1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Generic process-grouping system.
4	*
5	* Based originally on the cpuset system, extracted by Paul Menage
6	* Copyright (C) 2006 Google, Inc
7	*
8	* Notifications support
9	* Copyright (C) 2009 Nokia Corporation
10	* Author: Kirill A. Shutemov
11	*
12	* Copyright notices from the original cpuset code:
13	* --------------------------------------------------
14	* Copyright (C) 2003 BULL SA.
15	* Copyright (C) 2004-2006 Silicon Graphics, Inc.
16	*
17	* Portions derived from Patrick Mochel's sysfs code.
18	* sysfs is Copyright (c) 2001-3 Patrick Mochel
19	*
20	* 2003-10-10 Written by Simon Derr.
21	* 2003-10-22 Updates by Stephen Hemminger.
22	* 2004 May-July Rework by Paul Jackson.
23	* ---------------------------------------------------
24	*/
25
26	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
27
28	#include "cgroup-internal.h"
29
30	#include <linux/bpf-cgroup.h>
31	#include <linux/cred.h>
32	#include <linux/errno.h>
33	#include <linux/init_task.h>
34	#include <linux/kernel.h>
35	#include <linux/magic.h>
36	#include <linux/mutex.h>
37	#include <linux/mount.h>
38	#include <linux/pagemap.h>
39	#include <linux/proc_fs.h>
40	#include <linux/rcupdate.h>
41	#include <linux/sched.h>
42	#include <linux/sched/task.h>
43	#include <linux/slab.h>
44	#include <linux/spinlock.h>
45	#include <linux/percpu-rwsem.h>
46	#include <linux/string.h>
47	#include <linux/hashtable.h>
48	#include <linux/idr.h>
49	#include <linux/kthread.h>
50	#include <linux/atomic.h>
51	#include <linux/cpuset.h>
52	#include <linux/proc_ns.h>
53	#include <linux/nsproxy.h>
54	#include <linux/file.h>
55	#include <linux/fs_parser.h>
56	#include <linux/sched/cputime.h>
57	#include <linux/sched/deadline.h>
58	#include <linux/psi.h>
59	#include <linux/nstree.h>
60	#include <linux/irq_work.h>
61	#include <net/sock.h>
62
63	#define CREATE_TRACE_POINTS
64	#include <trace/events/cgroup.h>
65
66	#define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
67	MAX_CFTYPE_NAME + 2)
68	/* let's not notify more than 100 times per second */
69	#define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
70
71	/*
72	* To avoid confusing the compiler (and generating warnings) with code
73	* that attempts to access what would be a 0-element array (i.e. sized
74	* to a potentially empty array when CGROUP_SUBSYS_COUNT == 0), this
75	* constant expression can be added.
76	*/
77	#define CGROUP_HAS_SUBSYS_CONFIG (CGROUP_SUBSYS_COUNT > 0)
78
79	/*
80	* cgroup_mutex is the master lock. Any modification to cgroup or its
81	* hierarchy must be performed while holding it.
82	*
83	* css_set_lock protects task->cgroups pointer, the list of css_set
84	* objects, and the chain of tasks off each css_set.
85	*
86	* These locks are exported if CONFIG_PROVE_RCU so that accessors in
87	* cgroup.h can use them for lockdep annotations.
88	*/
89	DEFINE_MUTEX(cgroup_mutex);
90	DEFINE_SPINLOCK(css_set_lock);
91
92	#if (defined CONFIG_PROVE_RCU || defined CONFIG_LOCKDEP)
93	EXPORT_SYMBOL_GPL(cgroup_mutex);
94	EXPORT_SYMBOL_GPL(css_set_lock);
95	#endif
96
97	struct blocking_notifier_head cgroup_lifetime_notifier =
98	BLOCKING_NOTIFIER_INIT(cgroup_lifetime_notifier);
99
100	DEFINE_SPINLOCK(trace_cgroup_path_lock);
101	char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
102	static bool cgroup_debug __read_mostly;
103
104	/*
105	* Protects cgroup_idr and css_idr so that IDs can be released without
106	* grabbing cgroup_mutex.
107	*/
108	static DEFINE_SPINLOCK(cgroup_idr_lock);
109
110	/*
111	* Protects cgroup_file->kn for !self csses. It synchronizes notifications
112	* against file removal/re-creation across css hiding.
113	*/
114	static DEFINE_SPINLOCK(cgroup_file_kn_lock);
115
116	DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
117
118	#define cgroup_assert_mutex_or_rcu_locked() \
119	RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
120	!lockdep_is_held(&cgroup_mutex), \
121	"cgroup_mutex or RCU read lock required");
122
123	/*
124	* cgroup destruction makes heavy use of work items and there can be a lot
125	* of concurrent destructions. Use a separate workqueue so that cgroup
126	* destruction work items don't end up filling up max_active of system_percpu_wq
127	* which may lead to deadlock.
