[for-next][PATCH 1/5] ftrace: Protect ftrace_graph_hash with ftrace_sync

[for-next][PATCH 1/5] ftrace: Protect ftrace_graph_hash with ftrace_sync

[Steven Rostedt]

From: "Steven Rostedt (VMware)" <rostedt@goodmis.org>

As function_graph tracer can run when RCU is not "watching", it can not be
protected by synchronize_rcu() it requires running a task on each CPU before
it can be freed. Calling schedule_on_each_cpu(ftrace_sync) needs to be used.

Link: https://lore.kernel.org/r/20200205131110.GT2935@paulmck-ThinkPad-P72

Cc: stable@vger.kernel.org
Fixes: b9b0c831bed26 ("ftrace: Convert graph filter to use hash tables")
Reported-by: "Paul E. McKenney" <paulmck@kernel.org>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
---
 kernel/trace/ftrace.c | 11 +++++++++--
 kernel/trace/trace.h  |  2 ++
 2 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index 481ede3eac13..3f7ee102868a 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -5867,8 +5867,15 @@ ftrace_graph_release(struct inode *inode, struct file *file)
 
 		mutex_unlock(&graph_lock);
 
-		/* Wait till all users are no longer using the old hash */
-		synchronize_rcu();
+		/*
+		 * We need to do a hard force of sched synchronization.
+		 * This is because we use preempt_disable() to do RCU, but
+		 * the function tracers can be called where RCU is not watching
+		 * (like before user_exit()). We can not rely on the RCU
+		 * infrastructure to do the synchronization, thus we must do it
+		 * ourselves.
+		 */
+		schedule_on_each_cpu(ftrace_sync);
 
 		free_ftrace_hash(old_hash);
 	}
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 8c52f5de9384..3c75d29bd861 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -979,6 +979,7 @@ static inline int ftrace_graph_addr(struct ftrace_graph_ent *trace)
 	 * Have to open code "rcu_dereference_sched()" because the
 	 * function graph tracer can be called when RCU is not
 	 * "watching".
+	 * Protected with schedule_on_each_cpu(ftrace_sync)
 	 */
 	hash = rcu_dereference_protected(ftrace_graph_hash, !preemptible());
 
@@ -1031,6 +1032,7 @@ static inline int ftrace_graph_notrace_addr(unsigned long addr)
 	 * Have to open code "rcu_dereference_sched()" because the
 	 * function graph tracer can be called when RCU is not
 	 * "watching".
+	 * Protected with schedule_on_each_cpu(ftrace_sync)
 	 */
 	notrace_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
 						 !preemptible());
-- 
2.24.1

Re: [for-next][PATCH 1/5] ftrace: Protect ftrace_graph_hash with ftrace_sync

[Paul E. McKenney]

On Wed, Feb 05, 2020 at 05:21:11PM -0500, Steven Rostedt wrote:
> From: "Steven Rostedt (VMware)" <rostedt@goodmis.org>
> 
> As function_graph tracer can run when RCU is not "watching", it can not be
> protected by synchronize_rcu() it requires running a task on each CPU before
> it can be freed. Calling schedule_on_each_cpu(ftrace_sync) needs to be used.
> 
> Link: https://lore.kernel.org/r/20200205131110.GT2935@paulmck-ThinkPad-P72
> 
> Cc: stable@vger.kernel.org
> Fixes: b9b0c831bed26 ("ftrace: Convert graph filter to use hash tables")
> Reported-by: "Paul E. McKenney" <paulmck@kernel.org>
> Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
> Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

Nice!  If there is much more call for this, perhaps I should take a hint
from the ftrace_sync() comment and add synchronize_rcu_rude().  ;-)

Reviewed-by: "Paul E. McKenney" <paulmck@kernel.org>

> ---
>  kernel/trace/ftrace.c | 11 +++++++++--
>  kernel/trace/trace.h  |  2 ++
>  2 files changed, 11 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index 481ede3eac13..3f7ee102868a 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -5867,8 +5867,15 @@ ftrace_graph_release(struct inode *inode, struct file *file)
>  
>  		mutex_unlock(&graph_lock);
>  
> -		/* Wait till all users are no longer using the old hash */
> -		synchronize_rcu();
> +		/*
> +		 * We need to do a hard force of sched synchronization.
> +		 * This is because we use preempt_disable() to do RCU, but
> +		 * the function tracers can be called where RCU is not watching
> +		 * (like before user_exit()). We can not rely on the RCU
> +		 * infrastructure to do the synchronization, thus we must do it
> +		 * ourselves.
> +		 */
> +		schedule_on_each_cpu(ftrace_sync);
>  
>  		free_ftrace_hash(old_hash);
>  	}
> diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
> index 8c52f5de9384..3c75d29bd861 100644
> --- a/kernel/trace/trace.h
> +++ b/kernel/trace/trace.h
> @@ -979,6 +979,7 @@ static inline int ftrace_graph_addr(struct ftrace_graph_ent *trace)
>  	 * Have to open code "rcu_dereference_sched()" because the
>  	 * function graph tracer can be called when RCU is not
>  	 * "watching".
> +	 * Protected with schedule_on_each_cpu(ftrace_sync)
>  	 */
>  	hash = rcu_dereference_protected(ftrace_graph_hash, !preemptible());
>  
> @@ -1031,6 +1032,7 @@ static inline int ftrace_graph_notrace_addr(unsigned long addr)
>  	 * Have to open code "rcu_dereference_sched()" because the
>  	 * function graph tracer can be called when RCU is not
>  	 * "watching".
> +	 * Protected with schedule_on_each_cpu(ftrace_sync)
>  	 */
>  	notrace_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
>  						 !preemptible());
> -- 
> 2.24.1
> 
>