cgroup: cgroup v2 freezer

Cgroup v1 implements the freezer controller, which provides an ability
to stop the workload in a cgroup and temporarily free up some
resources (cpu, io, network bandwidth and, potentially, memory)
for some other tasks. Cgroup v2 lacks this functionality.

This patch implements freezer for cgroup v2.

Cgroup v2 freezer tries to put tasks into a state similar to jobctl
stop. This means that tasks can be killed, ptraced (using
PTRACE_SEIZE*), and interrupted. It is possible to attach to
a frozen task, get some information (e.g. read registers) and detach.
It's also possible to migrate a frozen tasks to another cgroup.

This differs cgroup v2 freezer from cgroup v1 freezer, which mostly
tried to imitate the system-wide freezer. However uninterruptible
sleep is fine when all tasks are going to be frozen (hibernation case),
it's not the acceptable state for some subset of the system.

Cgroup v2 freezer is not supporting freezing kthreads.
If a non-root cgroup contains kthread, the cgroup still can be frozen,
but the kthread will remain running, the cgroup will be shown
as non-frozen, and the notification will not be delivered.

* PTRACE_ATTACH is not working because non-fatal signal delivery
is blocked in frozen state.

There are some interface differences between cgroup v1 and cgroup v2
freezer too, which are required to conform the cgroup v2 interface
design principles:
1) There is no separate controller, which has to be turned on:
the functionality is always available and is represented by
cgroup.freeze and cgroup.events cgroup control files.
2) The desired state is defined by the cgroup.freeze control file.
Any hierarchical configuration is allowed.
3) The interface is asynchronous. The actual state is available
using cgroup.events control file ("frozen" field). There are no
dedicated transitional states.
4) It's allowed to make any changes with the cgroup hierarchy
(create new cgroups, remove old cgroups, move tasks between cgroups)
no matter if some cgroups are frozen.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
No-objection-from-me-by: Oleg Nesterov <oleg@redhat.com>
Cc: kernel-team@fb.com

3 files changed, 424 insertions, 5 deletions

diff --git a/kernel/cgroup/Makefile b/kernel/cgroup/Makefile
index 8d5689ca94b9..5d7a76bfbbb7 100644
--- a/kernel/cgroup/Makefile
+++ b/kernel/cgroup/Makefile
@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: GPL-2.0
-obj-y := cgroup.o rstat.o namespace.o cgroup-v1.o
+obj-y := cgroup.o rstat.o namespace.o cgroup-v1.o freezer.o
 
 obj-$(CONFIG_CGROUP_FREEZER) += legacy_freezer.o
 obj-$(CONFIG_CGROUP_PIDS) += pids.o
diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index 786ceef2f222..6895464b54c6 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -2435,8 +2435,15 @@ static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
 			get_css_set(to_cset);
 			to_cset->nr_tasks++;
 			css_set_move_task(task, from_cset, to_cset, true);
-			put_css_set_locked(from_cset);
 			from_cset->nr_tasks--;
+			/*
+			 * If the source or destination cgroup is frozen,
+			 * the task might require to change its state.
+			 */
+			cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
+						    to_cset->dfl_cgrp);
+			put_css_set_locked(from_cset);
+
 		}
 	}
 	spin_unlock_irq(&css_set_lock);
@@ -3477,8 +3484,11 @@ static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
 
 static int cgroup_events_show(struct seq_file *seq, void *v)
 {
-	seq_printf(seq, "populated %d\n",
-		   cgroup_is_populated(seq_css(seq)->cgroup));
+	struct cgroup *cgrp = seq_css(seq)->cgroup;
+
+	seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
+	seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
+
 	return 0;
 }
 
@@ -3540,6 +3550,40 @@ static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
 }
 #endif
 
