * need their data freed. Since, function tracing is
> -		 * not currently active, we can just free them
> -		 * without synchronizing all CPUs.
> +		 * If these are dynamic, they still need their data freed. If
> +		 * function tracing is currently active, we neet to synchronize
> +		 * all CPUs before we can release them.
>  		 */
> -		if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
> +		if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
> +			if (ftrace_enabled)
> +				goto sync_rcu;
> +
>  			goto free_ops;

Change the above just to "goto out;"

> +		}
>  
>  		return 0;
>  	}
> @@ -3083,6 +3086,7 @@ int ftrace_shutdown(struct ftrace_ops *ops, int command)
>  	 * ops.
>  	 */

Add here:

 out:

>  	if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
> + sync_rcu:
>  		/*
>  		 * We need to do a hard force of sched synchronization.
>  		 * This is because we use preempt_disable() to do RCU, but

And get rid of the labels in the if block.

Thanks!

-- Steve