128	*
129	* A cgroup destruction should enqueue work sequentially to:
130	* cgroup_offline_wq: use for css offline work
131	* cgroup_release_wq: use for css release work
132	* cgroup_free_wq: use for free work
133	*
134	* Rationale for using separate workqueues:
135	* The cgroup root free work may depend on completion of other css offline
136	* operations. If all tasks were enqueued to a single workqueue, this could
137	* create a deadlock scenario where:
138	* - Free work waits for other css offline work to complete.
139	* - But other css offline work is queued after free work in the same queue.
140	*
141	* Example deadlock scenario with single workqueue (cgroup_destroy_wq):
142	* 1. umount net_prio
143	* 2. net_prio root destruction enqueues work to cgroup_destroy_wq (CPUx)
144	* 3. perf_event CSS A offline enqueues work to same cgroup_destroy_wq (CPUx)
145	* 4. net_prio cgroup_destroy_root->cgroup_lock_and_drain_offline.
146	* 5. net_prio root destruction blocks waiting for perf_event CSS A offline,
147	* which can never complete as it's behind in the same queue and
148	* workqueue's max_active is 1.
149	*/
150	static struct workqueue_struct *cgroup_offline_wq;
151	static struct workqueue_struct *cgroup_release_wq;
152	static struct workqueue_struct *cgroup_free_wq;
153
154	/* generate an array of cgroup subsystem pointers */
155	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
156	struct cgroup_subsys *cgroup_subsys[] = {
157	#include <linux/cgroup_subsys.h>
158	};
159	#undef SUBSYS
160
161	/* array of cgroup subsystem names */
162	#define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
163	static const char *cgroup_subsys_name[] = {
164	#include <linux/cgroup_subsys.h>
165	};
166	#undef SUBSYS
167
168	/* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
169	#define SUBSYS(_x) \
170	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
171	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
172	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
173	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
174	#include <linux/cgroup_subsys.h>
175	#undef SUBSYS
176
177	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
178	static struct static_key_true *cgroup_subsys_enabled_key[] = {
179	#include <linux/cgroup_subsys.h>
180	};
181	#undef SUBSYS
182
183	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
184	static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
185	#include <linux/cgroup_subsys.h>
186	};
187	#undef SUBSYS
188
189	static DEFINE_PER_CPU(struct css_rstat_cpu, root_rstat_cpu);
190	static DEFINE_PER_CPU(struct cgroup_rstat_base_cpu, root_rstat_base_cpu);
191
192	/* the default hierarchy */
193	struct cgroup_root cgrp_dfl_root = {
194	.cgrp.self.rstat_cpu = &root_rstat_cpu,
195	.cgrp.rstat_base_cpu = &root_rstat_base_cpu,
196	};
197	EXPORT_SYMBOL_GPL(cgrp_dfl_root);
198
199	/*
200	* The default hierarchy always exists but is hidden until mounted for the
201	* first time. This is for backward compatibility.
202	*/
203	bool cgrp_dfl_visible;
204
205	/* some controllers are not supported in the default hierarchy */
206	static u16 cgrp_dfl_inhibit_ss_mask;
207
208	/* some controllers are implicitly enabled on the default hierarchy */
209	static u16 cgrp_dfl_implicit_ss_mask;
210
211	/* some controllers can be threaded on the default hierarchy */
212	static u16 cgrp_dfl_threaded_ss_mask;
213
214	/* The list of hierarchy roots */
215	LIST_HEAD(cgroup_roots);
216	static int cgroup_root_count;
217
218	/* hierarchy ID allocation and mapping, protected by cgroup_mutex */
219	static DEFINE_IDR(cgroup_hierarchy_idr);
220
221	/*
222	* Assign a monotonically increasing serial number to csses. It guarantees
223	* cgroups with bigger numbers are newer than those with smaller numbers.
224	* Also, as csses are always appended to the parent's ->children list, it
225	* guarantees that sibling csses are always sorted in the ascending serial
226	* number order on the list. Protected by cgroup_mutex.
227	*/
228	static u64 css_serial_nr_next = 1;
229
230	/*
231	* These bitmasks identify subsystems with specific features to avoid
232	* having to do iterative checks repeatedly.
233	*/
234	static u16 have_fork_callback __read_mostly;
235	static u16 have_exit_callback __read_mostly;
236	static u16 have_release_callback __read_mostly;
237	static u16 have_canfork_callback __read_mostly;
238
239	static bool have_favordynmods __ro_after_init = IS_ENABLED(CONFIG_CGROUP_FAVOR_DYNMODS);
240
241	/*
242	* Write protected by cgroup_mutex and write-lock of cgroup_threadgroup_rwsem,
243	* read protected by either.
244	*
245	* Can only be turned on, but not turned off.
