[Devel] Re: [RFC][PATCH 2/2] Virtualization of IPC

[Eric W. Biederman]

Kirill Korotaev <dev at sw.ru> writes:

> This patch introduces IPC namespaces, which allow to create isolated IPC users
> or containers.
> Introduces CONFIG_IPC_NS and ipc_namespace structure.
> It also uses current->ipc_ns as a pointer to current namespace, which reduces
> places where additional argument to functions should be added.

I don't see where we are freeing the shared memory segments,
the message queues and the semaphores when the last user of the namespace
goes away.  Am I missing something?

> --- a/include/linux/ipc.h
> +++ b/include/linux/ipc.h
> @@ -70,6 +70,50 @@ struct kern_ipc_perm
>  
>  #endif /* __KERNEL__ */
>  
> +#include <linux/config.h>
> +
> +#ifdef CONFIG_IPC_NS
> +#include <asm/atomic.h>
> +
> +struct ipc_ids;
> +struct ipc_namespace {
> +	atomic_t cnt;
> +
> +	struct ipc_ids *sem_ids;
> +	int sem_ctls[4];
> +	int used_sems;
> +
> +	struct ipc_ids *msg_ids;
> +	int msg_ctlmax;
> +	int msg_ctlmnb;
> +	int msg_ctlmni;
> +
> +	struct ipc_ids *shm_ids;
> +	size_t	shm_ctlmax;
> +	size_t 	shm_ctlall;
> +	int 	shm_ctlmni;
> +	int	shm_total;
> +};

I believe there is a small problem with this implementation.
per namespace counts and limits are fine.  But I think we want
to maintain true global limits as well.   I know
concerns of that nature have been expressed in regards
to Daves patch.

> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -1193,6 +1193,7 @@ static task_t *copy_process(unsigned lon
>  	attach_pid(p, PIDTYPE_TGID, p->tgid);
>  	attach_pid(p, PIDTYPE_PID, p->pid);
>  	get_uts_ns(p->uts_ns);
> +	get_ipc_ns(p->ipc_ns);
>  
>  	nr_threads++;
>  	total_forks++;

Again please move the get outside of the tasklist_lock.

Eric