+static int cgroup_freeze_show(struct seq_file *seq, void *v)
+{
+	struct cgroup *cgrp = seq_css(seq)->cgroup;
+
+	seq_printf(seq, "%d\n", cgrp->freezer.freeze);
+
+	return 0;
+}
+
+static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
+				   char *buf, size_t nbytes, loff_t off)
+{
+	struct cgroup *cgrp;
+	ssize_t ret;
+	int freeze;
+
+	ret = kstrtoint(strstrip(buf), 0, &freeze);
+	if (ret)
+		return ret;
+
+	if (freeze < 0 || freeze > 1)
+		return -ERANGE;
+
+	cgrp = cgroup_kn_lock_live(of->kn, false);
+	if (!cgrp)
+		return -ENOENT;
+
+	cgroup_freeze(cgrp, freeze);
+
+	cgroup_kn_unlock(of->kn);
+
+	return nbytes;
+}
+
 static int cgroup_file_open(struct kernfs_open_file *of)
 {
 	struct cftype *cft = of->kn->priv;
@@ -4684,6 +4728,12 @@ static struct cftype cgroup_base_files[] = {
 		.seq_show = cgroup_stat_show,
 	},
 	{
+		.name = "cgroup.freeze",
+		.flags = CFTYPE_NOT_ON_ROOT,
+		.seq_show = cgroup_freeze_show,
+		.write = cgroup_freeze_write,
+	},
+	{
 		.name = "cpu.stat",
 		.flags = CFTYPE_NOT_ON_ROOT,
 		.seq_show = cpu_stat_show,
@@ -5033,12 +5083,29 @@ static struct cgroup *cgroup_create(struct cgroup *parent)
 	if (ret)
 		goto out_psi_free;
 
+	/*
+	 * New cgroup inherits effective freeze counter, and
+	 * if the parent has to be frozen, the child has too.
+	 */
+	cgrp->freezer.e_freeze = parent->freezer.e_freeze;
+	if (cgrp->freezer.e_freeze)
+		set_bit(CGRP_FROZEN, &cgrp->flags);
+
 	spin_lock_irq(&css_set_lock);
 	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
 		cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
 
-		if (tcgrp != cgrp)
+		if (tcgrp != cgrp) {
 			tcgrp->nr_descendants++;
+
+			/*
+			 * If the new cgroup is frozen, all ancestor cgroups
+			 * get a new frozen descendant, but their state can't
+			 * change because of this.
+			 */
+			if (cgrp->freezer.e_freeze)
+				tcgrp->freezer.nr_frozen_descendants++;
+		}
 	}
 	spin_unlock_irq(&css_set_lock);
 
@@ -5329,6 +5396,12 @@ static int cgroup_destroy_locked(struct cgroup *cgrp)
 	for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
 		tcgrp->nr_descendants--;
 		tcgrp->nr_dying_descendants++;
+		/*
+		 * If the dying cgroup is frozen, decrease frozen descendants
+		 * counters of ancestor cgroups.
+		 */
+		if (test_bit(CGRP_FROZEN, &cgrp->flags))
+			tcgrp->freezer.nr_frozen_descendants--;
 	}
 	spin_unlock_irq(&css_set_lock);
 
@@ -5782,6 +5855,29 @@ void cgroup_post_fork(struct task_struct *child)
 			cset->nr_tasks++;
 			css_set_move_task(child, NULL, cset, false);
 		}
+
+		/*
+		 * If the cgroup has to be frozen, the new task has too.
+		 * Let's set the JOBCTL_TRAP_FREEZE jobctl bit to get
+		 * the task into the frozen state.
+		 */
+		if (unlikely(cgroup_task_freeze(child))) {
+			struct cgroup *cgrp;
+
+			spin_lock(&child->sighand->siglock);
+			WARN_ON_ONCE(child->frozen);
+			cgrp = cset->dfl_cgrp;
+			child->jobctl |= JOBCTL_TRAP_FREEZE;
+			spin_unlock(&child->sighand->siglock);
+
+			/*
+			 * Calling cgroup_update_frozen() isn't required here,
+			 * because it will be called anyway a bit later
+			 * from do_freezer_trap(). So we avoid cgroup's
+			 * transient switch from the frozen state and back.
+			 */
+		}
+
 		spin_unlock_irq(&css_set_lock);
 	}
 