246	*/
247	bool cgroup_enable_per_threadgroup_rwsem __read_mostly;
248
249	/* cgroup namespace for init task */
250	struct cgroup_namespace init_cgroup_ns = {
251	.ns = NS_COMMON_INIT(init_cgroup_ns),
252	.user_ns = &init_user_ns,
253	.root_cset = &init_css_set,
254	};
255
256	static struct file_system_type cgroup2_fs_type;
257	static struct cftype cgroup_base_files[];
258	static struct cftype cgroup_psi_files[];
259
260	/* cgroup optional features */
261	enum cgroup_opt_features {
262	#ifdef CONFIG_PSI
263	OPT_FEATURE_PRESSURE,
264	#endif
265	OPT_FEATURE_COUNT
266	};
267
268	static const char *cgroup_opt_feature_names[OPT_FEATURE_COUNT] = {
269	#ifdef CONFIG_PSI
270	"pressure",
271	#endif
272	};
273
274	static u16 cgroup_feature_disable_mask __read_mostly;
275
276	static int cgroup_apply_control(struct cgroup *cgrp);
277	static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
278	static void css_task_iter_skip(struct css_task_iter *it,
279	struct task_struct *task);
280	static int cgroup_destroy_locked(struct cgroup *cgrp);
281	static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
282	struct cgroup_subsys *ss);
283	static void css_release(struct percpu_ref *ref);
284	static void kill_css(struct cgroup_subsys_state *css);
285	static int cgroup_addrm_files(struct cgroup_subsys_state *css,
286	struct cgroup *cgrp, struct cftype cfts[],
287	bool is_add);
288	static void cgroup_rt_init(void);
289
290	#ifdef CONFIG_DEBUG_CGROUP_REF
291	#define CGROUP_REF_FN_ATTRS noinline
292	#define CGROUP_REF_EXPORT(fn) EXPORT_SYMBOL_GPL(fn);
293	#include <linux/cgroup_refcnt.h>
294	#endif
295
296	/**
297	* cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
298	* @ssid: subsys ID of interest
299	*
300	* cgroup_subsys_enabled() can only be used with literal subsys names which
301	* is fine for individual subsystems but unsuitable for cgroup core. This
302	* is slower static_key_enabled() based test indexed by @ssid.
303	*/
304	bool cgroup_ssid_enabled(int ssid)
305	{
306	if (!CGROUP_HAS_SUBSYS_CONFIG)
307	return false;
308
309	return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
310	}
311
312	/**
313	* cgroup_on_dfl - test whether a cgroup is on the default hierarchy
314	* @cgrp: the cgroup of interest
315	*
316	* The default hierarchy is the v2 interface of cgroup and this function
317	* can be used to test whether a cgroup is on the default hierarchy for
318	* cases where a subsystem should behave differently depending on the
319	* interface version.
320	*
321	* List of changed behaviors:
322	*
323	* - Mount options "noprefix", "xattr", "clone_children", "release_agent"
324	* and "name" are disallowed.
325	*
326	* - When mounting an existing superblock, mount options should match.
327	*
328	* - rename(2) is disallowed.
329	*
330	* - "tasks" is removed. Everything should be at process granularity. Use
331	* "cgroup.procs" instead.
332	*
333	* - "cgroup.procs" is not sorted. pids will be unique unless they got
334	* recycled in-between reads.
335	*
336	* - "release_agent" and "notify_on_release" are removed. Replacement
337	* notification mechanism will be implemented.
338	*
339	* - "cgroup.clone_children" is removed.
340	*
341	* - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
342	* and its descendants contain no task; otherwise, 1. The file also
343	* generates kernfs notification which can be monitored through poll and
344	* [di]notify when the value of the file changes.
345	*
346	* - cpuset: tasks will be kept in empty cpusets when hotplug happens and
347	* take masks of ancestors with non-empty cpus/mems, instead of being
348	* moved to an ancestor.
349	*
350	* - cpuset: a task can be moved into an empty cpuset, and again it takes
351	* masks of ancestors.
352	*
353	* - blkcg: blk-throttle becomes properly hierarchical.
