[Devel] [PATCH v10 6/9] ve/cgroup: unmark ve-root cgroups at container stop

[Kirill Tkhai]

On 17.04.2020 15:14, Kirill Tkhai wrote:
> On 17.04.2020 15:11, Valeriy Vdovin wrote:
>>> Will the function work correctly in case jf we completely remove above "if"? Is this "if" optimization for init_css_set?
>>
>> This is indeed an optimization. And also it looks like it's the most common case.
> 
> (Please, avoid top-posting in your replies. Top-posting is not welcomed. See in Documentation/development-process/2.Process)
> 
> Then, we should add a comment about that this is optimization, I think.

With an appropriate comment about this is optimization, you may add my Reviewed-by: to this patch in next resend.

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>> *From:* Kirill Tkhai <ktkhai at virtuozzo.com>
>> *Sent:* Friday, April 17, 2020 12:15 PM
>> *To:* Valeriy Vdovin <Valeriy.Vdovin at virtuozzo.com>; devel at openvz.org <devel at openvz.org>; Konstantin Khorenko <khorenko at virtuozzo.com>; Pavel Tikhomirov <ptikhomirov at virtuozzo.com>
>> *Subject:* Re: [PATCH v10 6/9] ve/cgroup: unmark ve-root cgroups at container stop
>>  
>> On 16.04.2020 19:14, Valeriy Vdovin wrote:
>>> Signed-off-by: Valeriy Vdovin <valeriy.vdovin at virtuozzo.com>
>>> Reviewed-by: Kirill Tkhai <ktkhai at virtuozzo.com>
>>> ---
>>>   include/linux/cgroup.h |  1 +
>>>   kernel/cgroup.c        | 42 ++++++++++++++++++++++++++++++++++++++++++
>>>   kernel/ve/ve.c         |  2 ++
>>>   3 files changed, 45 insertions(+)
>>>
>>> diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
>>> index 0a42f93..1bd0fe7 100644
>>> --- a/include/linux/cgroup.h
>>> +++ b/include/linux/cgroup.h
>>> @@ -636,6 +636,7 @@ int cgroup_task_count(const struct cgroup *cgrp);
>>>   
>>>   #ifdef CONFIG_VE
>>>   void cgroup_mark_ve_roots(struct ve_struct *ve);
>>> +void cgroup_unmark_ve_roots(struct ve_struct *ve);
>>>   #endif
>>>   
>>>   /*
>>> diff --git a/kernel/cgroup.c b/kernel/cgroup.c
>>> index 60a4ca4..1dd36ed 100644
>>> --- a/kernel/cgroup.c
>>> +++ b/kernel/cgroup.c
>>> @@ -635,6 +635,35 @@ static struct css_set *find_css_set(
>>>   }
>>>   
>>>   /*
>>> + * Walk each cgroup link of a given css_set and find a cgroup that
>>> + * is the child of cgroupfs_root in argument.
>>> + */
>>> +static struct cgroup *css_cgroup_from_root(struct css_set *css_set,
>>> +                                         struct cgroupfs_root *root)
>>> +{
>>> +     struct cgroup *res = NULL;
>>> +
>>> +     BUG_ON(!mutex_is_locked(&cgroup_mutex));
>>> +     read_lock(&css_set_lock);
>>> +
>>> +     if (css_set == &init_css_set) {
>>> +             res = &root->top_cgroup;
>>
>> Will the function work correctly in case jf we completely remove above "if"?
>> Is this "if" optimization for init_css_set?
>>
>>> +     } else {
>>> +             struct cg_cgroup_link *link;
>>> +             list_for_each_entry(link, &css_set->cg_links, cg_link_list) {
>>> +                     struct cgroup *c = link->cgrp;
>>> +                     if (c->root == root) {
>>> +                             res = c;
>>> +                             break;
>>> +                     }
>>> +             }
>>> +     }
>>> +     read_unlock(&css_set_lock);
>>> +     BUG_ON(!res);
>>> +     return res;
>>> +}
>>> +
>>> +/*
>>>    * Return the cgroup for "task" from the given hierarchy. Must be
>>>    * called with cgroup_mutex held.
>>>    */
>>> @@ -4237,6 +4266,19 @@ void cgroup_mark_ve_roots(struct ve_struct *ve)
>>>         mutex_unlock(&cgroup_mutex);
>>>   }
>>>   
>>> +void cgroup_unmark_ve_roots(struct ve_struct *ve)
>>> +{
>>> +     struct cgroup *cgrp;
>>> +     struct cgroupfs_root *root;
>>> +
>>> +     mutex_lock(&cgroup_mutex);
>>> +     for_each_active_root(root) {
>>> +             cgrp = css_cgroup_from_root(ve->root_css_set, root);
>>> +             clear_bit(CGRP_VE_ROOT, &cgrp->flags);
>>> +     }
>>> +     mutex_unlock(&cgroup_mutex);
>>> +}
>>> +
>>>   struct cgroup *cgroup_get_ve_root(struct cgroup *cgrp)
>>>   {
>>>         struct cgroup *ve_root = NULL;
>>> diff --git a/kernel/ve/ve.c b/kernel/ve/ve.c
>>> index eedb85d..db3b600 100644
>>> --- a/kernel/ve/ve.c
>>> +++ b/kernel/ve/ve.c
>>> @@ -618,6 +618,8 @@ void ve_exit_ns(struct pid_namespace *pid_ns)
>>>         if (!ve->ve_ns || ve->ve_ns->pid_ns != pid_ns)
>>>                 return;
>>>   
>>> +     cgroup_unmark_ve_roots(ve);
>>> +
>>>         ve_workqueue_stop(ve);
>>>   
>>>         /*
>>>
>>
>