On 7/11/24 8:56 AM, Roman Gushchin wrote:
> On Wed, Jul 10, 2024 at 10:51:53PM -0400, Waiman Long wrote:
>> Cgroup subsystem state (CSS) is an abstraction in the cgroup layer to
>> help manage different structures in various cgroup subsystems by being
>> an embedded element inside a larger structure like cpuset or mem_cgroup.
>>
>> The /proc/cgroups file shows the number of cgroups for each of the
>> subsystems. With cgroup v1, the number of CSSes is the same as the
>> number of cgroups. That is not the case anymore with cgroup v2. The
>> /proc/cgroups file cannot show the actual number of CSSes for the
>> subsystems that are bound to cgroup v2.
>>
>> So if a v2 cgroup subsystem is leaking cgroups (usually memory cgroup),
>> we can't tell by looking at /proc/cgroups which cgroup subsystems may
>> be responsible.
>>
>> As cgroup v2 had deprecated the use of /proc/cgroups, the hierarchical
>> cgroup.stat file is now being extended to show the number of live and
>> dying CSSes associated with all the non-inhibited cgroup subsystems
>> that have been bound to cgroup v2 as long as it is not zero. The number
>> includes CSSes in the current cgroup as well as in all the descendants
>> underneath it. This will help us pinpoint which subsystems are
>> responsible for the increasing number of dying (nr_dying_descendants)
>> cgroups.
>>
>> The cgroup-v2.rst file is updated to discuss this new behavior.
>>
>> With this patch applied, a sample output from root cgroup.stat file
>> was shown below.
>>
>> nr_descendants 55
>> nr_dying_descendants 35
>> nr_subsys_cpuset 1
>> nr_subsys_cpu 40
>> nr_subsys_io 40
>> nr_subsys_memory 55
>> nr_dying_subsys_memory 35
>> nr_subsys_perf_event 56
>> nr_subsys_hugetlb 1
>> nr_subsys_pids 55
>> nr_subsys_rdma 1
>> nr_subsys_misc 1
>>
>> Another sample output from system.slice/cgroup.stat was:
>>
>> nr_descendants 32
>> nr_dying_descendants 33
>> nr_subsys_cpu 30
>> nr_subsys_io 30
>> nr_subsys_memory 32
>> nr_dying_subsys_memory 33
>> nr_subsys_perf_event 33
>> nr_subsys_pids 32
>>
>> Signed-off-by: Waiman Long <longman@xxxxxxxxxx>
>> ---
>> Documentation/admin-guide/cgroup-v2.rst | 14 ++++++-
>> include/linux/cgroup-defs.h | 7 ++++
>> kernel/cgroup/cgroup.c | 52 ++++++++++++++++++++++++-
>> 3 files changed, 70 insertions(+), 3 deletions(-)
>>
>> diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
>> index 52763d6b2919..356cd430c888 100644
>> --- a/Documentation/admin-guide/cgroup-v2.rst
>> +++ b/Documentation/admin-guide/cgroup-v2.rst
>> @@ -981,6 +981,16 @@ All cgroup core files are prefixed with "cgroup."
>> A dying cgroup can consume system resources not exceeding
>> limits, which were active at the moment of cgroup deletion.
>>
>> + nr_subsys_<cgroup_subsys>
>> + Total number of live cgroups associated with that cgroup
>> + subsystem (e.g. memory) at and beneath the current
>> + cgroup. An entry will only be shown if it is not zero.
>> +
>> + nr_dying_subsys_<cgroup_subsys>
>> + Total number of dying cgroups associated with that cgroup
>> + subsystem (e.g. memory) beneath the current cgroup.
>> + An entry will only be shown if it is not zero.
>> +
>> cgroup.freeze
>> A read-write single value file which exists on non-root cgroups.
>> Allowed values are "0" and "1". The default is "0".
>> @@ -2930,8 +2940,8 @@ Deprecated v1 Core Features
>>
>> - "cgroup.clone_children" is removed.
>>
>> -- /proc/cgroups is meaningless for v2. Use "cgroup.controllers" file
>> - at the root instead.
>> +- /proc/cgroups is meaningless for v2. Use "cgroup.controllers" or
>> + "cgroup.stat" files at the root instead.
>>
>>
>> Issues with v1 and Rationales for v2
>> diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h
>> index b36690ca0d3f..62de18874508 100644
>> --- a/include/linux/cgroup-defs.h
>> +++ b/include/linux/cgroup-defs.h
>> @@ -210,6 +210,13 @@ struct cgroup_subsys_state {
>> * fields of the containing structure.