354	*/
355	bool cgroup_on_dfl(const struct cgroup *cgrp)
356	{
357	return cgrp->root == &cgrp_dfl_root;
358	}
359
360	/* IDR wrappers which synchronize using cgroup_idr_lock */
361	static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
362	gfp_t gfp_mask)
363	{
364	int ret;
365
366	idr_preload(gfp_mask);
367	spin_lock_bh(lock: &cgroup_idr_lock);
368	ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
369	spin_unlock_bh(lock: &cgroup_idr_lock);
370	idr_preload_end();
371	return ret;
372	}
373
374	static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
375	{
376	void *ret;
377
378	spin_lock_bh(lock: &cgroup_idr_lock);
379	ret = idr_replace(idr, ptr, id);
380	spin_unlock_bh(lock: &cgroup_idr_lock);
381	return ret;
382	}
383
384	static void cgroup_idr_remove(struct idr *idr, int id)
385	{
386	spin_lock_bh(lock: &cgroup_idr_lock);
387	idr_remove(idr, id);
388	spin_unlock_bh(lock: &cgroup_idr_lock);
389	}
390
391	static bool cgroup_has_tasks(struct cgroup *cgrp)
392	{
393	return cgrp->nr_populated_csets;
394	}
395
396	static bool cgroup_is_threaded(struct cgroup *cgrp)
397	{
398	return cgrp->dom_cgrp != cgrp;
399	}
400
401	/* can @cgrp host both domain and threaded children? */
402	static bool cgroup_is_mixable(struct cgroup *cgrp)
403	{
404	/*
405	* Root isn't under domain level resource control exempting it from
406	* the no-internal-process constraint, so it can serve as a thread
407	* root and a parent of resource domains at the same time.
408	*/
409	return !cgroup_parent(cgrp);
410	}
411
412	/* can @cgrp become a thread root? Should always be true for a thread root */
413	static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
414	{
415	/* mixables don't care */
416	if (cgroup_is_mixable(cgrp))
417	return true;
418
419	/* domain roots can't be nested under threaded */
420	if (cgroup_is_threaded(cgrp))
421	return false;
422
423	/* can only have either domain or threaded children */
424	if (cgrp->nr_populated_domain_children)
425	return false;
426
427	/* and no domain controllers can be enabled */
428	if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
429	return false;
430
431	return true;
432	}
433
434	/* is @cgrp root of a threaded subtree? */
435	static bool cgroup_is_thread_root(struct cgroup *cgrp)
436	{
437	/* thread root should be a domain */
438	if (cgroup_is_threaded(cgrp))
439	return false;
440
441	/* a domain w/ threaded children is a thread root */
442	if (cgrp->nr_threaded_children)
443	return true;
444
445	/*
446	* A domain which has tasks and explicit threaded controllers
447	* enabled is a thread root.
448	*/
449	if (cgroup_has_tasks(cgrp) &&
450	(cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
451	return true;
452
453	return false;
454	}
455
456	/* a domain which isn't connected to the root w/o brekage can't be used */
457	static bool cgroup_is_valid_domain(struct cgroup *cgrp)
458	{
459	/* the cgroup itself can be a thread root */
460	if (cgroup_is_threaded(cgrp))
461	return false;
462
463	/* but the ancestors can't be unless mixable */
464	while ((cgrp = cgroup_parent(cgrp))) {
465	if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
466	return false;
467	if (cgroup_is_threaded(cgrp))
468	return false;
469	}
470
471	return true;
472	}
473
474	/* subsystems visibly enabled on a cgroup */
475	static u16 cgroup_control(struct cgroup *cgrp)
476	{
477	struct cgroup *parent = cgroup_parent(cgrp);
478	u16 root_ss_mask = cgrp->root->subsys_mask;
479
480	if (parent) {
481	u16 ss_mask = parent->subtree_control;
482
483	/* threaded cgroups can only have threaded controllers */
484	if (cgroup_is_threaded(cgrp))
485	ss_mask &= cgrp_dfl_threaded_ss_mask;
486	return ss_mask;
487	}
488
489	if (cgroup_on_dfl(cgrp))
490	root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
491	cgrp_dfl_implicit_ss_mask);
492	return root_ss_mask;
493	}
494
495	/* subsystems enabled on a cgroup */
496	static u16 cgroup_ss_mask(struct cgroup *cgrp)
497	{
498	struct cgroup *parent = cgroup_parent(cgrp);
499
500	if (parent) {
501	u16 ss_mask = parent->subtree_ss_mask;
502
503	/* threaded cgroups can only have threaded controllers */
504	if (cgroup_is_threaded(cgrp))
505	ss_mask &= cgrp_dfl_threaded_ss_mask;
506	return ss_mask;
507	}
508
509	return cgrp->root->subsys_mask;
510	}
511
512	/**
513	* cgroup_css - obtain a cgroup's css for the specified subsystem
514	* @cgrp: the cgroup of interest
515	* @ss: the subsystem of interest (%NULL returns @cgrp->self)
516	*
517	* Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
518	* function must be called either under cgroup_mutex or rcu_read_lock() and
519	* the caller is responsible for pinning the returned css if it wants to
520	* keep accessing it outside the said locks. This function may return
521	* %NULL if @cgrp doesn't have @subsys_id enabled.
