[PATCH 1/2] sched/psi: Create the psimon kthread outside of cgroup_mutex

a5b98009f16d ("sched/psi: fix race between file release and pressure write")
made pressure_write() hold cgroup_mutex across psi_trigger_create(), which
forks the psimon kthread for the first rtpoll trigger. As kthread creation
depends on the whole fork path, the commit inadvertently created a lot of
unwanted locking dependencies from cgroup_mutex.

sched_ext got hit by one: its enable path blocks forks and then grabs
cgroup_mutex, so a pressure write racing a scheduler enable deadlocks, with
every other fork piling up behind.

Fix it by splitting trigger creation so that the worker is forked with
cgroup_mutex dropped and the kernfs active reference left broken. The latter
matters because rmdir and cgroup.pressure writes drain active references
under cgroup_mutex. Publishing the trigger last keeps error reporting
synchronous and preserves the of->priv lifetime rules.

The trigger registered in the first stage pins the group's rtpoll machinery
across the unlocked window, leaving only creation races to resolve. The
catch-up poll on installation covers scheduling attempts dropped while there
was no worker.

v2: Retagged sched/psi (was cgroup).

Fixes: a5b98009f16d ("sched/psi: fix race between file release and pressure write")
Cc: [email protected]
Cc: Johannes Weiner <[email protected]>
Cc: Edward Adam Davis <[email protected]>
Cc: Chen Ridong <[email protected]>
Reported-by: Matt Fleming <[email protected]>
Closes: https://lore.kernel.org/all/[email protected]/
Signed-off-by: Tejun Heo <[email protected]>
---
 include/linux/psi.h    |  4 ++-
 kernel/cgroup/cgroup.c | 23 +++++++++++++-
 kernel/sched/psi.c     | 69 ++++++++++++++++++++++++++++++++----------
 3 files changed, 78 insertions(+), 18 deletions(-)

diff --git a/include/linux/psi.h b/include/linux/psi.h
index e0745873e3f2..7966e3ac03b9 100644
--- a/include/linux/psi.h
+++ b/include/linux/psi.h
@@ -25,7 +25,9 @@ void psi_memstall_leave(unsigned long *flags);
 int psi_show(struct seq_file *s, struct psi_group *group, enum psi_res res);
 struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
 				       enum psi_res res, struct file *file,
-				       struct kernfs_open_file *of);
+				       struct kernfs_open_file *of,
+				       bool *need_rtpoll_worker);
+int psi_trigger_create_rtpoll_worker(struct psi_group *group);
 void psi_trigger_destroy(struct psi_trigger *t);
 
 __poll_t psi_trigger_poll(void **trigger_ptr, struct file *file,
diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index 38f8d9df8fbc..b5b461d4418b 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -3996,6 +3996,7 @@ static ssize_t pressure_write(struct kernfs_open_file *of, char *buf,
 	struct psi_trigger *new;
 	struct cgroup *cgrp;
 	struct psi_group *psi;
+	bool need_rtpoll_worker;
 	ssize_t ret = 0;
 
 	cgrp = cgroup_kn_lock_live(of->kn, false);
@@ -4015,12 +4016,32 @@ static ssize_t pressure_write(struct kernfs_open_file *of, char *buf,
 	}
 
 	psi = cgroup_psi(cgrp);
-	new = psi_trigger_create(psi, buf, res, of->file, of);
+	new = psi_trigger_create(psi, buf, res, of->file, of,
+				 &need_rtpoll_worker);
 	if (IS_ERR(new)) {
 		ret = PTR_ERR(new);
 		goto out_unlock;
 	}
 
+	/*
+	 * The worker fork must run with neither cgroup_mutex nor the file's
+	 * kernfs active reference held. The latter is broken since
+	 * cgroup_kn_lock_live(). @of->priv may be released while unlocked, so
+	 * recheck before publishing @new.
+	 */
+	if (need_rtpoll_worker) {
+		cgroup_unlock();
+		ret = psi_trigger_create_rtpoll_worker(psi);
+		cgroup_lock();
+
+		if (!ret && !of->priv)
+			ret = -ENODEV;
+		if (ret) {
+			psi_trigger_destroy(new);
+			goto out_unlock;
+		}
+	}
+
 	smp_store_release(&ctx->psi.trigger, new);
 
 out_unlock:
diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
index d9c9d9480a45..565ec7b80743 100644
--- a/kernel/sched/psi.c
+++ b/kernel/sched/psi.c
@@ -1292,9 +1292,44 @@ int psi_show(struct seq_file *m, struct psi_group *group, enum psi_res res)
 	return 0;
 }
 
