[PATCH v2 0/2] cpuset: Allow v2 behavior in v1 cgroup

[PATCH v2 0/2] cpuset: Allow v2 behavior in v1 cgroup

[Waiman Long]

v1->v2:
 - Drop the kernel command line option and use cgroupfs mount option
   instead to enable v2 controller behavior in v1 cgroup.

v1: https://lkml.org/lkml/2017/8/15/570

This patchset adds a new cgroupfs mount option "v2_mode" that serves
as a hint that a v1 controller can use v2 hehavior, if applicable.
The cpuset controller is then modified to observe this mode.

Patch 1 adds the v1 cgroupfs mount option "v2_mode" which set a flag
in the cgroup_root flags.

Patch 2 modifies the cpuset controller to check the cgroup_root flag
and switch to v2 behavior even in a v1 cgroup.

Waiman Long (2):
  cgroup: Add a hint to allow v1 controller to use v2 behavior
  cpuset: Allow v2 behavior in v1 cgroup

 include/linux/cgroup-defs.h |  5 +++++
 kernel/cgroup/cgroup-v1.c   |  6 ++++++
 kernel/cgroup/cpuset.c      | 33 ++++++++++++++++++++-------------
 3 files changed, 31 insertions(+), 13 deletions(-)

-- 
1.8.3.1

[PATCH v2 1/2] cgroup: Add a hint to allow v1 controller to use v2 behavior

[Waiman Long]

For some cgroup controllers, the behavior on cgroup v1 differs somewhat
from cgroup v2. To indicate that a controller can use v2 behavior
in cgroup v1, a new cgroupfs mount option "v2_mode" is added. This
is essentially a hint, there is no obligation for a controller to
comply with this semantics.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 include/linux/cgroup-defs.h | 5 +++++
 kernel/cgroup/cgroup-v1.c   | 6 ++++++
 2 files changed, 11 insertions(+)

diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h
index 09f4c7d..fae33ed 100644
--- a/include/linux/cgroup-defs.h
+++ b/include/linux/cgroup-defs.h
@@ -74,6 +74,11 @@ enum {
 	 * aren't writeable from inside the namespace.
 	 */
 	CGRP_ROOT_NS_DELEGATE	= (1 << 3),
+
+	/*
+	 * A hint that v1 cgroup should use v2 behavior, if applicable.
+	 */
+	CGRP_ROOT_V2_MODE	= (1 << 4),
 };
 
 /* cftype->flags */
diff --git a/kernel/cgroup/cgroup-v1.c b/kernel/cgroup/cgroup-v1.c
index 7bf4b15..77e22a4 100644
--- a/kernel/cgroup/cgroup-v1.c
+++ b/kernel/cgroup/cgroup-v1.c
@@ -846,6 +846,8 @@ static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_roo
 		seq_puts(seq, ",noprefix");
 	if (root->flags & CGRP_ROOT_XATTR)
 		seq_puts(seq, ",xattr");
+	if (root->flags & CGRP_ROOT_V2_MODE)
+		seq_puts(seq, ",v2_mode");
 
 	spin_lock(&release_agent_path_lock);
 	if (strlen(root->release_agent_path))
@@ -900,6 +902,10 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
 			opts->cpuset_clone_children = true;
 			continue;
 		}
+		if (!strcmp(token, "v2_mode")) {
+			opts->flags |= CGRP_ROOT_V2_MODE;
+			continue;
+		}
 		if (!strcmp(token, "xattr")) {
 			opts->flags |= CGRP_ROOT_XATTR;
 			continue;
-- 
1.8.3.1

Re: [PATCH v2 1/2] cgroup: Add a hint to allow v1 controller to use v2 behavior

[Tejun Heo]

Hello, Waiman.

On Thu, Aug 17, 2017 at 02:25:38PM -0400, Waiman Long wrote:
> For some cgroup controllers, the behavior on cgroup v1 differs somewhat
> from cgroup v2. To indicate that a controller can use v2 behavior
> in cgroup v1, a new cgroupfs mount option "v2_mode" is added. This
> is essentially a hint, there is no obligation for a controller to
> comply with this semantics.

Heh, I don't think we can make it a generic flag.  We can't have one
kernel interpreting the option one way and later kernels doing
differently.  Let's just make it cpuset specific - something like
cpuset_v2_mode.

Thanks.

-- 
tejun

Re: [PATCH v2 1/2] cgroup: Add a hint to allow v1 controller to use v2 behavior

[Waiman Long]

On 08/17/2017 02:48 PM, Tejun Heo wrote:
> Hello, Waiman.
>
> On Thu, Aug 17, 2017 at 02:25:38PM -0400, Waiman Long wrote:
>> For some cgroup controllers, the behavior on cgroup v1 differs somewhat
>> from cgroup v2. To indicate that a controller can use v2 behavior
>> in cgroup v1, a new cgroupfs mount option "v2_mode" is added. This
>> is essentially a hint, there is no obligation for a controller to
>> comply with this semantics.
> Heh, I don't think we can make it a generic flag.  We can't have one
> kernel interpreting the option one way and later kernels doing
> differently.  Let's just make it cpuset specific - something like
> cpuset_v2_mode.
>
> Thanks.
>
That is fine for me. I will update the patchset accordingly.