522	*/
523	static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
524	struct cgroup_subsys *ss)
525	{
526	if (CGROUP_HAS_SUBSYS_CONFIG && ss)
527	return rcu_dereference_check(cgrp->subsys[ss->id],
528	lockdep_is_held(&cgroup_mutex));
529	else
530	return &cgrp->self;
531	}
532
533	/**
534	* cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
535	* @cgrp: the cgroup of interest
536	* @ss: the subsystem of interest (%NULL returns @cgrp->self)
537	*
538	* Similar to cgroup_css() but returns the effective css, which is defined
539	* as the matching css of the nearest ancestor including self which has @ss
540	* enabled. If @ss is associated with the hierarchy @cgrp is on, this
541	* function is guaranteed to return non-NULL css.
542	*/
543	static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
544	struct cgroup_subsys *ss)
545	{
546	lockdep_assert_held(&cgroup_mutex);
547
548	if (!ss)
549	return &cgrp->self;
550
551	/*
552	* This function is used while updating css associations and thus
553	* can't test the csses directly. Test ss_mask.
554	*/
555	while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
556	cgrp = cgroup_parent(cgrp);
557	if (!cgrp)
558	return NULL;
559	}
560
561	return cgroup_css(cgrp, ss);
562	}
563
564	/**
565	* cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
566	* @cgrp: the cgroup of interest
567	* @ss: the subsystem of interest
568	*
569	* Find and get the effective css of @cgrp for @ss. The effective css is
570	* defined as the matching css of the nearest ancestor including self which
571	* has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
572	* the root css is returned, so this function always returns a valid css.
573	*
574	* The returned css is not guaranteed to be online, and therefore it is the
575	* callers responsibility to try get a reference for it.
576	*/
577	struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
578	struct cgroup_subsys *ss)
579	{
580	struct cgroup_subsys_state *css;
581
582	if (!CGROUP_HAS_SUBSYS_CONFIG)
583	return NULL;
584
585	do {
586	css = cgroup_css(cgrp, ss);
587
588	if (css)
589	return css;
590	cgrp = cgroup_parent(cgrp);
591	} while (cgrp);
592
593	return init_css_set.subsys[ss->id];
594	}
595
596	/**
597	* cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
598	* @cgrp: the cgroup of interest
599	* @ss: the subsystem of interest
600	*
601	* Find and get the effective css of @cgrp for @ss. The effective css is
602	* defined as the matching css of the nearest ancestor including self which
603	* has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
604	* the root css is returned, so this function always returns a valid css.
605	* The returned css must be put using css_put().
606	*/
607	struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
608	struct cgroup_subsys *ss)
609	{
610	struct cgroup_subsys_state *css;
611
612	if (!CGROUP_HAS_SUBSYS_CONFIG)
613	return NULL;
614
615	rcu_read_lock();
616
617	do {
618	css = cgroup_css(cgrp, ss);
619
620	if (css && css_tryget_online(css))
621	goto out_unlock;
622	cgrp = cgroup_parent(cgrp);
623	} while (cgrp);
624
625	css = init_css_set.subsys[ss->id];
626	css_get(css);
627	out_unlock:
628	rcu_read_unlock();
629	return css;
630	}
631	EXPORT_SYMBOL_GPL(cgroup_get_e_css);
632
633	static void cgroup_get_live(struct cgroup *cgrp)
634	{
635	WARN_ON_ONCE(cgroup_is_dead(cgrp));
636	cgroup_get(cgrp);
637	}
638
639	/**
640	* __cgroup_task_count - count the number of tasks in a cgroup. The caller
641	* is responsible for taking the css_set_lock.
642	* @cgrp: the cgroup in question
643	*/
644	int __cgroup_task_count(const struct cgroup *cgrp)
645	{
646	int count = 0;
647	struct cgrp_cset_link *link;
648
649	lockdep_assert_held(&css_set_lock);
650
651	list_for_each_entry(link, &cgrp->cset_links, cset_link)
652	count += link->cset->nr_tasks;
653
654	return count;
655	}
656
657	/**
658	* cgroup_task_count - count the number of tasks in a cgroup.
659	* @cgrp: the cgroup in question
660	*/
661	int cgroup_task_count(const struct cgroup *cgrp)
662	{
663	int count;
664
665	spin_lock_irq(lock: &css_set_lock);
666	count = __cgroup_task_count(cgrp);
667	spin_unlock_irq(lock: &css_set_lock);
668
669	return count;
670	}
671
672	static struct cgroup *kn_priv(struct kernfs_node *kn)
673	{
674	struct kernfs_node *parent;
675	/*
676	* The parent can not be replaced due to KERNFS_ROOT_INVARIANT_PARENT.
677	* Therefore it is always safe to dereference this pointer outside of a
678	* RCU section.
679	*/
680	parent = rcu_dereference_check(kn->__parent,
681	kernfs_root_flags(kn) & KERNFS_ROOT_INVARIANT_PARENT);
682	return parent->priv;
683	}
684
685	struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
686	{
687	struct cgroup *cgrp = kn_priv(kn: of->kn);
688	struct cftype *cft = of_cft(of);
689
690	/*
691	* This is open and unprotected implementation of cgroup_css().