+/*
+ * Create @group's rtpoll worker after psi_trigger_create() reported the need
+ * for one. kthread creation depends on the whole fork path and we don't want
+ * all of that nested inside cgroup_mutex, so the caller must drop it and any
+ * other lock that forks can wait behind. If two callers race, the loser stops
+ * its never-woken kthread.
+ */
+int psi_trigger_create_rtpoll_worker(struct psi_group *group)
+{
+	struct task_struct *task;
+
+	task = kthread_create(psi_rtpoll_worker, group, "psimon");
+	if (IS_ERR(task))
+		return PTR_ERR(task);
+
+	scoped_guard(mutex, &group->rtpoll_trigger_lock) {
+		if (!rcu_access_pointer(group->rtpoll_task)) {
+			atomic_set(&group->rtpoll_wakeup, 0);
+			wake_up_process(task);
+			rcu_assign_pointer(group->rtpoll_task, task);
+
+			/*
+			 * Poll once to catch up on scheduling attempts dropped
+			 * while there was no rtpoll worker.
+			 */
+			psi_schedule_rtpoll_work(group, 1, true);
+			return 0;
+		}
+	}
+
+	kthread_stop(task);
+	return 0;
+}
+
 struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
 				       enum psi_res res, struct file *file,
-				       struct kernfs_open_file *of)
+				       struct kernfs_open_file *of,
+				       bool *need_rtpoll_worker)
 {
 	struct psi_trigger *t;
 	enum psi_states state;
@@ -1302,6 +1337,8 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
 	bool privileged;
 	u32 window_us;
 
+	*need_rtpoll_worker = false;
+
 	if (static_branch_likely(&psi_disabled))
 		return ERR_PTR(-EOPNOTSUPP);
 
@@ -1362,26 +1399,14 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
 	if (privileged) {
 		mutex_lock(&group->rtpoll_trigger_lock);
 
-		if (!rcu_access_pointer(group->rtpoll_task)) {
-			struct task_struct *task;
-
-			task = kthread_create(psi_rtpoll_worker, group, "psimon");
-			if (IS_ERR(task)) {
-				kfree(t);
-				mutex_unlock(&group->rtpoll_trigger_lock);
-				return ERR_CAST(task);
-			}
-			atomic_set(&group->rtpoll_wakeup, 0);
-			wake_up_process(task);
-			rcu_assign_pointer(group->rtpoll_task, task);
-		}
-
 		list_add(&t->node, &group->rtpoll_triggers);
 		group->rtpoll_min_period = min(group->rtpoll_min_period,
 			div_u64(t->win.size, UPDATES_PER_WINDOW));
 		group->rtpoll_nr_triggers[t->state]++;
 		group->rtpoll_states |= (1 << t->state);
 
+		*need_rtpoll_worker = !rcu_access_pointer(group->rtpoll_task);
+
 		mutex_unlock(&group->rtpoll_trigger_lock);
 	} else {
 		mutex_lock(&group->avgs_lock);
@@ -1541,6 +1566,8 @@ static ssize_t psi_write(struct file *file, const char __user *user_buf,
 	size_t buf_size;
 	struct seq_file *seq;
 	struct psi_trigger *new;
+	bool need_rtpoll_worker;
+	int ret;
 
 	if (static_branch_likely(&psi_disabled))
 		return -EOPNOTSUPP;
@@ -1565,12 +1592,22 @@ static ssize_t psi_write(struct file *file, const char __user *user_buf,
 		return -EBUSY;
 	}
 
-	new = psi_trigger_create(&psi_system, buf, res, file, NULL);
+	new = psi_trigger_create(&psi_system, buf, res, file, NULL,
+				 &need_rtpoll_worker);
 	if (IS_ERR(new)) {
 		mutex_unlock(&seq->lock);
 		return PTR_ERR(new);
 	}
 
+	if (need_rtpoll_worker) {
+		ret = psi_trigger_create_rtpoll_worker(&psi_system);
+		if (ret) {
+			psi_trigger_destroy(new);
+			mutex_unlock(&seq->lock);
+			return ret;
+		}
+	}
+
 	smp_store_release(&seq->private, new);
 	mutex_unlock(&seq->lock);
 
-- 
2.55.0