@@ -5830,6 +5926,12 @@ void cgroup_exit(struct task_struct *tsk)
 		spin_lock_irq(&css_set_lock);
 		css_set_move_task(tsk, cset, NULL, false);
 		cset->nr_tasks--;
+
+		if (unlikely(cgroup_task_frozen(tsk)))
+			cgroup_freezer_frozen_exit(tsk);
+		else if (unlikely(cgroup_task_freeze(tsk)))
+			cgroup_update_frozen(task_dfl_cgroup(tsk));
+
 		spin_unlock_irq(&css_set_lock);
 	} else {
 		get_css_set(cset);
diff --git a/kernel/cgroup/freezer.c b/kernel/cgroup/freezer.c
new file mode 100644
index 000000000000..9d8cda478fc9
--- /dev/null
+++ b/kernel/cgroup/freezer.c
@@ -0,0 +1,317 @@
+//SPDX-License-Identifier: GPL-2.0
+#include <linux/cgroup.h>
+#include <linux/sched.h>
+#include <linux/sched/task.h>
+#include <linux/sched/signal.h>
+
+#include "cgroup-internal.h"
+
+/*
+ * Propagate the cgroup frozen state upwards by the cgroup tree.
+ */
+static void cgroup_propagate_frozen(struct cgroup *cgrp, bool frozen)
+{
+	int desc = 1;
+
+	/*
+	 * If the new state is frozen, some freezing ancestor cgroups may change
+	 * their state too, depending on if all their descendants are frozen.
+	 *
+	 * Otherwise, all ancestor cgroups are forced into the non-frozen state.
+	 */
+	while ((cgrp = cgroup_parent(cgrp))) {
+		if (frozen) {
+			cgrp->freezer.nr_frozen_descendants += desc;
+			if (!test_bit(CGRP_FROZEN, &cgrp->flags) &&
+			    test_bit(CGRP_FREEZE, &cgrp->flags) &&
+			    cgrp->freezer.nr_frozen_descendants ==
+			    cgrp->nr_descendants) {
+				set_bit(CGRP_FROZEN, &cgrp->flags);
+				cgroup_file_notify(&cgrp->events_file);
+				desc++;
+			}
+		} else {
+			cgrp->freezer.nr_frozen_descendants -= desc;
+			if (test_bit(CGRP_FROZEN, &cgrp->flags)) {
+				clear_bit(CGRP_FROZEN, &cgrp->flags);
+				cgroup_file_notify(&cgrp->events_file);
+				desc++;
+			}
+		}
+	}
+}
+
+/*
+ * Revisit the cgroup frozen state.
+ * Checks if the cgroup is really frozen and perform all state transitions.
+ */
+void cgroup_update_frozen(struct cgroup *cgrp)
+{
+	bool frozen;
+
+	lockdep_assert_held(&css_set_lock);
+
+	/*
+	 * If the cgroup has to be frozen (CGRP_FREEZE bit set),
+	 * and all tasks are frozen and/or stopped, let's consider
+	 * the cgroup frozen. Otherwise it's not frozen.
+	 */
+	frozen = test_bit(CGRP_FREEZE, &cgrp->flags) &&
+		cgrp->freezer.nr_frozen_tasks == __cgroup_task_count(cgrp);
+
+	if (frozen) {
+		/* Already there? */
+		if (test_bit(CGRP_FROZEN, &cgrp->flags))
+			return;
+
+		set_bit(CGRP_FROZEN, &cgrp->flags);
+	} else {
+		/* Already there? */
+		if (!test_bit(CGRP_FROZEN, &cgrp->flags))
+			return;
+
+		clear_bit(CGRP_FROZEN, &cgrp->flags);
+	}
+	cgroup_file_notify(&cgrp->events_file);
+
+	/* Update the state of ancestor cgroups. */
+	cgroup_propagate_frozen(cgrp, frozen);
+}
+
+/*
+ * Increment cgroup's nr_frozen_tasks.
+ */
+static void cgroup_inc_frozen_cnt(struct cgroup *cgrp)
+{