692	* seq_css() is only called from a kernfs file operation which has
693	* an active reference on the file. Because all the subsystem
694	* files are drained before a css is disassociated with a cgroup,
695	* the matching css from the cgroup's subsys table is guaranteed to
696	* be and stay valid until the enclosing operation is complete.
697	*/
698	if (CGROUP_HAS_SUBSYS_CONFIG && cft->ss)
699	return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
700	else
701	return &cgrp->self;
702	}
703	EXPORT_SYMBOL_GPL(of_css);
704
705	/**
706	* for_each_css - iterate all css's of a cgroup
707	* @css: the iteration cursor
708	* @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
709	* @cgrp: the target cgroup to iterate css's of
710	*
711	* Should be called under cgroup_mutex.
712	*/
713	#define for_each_css(css, ssid, cgrp) \
714	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
715	if (!((css) = rcu_dereference_check( \
716	(cgrp)->subsys[(ssid)], \
717	lockdep_is_held(&cgroup_mutex)))) { } \
718	else
719
720	/**
721	* do_each_subsys_mask - filter for_each_subsys with a bitmask
722	* @ss: the iteration cursor
723	* @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
724	* @ss_mask: the bitmask
725	*
726	* The block will only run for cases where the ssid-th bit (1 << ssid) of
727	* @ss_mask is set.
728	*/
729	#define do_each_subsys_mask(ss, ssid, ss_mask) do { \
730	unsigned long __ss_mask = (ss_mask); \
731	if (!CGROUP_HAS_SUBSYS_CONFIG) { \
732	(ssid) = 0; \
733	break; \
734	} \
735	for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
736	(ss) = cgroup_subsys[ssid]; \
737	{
738
739	#define while_each_subsys_mask() \
740	} \
741	} \
742	} while (false)
743
744	/* iterate over child cgrps, lock should be held throughout iteration */
745	#define cgroup_for_each_live_child(child, cgrp) \
746	list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
747	if (({ lockdep_assert_held(&cgroup_mutex); \
748	cgroup_is_dead(child); })) \
749	; \
750	else
751
752	/* walk live descendants in pre order */
753	#define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
754	css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
755	if (({ lockdep_assert_held(&cgroup_mutex); \
756	(dsct) = (d_css)->cgroup; \
757	cgroup_is_dead(dsct); })) \
758	; \
759	else
760
761	/* walk live descendants in postorder */
762	#define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
763	css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
764	if (({ lockdep_assert_held(&cgroup_mutex); \
765	(dsct) = (d_css)->cgroup; \
766	cgroup_is_dead(dsct); })) \
767	; \
768	else
769
770	/*
771	* The default css_set - used by init and its children prior to any
772	* hierarchies being mounted. It contains a pointer to the root state
773	* for each subsystem. Also used to anchor the list of css_sets. Not
774	* reference-counted, to improve performance when child cgroups
775	* haven't been created.
776	*/
777	struct css_set init_css_set = {
778	.refcount = REFCOUNT_INIT(1),
779	.dom_cset = &init_css_set,
780	.tasks = LIST_HEAD_INIT(init_css_set.tasks),
781	.mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
782	.dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks),
783	.task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
784	.threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
785	.cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
786	.mg_src_preload_node = LIST_HEAD_INIT(init_css_set.mg_src_preload_node),
787	.mg_dst_preload_node = LIST_HEAD_INIT(init_css_set.mg_dst_preload_node),
788	.mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
789
790	/*
791	* The following field is re-initialized when this cset gets linked
792	* in cgroup_init(). However, let's initialize the field
793	* statically too so that the default cgroup can be accessed safely
794	* early during boot.
795	*/
796	.dfl_cgrp = &cgrp_dfl_root.cgrp,
797	};
798
799	static int css_set_count = 1; /* 1 for init_css_set */
800
801	static bool css_set_threaded(struct css_set *cset)
802	{
803	return cset->dom_cset != cset;
804	}
805
806	/**
807	* css_set_populated - does a css_set contain any tasks?
808	* @cset: target css_set
809	*
810	* css_set_populated() should be the same as !!cset->nr_tasks at steady
811	* state. However, css_set_populated() can be called while a task is being
812	* added to or removed from the linked list before the nr_tasks is
813	* properly updated. Hence, we can't just look at ->nr_tasks here.
814	*/
815	static bool css_set_populated(struct css_set *cset)
816	{
817	lockdep_assert_held(&css_set_lock);
818
819	return !list_empty(head: &cset->tasks) || !list_empty(head: &cset->mg_tasks);
820	}
821
822	/**
823	* cgroup_update_populated - update the populated count of a cgroup
824	* @cgrp: the target cgroup
825	* @populated: inc or dec populated count
826	*
827	* One of the css_sets associated with @cgrp is either getting its first
828	* task or losing the last. Update @cgrp->nr_populated_* accordingly. The
829	* count is propagated towards root so that a given cgroup's
830	* nr_populated_children is zero iff none of its descendants contain any
831	* tasks.
