Re: [Drbd-dev] + exec-allow-core_pipe-recursion-check-to-look-for-a-value-of-1-rather -than-0.patch added to -mm tree

[Oleg Nesterov <oleg@redhat.com>]

On 01/26, Andrew Morton wrote:
>
> From: Neil Horman <nhorman@tuxdriver.com>
>
> What I've done is modify the
> call_usermodehelper() api such that an extra parameter is added, a
> function pointer which will be called by the user helper task, after it
> forks, but before it execs the required process.

Personally I agree, I think this fptr can be useful, not only for coredump.

> This will give the
> caller the opportunity to get a callback in the process's context,
> allowing it to do whatever it needs to to the process in the kernel

in this case it probably needs "void *data" argument, otherwise the
usage is very limited.

Currently only d_coredump() needs this new feature, but please note
that ____call_usermodehelper() was already "uglified" for the coredumping
over the pipe.

If we add sub_info->finit(), then probably we should move the code
under "if (sub_info->stdin)" from ____call_usermodehelper() to
core_pipe_setup() ?

> +/*
> + * This is used as a helper to set up the task that execs
> + * our user space core collector application
> + * Its called in the context of the task thats going to
> + * exec itself to be the helper, so we can modify current here
> + */

very minor nit, perhaps the comment should explain what is the meaning
of the magical rlim_cur = 1 value? It is not immediately obvious we
check cprm.limit == 1 below.

> +void core_pipe_setup(void)
> +{
> +	task_lock(current->group_leader);
> +	current->signal->rlim[RLIMIT_CORE].rlim_cur = 1;
> +	task_unlock(current->group_leader);
> +}

Well, this thread must be the kernel thread and thus it should be
->group_leader and I don't think we really need task_lock() her,
but this is minor and perhaps ->group_leader + task_lock() look
better even if not needed.

Oleg.