[PATCH? 3/3] move daemonized kernel threads into the swapper's session

[PATCH? 3/3] move daemonized kernel threads into the swapper's session

[Oleg Nesterov]

Any reason why daemonized kthread still uses 1,1 special pids? This patch
sets 0,0 pids, this matches kthread_create'ed threads.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>

--- PT/kernel/exit.c~3_daemonize_swapper	2007-11-26 16:47:35.000000000 +0300
+++ PT/kernel/exit.c	2007-11-26 17:10:41.000000000 +0300
@@ -390,7 +390,7 @@ void daemonize(const char *name, ...)
 		get_nsproxy(&init_nsproxy);
 		switch_task_namespaces(current, &init_nsproxy);
 	}
-	set_special_pids(find_pid(1));
+	set_special_pids(&init_struct_pid);
 	proc_clear_tty(current);
 
 	/* Block and flush all signals */

Re: [PATCH? 3/3] move daemonized kernel threads into the swapper's session

[Oleg Nesterov]

On 11/26, Oleg Nesterov wrote:
>
> Any reason why daemonized kthread still uses 1,1 special pids?

Actually, I guess the reason is that this was not possible before.
Now that set_special_pids uses "struct pid" we can do this.

Perhaps it is better to move the callsite to reparent_to_kthreadd()
and kill set_sepcial_pids().

Oleg.

Re: [PATCH? 3/3] move daemonized kernel threads into the swapper's session

[Eric W. Biederman]

Oleg Nesterov <oleg@tv-sign.ru> writes:

> Any reason why daemonized kthread still uses 1,1 special pids? This patch
> sets 0,0 pids, this matches kthread_create'ed threads.

You got it in your reply...

> Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>

Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>



Hmm.  I wonder if we could ensure kernel_thread always
returns a global pid_t, and then reallocate the struct
pid for the task we have just daemonized so that it only
has a single pid in the init_pid_ns.

That would remove our capturing of child pid namespaces.

Eric

Re: [PATCH? 3/3] move daemonized kernel threads into the swapper's session

[Eric W. Biederman]

ebiederm@xmission.com (Eric W. Biederman) writes:


> Hmm.  I wonder if we could ensure kernel_thread always
> returns a global pid_t, and then reallocate the struct
> pid for the task we have just daemonized so that it only
> has a single pid in the init_pid_ns.

Well we can't reallocate the struct pid, if anyone was using
it we would have problems, but we can certainly remove all
pid_t values except for those in the init_pid_ns.

We definitely have some serious issues with kernel_thread
and signal sending though.

Eric