832	*
833	* @cgrp's interface file "cgroup.populated" is zero if both
834	* @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
835	* 1 otherwise. When the sum changes from or to zero, userland is notified
836	* that the content of the interface file has changed. This can be used to
837	* detect when @cgrp and its descendants become populated or empty.
838	*/
839	static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
840	{
841	struct cgroup *child = NULL;
842	int adj = populated ? 1 : -1;
843
844	lockdep_assert_held(&css_set_lock);
845
846	do {
847	bool was_populated = cgroup_is_populated(cgrp);
848
849	if (!child) {
850	cgrp->nr_populated_csets += adj;
851	} else {
852	if (cgroup_is_threaded(cgrp: child))
853	cgrp->nr_populated_threaded_children += adj;
854	else
855	cgrp->nr_populated_domain_children += adj;
856	}
857
858	if (was_populated == cgroup_is_populated(cgrp))
859	break;
860
861	cgroup1_check_for_release(cgrp);
862	TRACE_CGROUP_PATH(notify_populated, cgrp,
863	cgroup_is_populated(cgrp));
864	cgroup_file_notify(cfile: &cgrp->events_file);
865
866	child = cgrp;
867	cgrp = cgroup_parent(cgrp);
868	} while (cgrp);
869	}
870
871	/**
872	* css_set_update_populated - update populated state of a css_set
873	* @cset: target css_set
874	* @populated: whether @cset is populated or depopulated
875	*
876	* @cset is either getting the first task or losing the last. Update the
877	* populated counters of all associated cgroups accordingly.
878	*/
879	static void css_set_update_populated(struct css_set *cset, bool populated)
880	{
881	struct cgrp_cset_link *link;
882
883	lockdep_assert_held(&css_set_lock);
884
885	list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
886	cgroup_update_populated(cgrp: link->cgrp, populated);
887	}
888
889	/*
890	* @task is leaving, advance task iterators which are pointing to it so
891	* that they can resume at the next position. Advancing an iterator might
892	* remove it from the list, use safe walk. See css_task_iter_skip() for
893	* details.
894	*/
895	static void css_set_skip_task_iters(struct css_set *cset,
896	struct task_struct *task)
897	{
898	struct css_task_iter *it, *pos;
899
900	list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
901	css_task_iter_skip(it, task);
902	}
903
904	/**
905	* css_set_move_task - move a task from one css_set to another
906	* @task: task being moved
907	* @from_cset: css_set @task currently belongs to (may be NULL)
908	* @to_cset: new css_set @task is being moved to (may be NULL)
909	* @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
910	*
911	* Move @task from @from_cset to @to_cset. If @task didn't belong to any
912	* css_set, @from_cset can be NULL. If @task is being disassociated
913	* instead of moved, @to_cset can be NULL.
914	*
915	* This function automatically handles populated counter updates and
916	* css_task_iter adjustments but the caller is responsible for managing
917	* @from_cset and @to_cset's reference counts.
918	*/
919	static void css_set_move_task(struct task_struct *task,
920	struct css_set *from_cset, struct css_set *to_cset,
921	bool use_mg_tasks)
922	{
923	lockdep_assert_held(&css_set_lock);
924
925	if (to_cset && !css_set_populated(cset: to_cset))
926	css_set_update_populated(cset: to_cset, populated: true);
927
928	if (from_cset) {
929	WARN_ON_ONCE(list_empty(&task->cg_list));
930
931	css_set_skip_task_iters(cset: from_cset, task);
932	list_del_init(entry: &task->cg_list);
933	if (!css_set_populated(cset: from_cset))
934	css_set_update_populated(cset: from_cset, populated: false);
935	} else {
936	WARN_ON_ONCE(!list_empty(&task->cg_list));
937	}
938
939	if (to_cset) {
940	/*
941	* We are synchronized through cgroup_threadgroup_rwsem
942	* against PF_EXITING setting such that we can't race
943	* against cgroup_task_dead()/cgroup_task_free() dropping
944	* the css_set.
945	*/
946	WARN_ON_ONCE(task->flags & PF_EXITING);
947
948	cgroup_move_task(p: task, to: to_cset);
949	list_add_tail(new: &task->cg_list, head: use_mg_tasks ? &to_cset->mg_tasks :
950	&to_cset->tasks);
951	}
952	}
953
954	/*
955	* hash table for cgroup groups. This improves the performance to find
956	* an existing css_set. This hash doesn't (currently) take into
957	* account cgroups in empty hierarchies.
