Re: [PATCHv5 01/37] ns: Introduce Time Namespace

From: Dmitry Safonov
Date: Thu Aug 01 2019 - 19:46:48 EST


Hi Andy,

Thank you for the review,

On 8/1/19 6:29 AM, Andy Lutomirski wrote:
> On Mon, Jul 29, 2019 at 2:58 PM Dmitry Safonov <dima@xxxxxxxxxx> wrote:
>>
>> From: Andrei Vagin <avagin@xxxxxxxxxx>
>>
>> Time Namespace isolates clock values.
>
>> +static int timens_install(struct nsproxy *nsproxy, struct ns_common *new)
>> +{
>> + struct time_namespace *ns = to_time_ns(new);
>> +
>> + if (!thread_group_empty(current))
>> + return -EINVAL;
>
> You also need to check for other users of the mm.

Oops. It seems like, if the check was

if (!current_is_single_threaded())
return -EUSERS;

instead of thread_group_empty(current), it would address the concerns
from 23/37 and 25/37 patches, too?

>
>> +
>> + if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
>> + !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
>> + return -EPERM;
>> +
>> + get_time_ns(ns);
>> + get_time_ns(ns);
>> + put_time_ns(nsproxy->time_ns);
>> + put_time_ns(nsproxy->time_ns_for_children);
>> + nsproxy->time_ns = ns;
>> + nsproxy->time_ns_for_children = ns;
>> + ns->initialized = true;
>
> I really really wish that setns() took an explicit flag for "change
> now" or "change for children", since the semantics are different. Oh
> well.
>

Thanks,
Dmitry