+	cgrp->freezer.nr_frozen_tasks++;
+}
+
+/*
+ * Decrement cgroup's nr_frozen_tasks.
+ */
+static void cgroup_dec_frozen_cnt(struct cgroup *cgrp)
+{
+	cgrp->freezer.nr_frozen_tasks--;
+	WARN_ON_ONCE(cgrp->freezer.nr_frozen_tasks < 0);
+}
+
+/*
+ * Enter frozen/stopped state, if not yet there. Update cgroup's counters,
+ * and revisit the state of the cgroup, if necessary.
+ */
+void cgroup_enter_frozen(void)
+{
+	struct cgroup *cgrp;
+
+	if (current->frozen)
+		return;
+
+	spin_lock_irq(&css_set_lock);
+	current->frozen = true;
+	cgrp = task_dfl_cgroup(current);
+	cgroup_inc_frozen_cnt(cgrp);
+	cgroup_update_frozen(cgrp);
+	spin_unlock_irq(&css_set_lock);
+}
+
+/*
+ * Conditionally leave frozen/stopped state. Update cgroup's counters,
+ * and revisit the state of the cgroup, if necessary.
+ *
+ * If always_leave is not set, and the cgroup is freezing,
+ * we're racing with the cgroup freezing. In this case, we don't
+ * drop the frozen counter to avoid a transient switch to
+ * the unfrozen state.
+ */
+void cgroup_leave_frozen(bool always_leave)
+{
+	struct cgroup *cgrp;
+
+	spin_lock_irq(&css_set_lock);
+	cgrp = task_dfl_cgroup(current);
+	if (always_leave || !test_bit(CGRP_FREEZE, &cgrp->flags)) {
+		cgroup_dec_frozen_cnt(cgrp);
+		cgroup_update_frozen(cgrp);
+		WARN_ON_ONCE(!current->frozen);
+		current->frozen = false;
+	}
+	spin_unlock_irq(&css_set_lock);
+
+	if (unlikely(current->frozen)) {
+		/*
+		 * If the task remained in the frozen state,
+		 * make sure it won't reach userspace without
+		 * entering the signal handling loop.
+		 */
+		spin_lock_irq(&current->sighand->siglock);
+		recalc_sigpending();
+		spin_unlock_irq(&current->sighand->siglock);
+	}
+}
+
+/*
+ * Freeze or unfreeze the task by setting or clearing the JOBCTL_TRAP_FREEZE
+ * jobctl bit.
+ */
+static void cgroup_freeze_task(struct task_struct *task, bool freeze)
+{
+	unsigned long flags;
+
+	/* If the task is about to die, don't bother with freezing it. */
+	if (!lock_task_sighand(task, &flags))
+		return;
+
+	if (freeze) {
+		task->jobctl |= JOBCTL_TRAP_FREEZE;
+		signal_wake_up(task, false);
+	} else {
+		task->jobctl &= ~JOBCTL_TRAP_FREEZE;
+		wake_up_process(task);
+	}
+
+	unlock_task_sighand(task, &flags);
+}
+
+/*
+ * Freeze or unfreeze all tasks in the given cgroup.
+ */
+static void cgroup_do_freeze(struct cgroup *cgrp, bool freeze)
+{
+	struct css_task_iter it;
+	struct task_struct *task;
+
+	lockdep_assert_held(&cgroup_mutex);
+
+	spin_lock_irq(&css_set_lock);
+	if (freeze)
+		set_bit(CGRP_FREEZE, &cgrp->flags);
+	else
+		clear_bit(CGRP_FREEZE, &cgrp->flags);
+	spin_unlock_irq(&css_set_lock);
+
+	css_task_iter_start(&cgrp->self, 0, &it);
+	while ((task = css_task_iter_next(&it))) {
+		/*
+		 * Ignore kernel threads here. Freezing cgroups containing
+		 * kthreads isn't supported.
+		 */
+		if (task->flags & PF_KTHREAD)
+			continue;
