sched_ext: Introduce cgroup_lifetime_notifier

Other subsystems may make use of the cgroup hierarchy with the cgroup_bpf
support being one such example. For such a feature, it's useful to be able
to hook into cgroup creation and destruction paths to perform
feature-specific initializations and cleanups.

Add cgroup_lifetime_notifier which generates CGROUP_LIFETIME_ONLINE and
CGROUP_LIFETIME_OFFLINE events whenever cgroups are created and destroyed,
respectively.

The next patch will convert cgroup_bpf to use the new notifier and other
uses are planned.

Signed-off-by: Tejun Heo <tj@kernel.org>

Author: htejun

include/linux/cgroup.h

@@ -19,6 +19,7 @@
 #include <linux/kernfs.h>
 #include <linux/jump_label.h>
 #include <linux/types.h>
+#include <linux/notifier.h>
 #include <linux/ns_common.h>
 #include <linux/nsproxy.h>
 #include <linux/user_namespace.h>
@@ -40,7 +41,7 @@ struct kernel_clone_args;
 
 #ifdef CONFIG_CGROUPS
 
-enum {
+enum css_task_iter_flags {
 	CSS_TASK_ITER_PROCS    = (1U << 0),  /* walk only threadgroup leaders */
 	CSS_TASK_ITER_THREADED = (1U << 1),  /* walk all threaded css_sets in the domain */
 	CSS_TASK_ITER_SKIPPED  = (1U << 16), /* internal flags */
@@ -66,10 +67,16 @@ struct css_task_iter {
 	struct list_head		iters_node;	/* css_set->task_iters */
 };
 
+enum cgroup_lifetime_events {
+	CGROUP_LIFETIME_ONLINE,
+	CGROUP_LIFETIME_OFFLINE,
+};
+
 extern struct file_system_type cgroup_fs_type;
 extern struct cgroup_root cgrp_dfl_root;
 extern struct css_set init_css_set;
 extern spinlock_t css_set_lock;
+extern struct blocking_notifier_head cgroup_lifetime_notifier;
 
 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
 #include <linux/cgroup_subsys.h>

kernel/cgroup/cgroup.c

@@ -95,6 +95,9 @@ EXPORT_SYMBOL_GPL(cgroup_mutex);
 EXPORT_SYMBOL_GPL(css_set_lock);
 #endif
 
+struct blocking_notifier_head cgroup_lifetime_notifier =
+	BLOCKING_NOTIFIER_INIT(cgroup_lifetime_notifier);
+
 DEFINE_SPINLOCK(trace_cgroup_path_lock);
 char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
 static bool cgroup_debug __read_mostly;
@@ -1326,6 +1329,7 @@ static void cgroup_destroy_root(struct cgroup_root *root)
 {
 	struct cgroup *cgrp = &root->cgrp;
 	struct cgrp_cset_link *link, *tmp_link;
+	int ret;
 
 	trace_cgroup_destroy_root(root);
 
@@ -1334,6 +1338,10 @@ static void cgroup_destroy_root(struct cgroup_root *root)
 	BUG_ON(atomic_read(&root->nr_cgrps));
 	BUG_ON(!list_empty(&cgrp->self.children));
 
+	ret = blocking_notifier_call_chain(&cgroup_lifetime_notifier,
+					   CGROUP_LIFETIME_OFFLINE, cgrp);
+	WARN_ON_ONCE(notifier_to_errno(ret));
+
 	/* Rebind all subsystems back to the default hierarchy */
 	WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
 
@@ -2140,6 +2148,10 @@ int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
 		WARN_ON_ONCE(ret);
 	}
 
+	ret = blocking_notifier_call_chain(&cgroup_lifetime_notifier,
+					   CGROUP_LIFETIME_ONLINE, root_cgrp);
+	WARN_ON_ONCE(notifier_to_errno(ret));
+
 	trace_cgroup_setup_root(root);
 
 	/*
@@ -5733,6 +5745,15 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 			goto out_psi_free;
 	}
 
+	ret = blocking_notifier_call_chain_robust(&cgroup_lifetime_notifier,
+						  CGROUP_LIFETIME_ONLINE,
+						  CGROUP_LIFETIME_OFFLINE, cgrp);
+	ret = notifier_to_errno(ret);
+	if (ret) {
+		cgroup_bpf_offline(cgrp);
+		goto out_psi_free;
+	}
+
 	/* allocation complete, commit to creation */
 	spin_lock_irq(&css_set_lock);
 	for (i = 0; i < level; i++) {
@@ -5966,7 +5987,7 @@ static int cgroup_destroy_locked(struct cgroup *cgrp)
 	struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
 	struct cgroup_subsys_state *css;
 	struct cgrp_cset_link *link;
-	int ssid;
+	int ssid, ret;
 
 	lockdep_assert_held(&cgroup_mutex);
 
@@ -6027,6 +6048,10 @@ static int cgroup_destroy_locked(struct cgroup *cgrp)
 	if (cgrp->root == &cgrp_dfl_root)
 		cgroup_bpf_offline(cgrp);
 
+	ret = blocking_notifier_call_chain(&cgroup_lifetime_notifier,
+					   CGROUP_LIFETIME_OFFLINE, cgrp);
+	WARN_ON_ONCE(notifier_to_errno(ret));
+
 	/* put the base reference */
 	percpu_ref_kill(&cgrp->self.refcnt);