Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

Am Dienstag, 19. Mai 2015, 15:22:27 schrieb Herbert Xu:

Hi Herbert,

On Tue, May 19, 2015 at 07:58:25AM +0200, Stephan Mueller wrote:

quoted

Herbert, do you have any ideas?

On the /dev/random side,

1) Add a struct module argument in addition to func/data.
2) Grab module ref count when func/data is registered.
3) Drop module ref count after func returns.

On the drbg side,

1) Allocate data pointer before func/data registration, it should
contain a flag indicating whether drbg is still alive.
2) In cra_exit, zap the flag in allocated data.
3) In func immediately return if flag indicates drbg is dead.
4) Free allocated data pointer when func is done.

Obviously you need to add some locking so that func doesn't race
against cra_exit or any other drbg paths that it intersects.

Thank you for the hints. I will follow your guidance.

Just for my edification: why is this (rather complex sounding) approach 
preferred over a simple cancel API? Other async APIs (e.g. the AIO syscalls 
with io_cancel) have such cancel operations.

Such cancel function would be as simple as:

void get_blocking_random_bytes_cancel(void *private)
{
        struct random_work *rw, *tmp;

        mutex_lock(&random_wait_list_mutex);
        list_for_each_entry_safe(rw, tmp, &random_wait_list, list) {
                if (private == rw->rw_private) {
                        list_del(&rw->list);
                        kfree(rw);
                        break;
                }
        }
        mutex_unlock(&random_wait_list_mutex);
}

-- 
Ciao
Stephan