Re: [PATCH v2] kernel: add pid_max to pid_namespace

[Joel Granados <joel.granados@kernel.org>]

On Tue, Nov 05, 2024 at 11:10:24AM +0800, Yun Zhou wrote:
> It is necessary to have a different pid_max in different containers.
> For example, multiple containers are running on a host, one of which
> is Android, and its 32 bit bionic libc only accepts pid <= 65535. So
> it requires the global pid_max <= 65535. This will cause configuration
> conflicts with other containers and also limit the maximum number of
> tasks for the entire system.
> 
> Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
> ---
>  - Remove sentinels from ctl_table arrays.
> v1 - https://lore.kernel.org/all/20241030052933.1041408-1-yun.zhou@windriver.com/
> ---
>  include/linux/pid_namespace.h     |  1 +
>  kernel/pid.c                      | 12 +++++------
>  kernel/pid_namespace.c            | 34 ++++++++++++++++++++++++++-----
>  kernel/sysctl.c                   |  9 --------
>  kernel/trace/pid_list.c           |  2 +-
>  kernel/trace/trace.h              |  2 --
>  kernel/trace/trace_sched_switch.c |  2 +-
>  7 files changed, 38 insertions(+), 24 deletions(-)

...

> diff --git a/kernel/pid_namespace.c b/kernel/pid_namespace.c
> index d70ab49d5b4a..a5a8254825d5 100644
> --- a/kernel/pid_namespace.c
> +++ b/kernel/pid_namespace.c
> @@ -111,6 +111,7 @@ static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns
>  	ns->user_ns = get_user_ns(user_ns);
>  	ns->ucounts = ucounts;
>  	ns->pid_allocated = PIDNS_ADDING;
> +	ns->pid_max = parent_pid_ns->pid_max;
>  #if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
>  	ns->memfd_noexec_scope = pidns_memfd_noexec_scope(parent_pid_ns);
>  #endif
> @@ -280,19 +281,44 @@ static int pid_ns_ctl_handler(const struct ctl_table *table, int write,
>  
>  	return ret;
>  }
> +#endif	/* CONFIG_CHECKPOINT_RESTORE */
> +
> +static int pid_max_ns_ctl_handler(const struct ctl_table *table, int write,
> +		void *buffer, size_t *lenp, loff_t *ppos)
> +{
> +	struct pid_namespace *pid_ns = task_active_pid_ns(current);
> +	struct ctl_table tmp = *table;
> +
> +	if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
> +		return -EPERM;
> +
> +	tmp.data = &pid_ns->pid_max;
> +	if (pid_ns->parent)
> +		tmp.extra2 = &pid_ns->parent->pid_max;
> +
> +	return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
> +}
>  
> -extern int pid_max;
>  static struct ctl_table pid_ns_ctl_table[] = {
> +#ifdef CONFIG_CHECKPOINT_RESTORE
>  	{
>  		.procname = "ns_last_pid",
>  		.maxlen = sizeof(int),
>  		.mode = 0666, /* permissions are checked in the handler */
>  		.proc_handler = pid_ns_ctl_handler,
>  		.extra1 = SYSCTL_ZERO,
> -		.extra2 = &pid_max,
> +		.extra2 = &init_pid_ns.pid_max,
>  	},
> -};
>  #endif	/* CONFIG_CHECKPOINT_RESTORE */
> +	{
> +		.procname = "pid_max",
> +		.maxlen = sizeof(int),
> +		.mode = 0644,
> +		.proc_handler = pid_max_ns_ctl_handler,
> +		.extra1 = &pid_max_min,
> +		.extra2 = &pid_max_max,
> +	},
> +};

I see here that the sysctls are without sentinel.
Reviewed-by: Joel Granados <joel.granados@kernel.org>


-- 

Joel Granados