+		cgroup_freeze_task(task, freeze);
+	}
+	css_task_iter_end(&it);
+
+	/*
+	 * Cgroup state should be revisited here to cover empty leaf cgroups
+	 * and cgroups which descendants are already in the desired state.
+	 */
+	spin_lock_irq(&css_set_lock);
+	if (cgrp->nr_descendants == cgrp->freezer.nr_frozen_descendants)
+		cgroup_update_frozen(cgrp);
+	spin_unlock_irq(&css_set_lock);
+}
+
+/*
+ * Adjust the task state (freeze or unfreeze) and revisit the state of
+ * source and destination cgroups.
+ */
+void cgroup_freezer_migrate_task(struct task_struct *task,
+				 struct cgroup *src, struct cgroup *dst)
+{
+	lockdep_assert_held(&css_set_lock);
+
+	/*
+	 * Kernel threads are not supposed to be frozen at all.
+	 */
+	if (task->flags & PF_KTHREAD)
+		return;
+
+	/*
+	 * Adjust counters of freezing and frozen tasks.
+	 * Note, that if the task is frozen, but the destination cgroup is not
+	 * frozen, we bump both counters to keep them balanced.
+	 */
+	if (task->frozen) {
+		cgroup_inc_frozen_cnt(dst);
+		cgroup_dec_frozen_cnt(src);
+	}
+	cgroup_update_frozen(dst);
+	cgroup_update_frozen(src);
+
+	/*
+	 * Force the task to the desired state.
+	 */
+	cgroup_freeze_task(task, test_bit(CGRP_FREEZE, &dst->flags));
+}
+
+void cgroup_freezer_frozen_exit(struct task_struct *task)
+{
+	struct cgroup *cgrp = task_dfl_cgroup(task);
+
+	lockdep_assert_held(&css_set_lock);
+
+	cgroup_dec_frozen_cnt(cgrp);
+	cgroup_update_frozen(cgrp);
+}
+
+void cgroup_freeze(struct cgroup *cgrp, bool freeze)
+{
+	struct cgroup_subsys_state *css;
+	struct cgroup *dsct;
+	bool applied = false;
+
+	lockdep_assert_held(&cgroup_mutex);
+
+	/*
+	 * Nothing changed? Just exit.
+	 */
+	if (cgrp->freezer.freeze == freeze)
+		return;
+
+	cgrp->freezer.freeze = freeze;
+
+	/*
+	 * Propagate changes downwards the cgroup tree.
+	 */
+	css_for_each_descendant_pre(css, &cgrp->self) {
+		dsct = css->cgroup;
+
+		if (cgroup_is_dead(dsct))
+			continue;
+
+		if (freeze) {
+			dsct->freezer.e_freeze++;
+			/*
+			 * Already frozen because of ancestor's settings?
+			 */
+			if (dsct->freezer.e_freeze > 1)
+				continue;
+		} else {
+			dsct->freezer.e_freeze--;
+			/*
+			 * Still frozen because of ancestor's settings?
+			 */
+			if (dsct->freezer.e_freeze > 0)
+				continue;
+
+			WARN_ON_ONCE(dsct->freezer.e_freeze < 0);
+		}
+
+		/*
+		 * Do change actual state: freeze or unfreeze.
+		 */
+		cgroup_do_freeze(dsct, freeze);
+		applied = true;
+	}
+
+	/*
+	 * Even if the actual state hasn't changed, let's notify a user.
+	 * The state can be enforced by an ancestor cgroup: the cgroup
+	 * can already be in the desired state or it can be locked in the
+	 * opposite state, so that the transition will never happen.
+	 * In both cases it's better to notify a user, that there is
+	 * nothing to wait for.
+	 */
+	if (!applied)
+		cgroup_file_notify(&cgrp->events_file);
+}