>> */
>> struct cgroup_subsys_state *parent;
>> +
>> + /*
>> + * Keep track of total numbers of visible and dying descendant CSSes.
>> + * Protected by cgroup_mutex.
>> + */
>> + int nr_descendants;
>> + int nr_dying_descendants;
>> };
>>
>> /*
>> diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
>> index c8e4b62b436a..cf4fc1c109e2 100644
>> --- a/kernel/cgroup/cgroup.c
>> +++ b/kernel/cgroup/cgroup.c
>> @@ -3669,12 +3669,36 @@ static int cgroup_events_show(struct seq_file *seq, void *v)
>> static int cgroup_stat_show(struct seq_file *seq, void *v)
>> {
>> struct cgroup *cgroup = seq_css(seq)->cgroup;
>> + struct cgroup_subsys_state *css;
>> + int ssid;
>>
>> seq_printf(seq, "nr_descendants %d\n",
>> cgroup->nr_descendants);
>> seq_printf(seq, "nr_dying_descendants %d\n",
>> cgroup->nr_dying_descendants);
>>
>> + /*
>> + * Show the number of live and dying csses associated with each of
>> + * non-inhibited cgroup subsystems bound to cgroup v2 if non-zero.
>> + *
>> + * Without proper lock protection, racing is possible. So the
>> + * numbers may not be consistent when that happens.
>> + */
>> + rcu_read_lock();
>> + for_each_css(css, ssid, cgroup) {
>> + if ((BIT(ssid) & cgrp_dfl_inhibit_ss_mask) ||
>> + (cgroup_subsys[ssid]->root != &cgrp_dfl_root))
>> + continue;
>> +
>> + seq_printf(seq, "nr_subsys_%s %d\n", cgroup_subsys[ssid]->name,
>> + css->nr_descendants + 1);
>> + /* Current css is online */
>> + if (css->nr_dying_descendants)
>> + seq_printf(seq, "nr_dying_subsys_%s %d\n",
>> + cgroup_subsys[ssid]->name,
>> + css->nr_dying_descendants);
>> + }
>> + rcu_read_unlock();
>> return 0;
>> }
>>
>> @@ -5424,6 +5448,8 @@ static void css_release_work_fn(struct work_struct *work)
>> list_del_rcu(&css->sibling);
>>
>> if (ss) {
>> + struct cgroup_subsys_state *parent_css;
>> +
>> /* css release path */
>> if (!list_empty(&css->rstat_css_node)) {
>> cgroup_rstat_flush(cgrp);
>> @@ -5433,6 +5459,14 @@ static void css_release_work_fn(struct work_struct *work)
>> cgroup_idr_replace(&ss->css_idr, NULL, css->id);
>> if (ss->css_released)
>> ss->css_released(css);
>> +
>> + WARN_ON_ONCE(css->nr_descendants || css->nr_dying_descendants);
>> + parent_css = css->parent;
>> + while (parent_css) {
>> + parent_css->nr_dying_descendants--;
>> + parent_css = parent_css->parent;
>> + }
>> + css_put(css->parent); /* Parent can be freed now */
>> } else {
>> struct cgroup *tcgrp;
>>
>> @@ -5517,8 +5551,11 @@ static int online_css(struct cgroup_subsys_state *css)
>> rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
>>
>> atomic_inc(&css->online_cnt);
>> - if (css->parent)
>> + if (css->parent) {
>> atomic_inc(&css->parent->online_cnt);
>> + while ((css = css->parent))
>> + css->nr_descendants++;
>> + }
>> }
>> return ret;
>> }
>> @@ -5540,6 +5577,19 @@ static void offline_css(struct cgroup_subsys_state *css)
>> RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
>>
>> wake_up_all(&css->cgroup->offline_waitq);
>> +
>> + /*
>> + * Get a reference to parent css to ensure reliable access to its
>> + * nr_dying_descendants until after this child css is ready to be
>> + * freed.
>> + */
>> + if (css->parent)
>> + css_get(css->parent);
>
> I think this blob can be dropped: css has a reference to their parent up to
> very last moment, see css_free_rwork_fn().
> And the corresponding css_put() can be dropped too.
>
> With this thing fixed, please, feel free to add
> Acked-by: Roman Gushchin <roman.gushchin@xxxxxxxxx> .
>
The nr_subsys_<subsy>, nr_dying_subsys_<subsys> format is more appealing.
I tested it with Roman's suggestion, with a simple test of creating,
enabling cpu, memory controllers, and removing 1000 cgroups as descendants.
While parallelly read /sys/fs/cgroup/cgroup.stat as root and non-root user
over a loop. I did run a few test runs and saw no issues.
--
Thanks,
Kamalesh