Thanks,
Longman

[PATCH v2 2/2] cpuset: Allow v2 behavior in v1 cgroup

[Waiman Long]

Cpuset v2 has some useful behaviors that are not present in v1 because
of backward compatibility concern. One of that is the restoration of
the original cpu and memory node mask after a hot removal and addition
event sequence.

This patch makes the cpuset controller to check the CGRP_ROOT_V2_MODE
flag and use the v2 behavior if it is set.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c | 33 ++++++++++++++++++++-------------
 1 file changed, 20 insertions(+), 13 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 9ed6a05..1c44df7 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -300,6 +300,16 @@ static inline int is_spread_slab(const struct cpuset *cs)
 static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq);
 
 /*
+ * Cgroup v2 behavior is used when on default hierarchy or the
+ * cgroup_v2_mode flag is set.
+ */
+static inline bool is_in_v2_mode(void)
+{
+	return cgroup_subsys_on_dfl(cpuset_cgrp_subsys) ||
+	      (cpuset_cgrp_subsys.root->flags & CGRP_ROOT_V2_MODE);
+}
+
+/*
  * This is ugly, but preserves the userspace API for existing cpuset
  * users. If someone tries to mount the "cpuset" filesystem, we
  * silently switch it to mount "cgroup" instead
@@ -489,8 +499,7 @@ static int validate_change(struct cpuset *cur, struct cpuset *trial)
 
 	/* On legacy hiearchy, we must be a subset of our parent cpuset. */
 	ret = -EACCES;
-	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
-	    !is_cpuset_subset(trial, par))
+	if (!is_in_v2_mode() && !is_cpuset_subset(trial, par))
 		goto out;
 
 	/*
@@ -903,8 +912,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
 		 * If it becomes empty, inherit the effective mask of the
 		 * parent, which is guaranteed to have some CPUs.
 		 */
-		if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
-		    cpumask_empty(new_cpus))
+		if (is_in_v2_mode() && cpumask_empty(new_cpus))
 			cpumask_copy(new_cpus, parent->effective_cpus);
 
 		/* Skip the whole subtree if the cpumask remains the same. */
@@ -921,7 +929,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
 		cpumask_copy(cp->effective_cpus, new_cpus);
 		spin_unlock_irq(&callback_lock);
 
-		WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
+		WARN_ON(!is_in_v2_mode() &&
 			!cpumask_equal(cp->cpus_allowed, cp->effective_cpus));
 
 		update_tasks_cpumask(cp);
@@ -1157,8 +1165,7 @@ static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems)
 		 * If it becomes empty, inherit the effective mask of the
 		 * parent, which is guaranteed to have some MEMs.
 		 */
-		if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
-		    nodes_empty(*new_mems))
+		if (is_in_v2_mode() && nodes_empty(*new_mems))
 			*new_mems = parent->effective_mems;
 
 		/* Skip the whole subtree if the nodemask remains the same. */
@@ -1175,7 +1182,7 @@ static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems)
 		cp->effective_mems = *new_mems;
 		spin_unlock_irq(&callback_lock);
 
-		WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
+		WARN_ON(!is_in_v2_mode() &&
 			!nodes_equal(cp->mems_allowed, cp->effective_mems));
 
 		update_tasks_nodemask(cp);
@@ -1467,7 +1474,7 @@ static int cpuset_can_attach(struct cgroup_taskset *tset)
 
 	/* allow moving tasks into an empty cpuset if on default hierarchy */
 	ret = -ENOSPC;
-	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
+	if (!is_in_v2_mode() &&
 	    (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)))
 		goto out_unlock;
 
@@ -1985,7 +1992,7 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
 	cpuset_inc();
 
 	spin_lock_irq(&callback_lock);
-	if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) {
+	if (is_in_v2_mode()) {
 		cpumask_copy(cs->effective_cpus, parent->effective_cpus);
 		cs->effective_mems = parent->effective_mems;
 	}
@@ -2062,7 +2069,7 @@ static void cpuset_bind(struct cgroup_subsys_state *root_css)
 	mutex_lock(&cpuset_mutex);
 	spin_lock_irq(&callback_lock);
 
-	if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) {
+	if (is_in_v2_mode()) {
 		cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask);
 		top_cpuset.mems_allowed = node_possible_map;
 	} else {
@@ -2256,7 +2263,7 @@ static void cpuset_hotplug_update_tasks(struct cpuset *cs)
 	cpus_updated = !cpumask_equal(&new_cpus, cs->effective_cpus);
 	mems_updated = !nodes_equal(new_mems, cs->effective_mems);
 
-	if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys))
+	if (is_in_v2_mode())
 		hotplug_update_tasks(cs, &new_cpus, &new_mems,
 				     cpus_updated, mems_updated);
 	else
@@ -2287,7 +2294,7 @@ static void cpuset_hotplug_workfn(struct work_struct *work)
 	static cpumask_t new_cpus;
 	static nodemask_t new_mems;
 	bool cpus_updated, mems_updated;
-	bool on_dfl = cgroup_subsys_on_dfl(cpuset_cgrp_subsys);
+	bool on_dfl = is_in_v2_mode();
 
 	mutex_lock(&cpuset_mutex);
 
-- 
1.8.3.1