Re: [PATCH 1/2] Take rcu read lock when iterating through threadgroup

[Oleg Nesterov]

On 03/01, Siddhesh Poyarekar wrote:
>
> Protect the iteration through thread group with rcu_read_lock when
> looking for tasks in the group that use the current vma as
> stack. Thanks KOSAKI Motohiro (kosaki.motohiro@xxxxxxxxx) for pointing
> it out.
>
> Signed-off-by: Siddhesh Poyarekar <siddhesh.poyarekar@xxxxxxxxx>
> ---
>  mm/memory.c |   15 +++++++++++----
>  1 files changed, 11 insertions(+), 4 deletions(-)
> 
> diff --git a/mm/memory.c b/mm/memory.c
> index 601a920..a88b764 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -3915,20 +3915,27 @@ void print_vma_addr(char *prefix, unsigned long ip)
>   * just check in the current task.
>   */
>  int vm_is_stack(struct task_struct *task,
> -			      struct vm_area_struct *vma, int in_group)
> +		struct vm_area_struct *vma, int in_group)
>  {
> +	int ret = 0;
> +
>  	if (vm_is_stack_for_task(task, vma))
>  		return 1;
>  
>  	if (in_group) {
>  		struct task_struct *t = task;
> +		rcu_read_lock();
>  		while_each_thread(task, t) {

This is the commont mistake. rcu_read_lock() can not help unless you
verify that ->thread_group.next still points to the rcu-protected
memory.

Just suppose that this task exits, then next_thread() exits too. Now
you take rcu_read_lock() but it is too late, ->next points to nowhere.

Also. In fact while_each_thread() is not safe under rcu. We are going
to fix this, but only for the case when while_each_thread() starts at
the thread group leader.

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/