[Devel] [PATCH RHEL7 v19 07/13] ve/cgroup: Added ve_owner field to cgroup

[Valeriy Vdovin]

Each cgroup representing a host or a container root of
cgroup subsystem hierarhy will have this field set to
a valid ve_struct, that owns this root. This way each
cgroup in a system will be able to know it's owning VE.
Non root cgroups will have this field set to NULL, this
is an optimization for cleanup code: at VE destruction
we only need to iterate over all root cgroups to clean
reference to former owning VE, rather than over all
cgroup hierarchy.
Still any cgroup that wants to know about it's owning
VE can find it's virtual root cgroup and read it's
ve_owner field.

cgroup->ve_owner is declared as RCU pointer, because it fits
RCU semantics - rare writes/often reads. ve_owner will be
read from multiple locations in code in further patches and
is only rarely set at cgroup_mark_ve_root/cgroup_mount.
cgroup_get_ve_owner is a read wrapper for this purpose.

Signed-off-by: Valeriy Vdovin <valeriy.vdovin at virtuozzo.com>
---
 include/linux/cgroup.h |  4 ++++
 kernel/cgroup.c        | 45 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 49 insertions(+)

diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
index 1bd0fe7..ad437a8 100644
--- a/include/linux/cgroup.h
+++ b/include/linux/cgroup.h
@@ -292,6 +292,9 @@ struct cgroup {
 	/* directory xattrs */
 	struct simple_xattrs xattrs;
 	u64 subgroups_limit;
+
+	/* ve_owner, responsible for running release agent. */
+	struct ve_struct __rcu *ve_owner;
 };
 
 #define MAX_CGROUP_ROOT_NAMELEN 64
@@ -637,6 +640,7 @@ 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);
+struct ve_struct *cgroup_get_ve_owner(struct cgroup *cgrp);
 #endif
 
 /*
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 010a23e..c4e0126 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -805,6 +805,42 @@ static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
 	return name;
 }
 
+struct cgroup *cgroup_get_local_root(struct cgroup *cgrp)
+{
+	/*
+	 * Find nearest root cgroup, which might be host cgroup root
+	 * or ve cgroup root.
+	 *
+	 *    <host_root_cgroup> -> local_root
+	 *     \                    ^
+	 *      <cgroup>            |
+	 *       \                  |
+	 *        <cgroup>   --->   from here
+	 *        \
+	 *         <ve_root_cgroup> -> local_root
+	 *         \                   ^
+	 *          <cgroup>           |
+	 *          \                  |
+	 *           <cgroup>  --->    from here
+	 */
+	while (cgrp->parent && !test_bit(CGRP_VE_ROOT, &cgrp->flags))
+		cgrp = cgrp->parent;
+
+	return cgrp;
+}
+
+struct ve_struct *cgroup_get_ve_owner(struct cgroup *cgrp)
+{
+	struct ve_struct *ve;
+	/* Caller should hold RCU */
+
+	cgrp = cgroup_get_local_root(cgrp);
+	ve = rcu_dereference(cgrp->ve_owner);
+	if (!ve)
+		ve = get_ve0();
+	return ve;
+}
+
 static void cgroup_free_fn(struct work_struct *work)
 {
 	struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
@@ -1627,6 +1663,8 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type,
 		ret = PTR_ERR(new_root);
 		goto drop_modules;
 	}
+
+	RCU_INIT_POINTER(new_root->top_cgroup.ve_owner, &ve0);
 	opts.new_root = new_root;
 
 	/* Locate an existing or new sb for this hierarchy */
@@ -4254,12 +4292,14 @@ void cgroup_mark_ve_roots(struct ve_struct *ve)
 	mutex_lock(&cgroup_mutex);
 	for_each_active_root(root) {
 		cgrp = task_cgroup_from_root(ve->init_task, root);
+		rcu_assign_pointer(cgrp->ve_owner, ve);
 		set_bit(CGRP_VE_ROOT, &cgrp->flags);
 
 		if (test_bit(cpu_cgroup_subsys_id, &root->subsys_mask))
 			link_ve_root_cpu_cgroup(cgrp);
 	}
 	mutex_unlock(&cgroup_mutex);
+	synchronize_rcu();
 }
 
 void cgroup_unmark_ve_roots(struct ve_struct *ve)
@@ -4270,9 +4310,14 @@ void cgroup_unmark_ve_roots(struct ve_struct *ve)
 	mutex_lock(&cgroup_mutex);
 	for_each_active_root(root) {
 		cgrp = css_cgroup_from_root(ve->root_css_set, root);
+		BUG_ON(!rcu_dereference_protected(cgrp->ve_owner,
+				lockdep_is_held(&cgroup_mutex)));
+		rcu_assign_pointer(cgrp->ve_owner, NULL);
 		clear_bit(CGRP_VE_ROOT, &cgrp->flags);
 	}
 	mutex_unlock(&cgroup_mutex);
+	/* ve_owner == NULL will be visible */
+	synchronize_rcu();
 }
 
 struct cgroup *cgroup_get_ve_root(struct cgroup *cgrp)
-- 
1.8.3.1