958	*/
959	#define CSS_SET_HASH_BITS 7
960	static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
961
962	static unsigned long css_set_hash(struct cgroup_subsys_state **css)
963	{
964	unsigned long key = 0UL;
965	struct cgroup_subsys *ss;
966	int i;
967
968	for_each_subsys(ss, i)
969	key += (unsigned long)css[i];
970	key = (key >> 16) ^ key;
971
972	return key;
973	}
974
975	void put_css_set_locked(struct css_set *cset)
976	{
977	struct cgrp_cset_link *link, *tmp_link;
978	struct cgroup_subsys *ss;
979	int ssid;
980
981	lockdep_assert_held(&css_set_lock);
982
983	if (!refcount_dec_and_test(r: &cset->refcount))
984	return;
985
986	WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
987
988	/* This css_set is dead. Unlink it and release cgroup and css refs */
989	for_each_subsys(ss, ssid) {
990	list_del(entry: &cset->e_cset_node[ssid]);
991	css_put(cset->subsys[ssid]);
992	}
993	hash_del(node: &cset->hlist);
994	css_set_count--;
995
996	list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
997	list_del(entry: &link->cset_link);
998	list_del(entry: &link->cgrp_link);
999	if (cgroup_parent(cgrp: link->cgrp))
1000	cgroup_put(cgrp: link->cgrp);
1001	kfree(objp: link);
1002	}
1003
1004	if (css_set_threaded(cset)) {
1005	list_del(entry: &cset->threaded_csets_node);
1006	put_css_set_locked(cset: cset->dom_cset);
1007	}
1008
1009	kfree_rcu(cset, rcu_head);
1010	}
1011
1012	/**
1013	* compare_css_sets - helper function for find_existing_css_set().
1014	* @cset: candidate css_set being tested
1015	* @old_cset: existing css_set for a task
1016	* @new_cgrp: cgroup that's being entered by the task
1017	* @template: desired set of css pointers in css_set (pre-calculated)
1018	*
1019	* Returns true if "cset" matches "old_cset" except for the hierarchy
1020	* which "new_cgrp" belongs to, for which it should match "new_cgrp".
1021	*/
1022	static bool compare_css_sets(struct css_set *cset,
1023	struct css_set *old_cset,
1024	struct cgroup *new_cgrp,
1025	struct cgroup_subsys_state *template[])
1026	{
1027	struct cgroup *new_dfl_cgrp;
1028	struct list_head *l1, *l2;
1029
1030	/*
1031	* On the default hierarchy, there can be csets which are
1032	* associated with the same set of cgroups but different csses.
1033	* Let's first ensure that csses match.
1034	*/
1035	if (memcmp(p: template, q: cset->subsys, size: sizeof(cset->subsys)))
1036	return false;
1037
1038
1039	/* @cset's domain should match the default cgroup's */
1040	if (cgroup_on_dfl(cgrp: new_cgrp))
1041	new_dfl_cgrp = new_cgrp;
1042	else
1043	new_dfl_cgrp = old_cset->dfl_cgrp;
1044
1045	if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
1046	return false;
1047
1048	/*
1049	* Compare cgroup pointers in order to distinguish between
1050	* different cgroups in hierarchies. As different cgroups may
1051	* share the same effective css, this comparison is always
1052	* necessary.
1053	*/
1054	l1 = &cset->cgrp_links;
1055	l2 = &old_cset->cgrp_links;
1056	while (1) {
1057	struct cgrp_cset_link *link1, *link2;
1058	struct cgroup *cgrp1, *cgrp2;
1059
1060	l1 = l1->next;
1061	l2 = l2->next;
1062	/* See if we reached the end - both lists are equal length. */
1063	if (l1 == &cset->cgrp_links) {
1064	BUG_ON(l2 != &old_cset->cgrp_links);
1065	break;
1066	} else {
1067	BUG_ON(l2 == &old_cset->cgrp_links);
1068	}
1069	/* Locate the cgroups associated with these links. */
1070	link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1071	link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1072	cgrp1 = link1->cgrp;
1073	cgrp2 = link2->cgrp;
1074	/* Hierarchies should be linked in the same order. */
1075	BUG_ON(cgrp1->root != cgrp2->root);
1076
1077	/*
1078	* If this hierarchy is the hierarchy of the cgroup
1079	* that's changing, then we need to check that this
1080	* css_set points to the new cgroup; if it's any other
1081	* hierarchy, then this css_set should point to the
1082	* same cgroup as the old css_set.
1083	*/
1084	if (cgrp1->root == new_cgrp->root) {
1085	if (cgrp1 != new_cgrp)
1086	return false;
1087	} else {
1088	if (cgrp1 != cgrp2)
1089	return false;
1090	}
1091	}
1092	return true;
1093	}
1094
1095	/**
1096	* find_existing_css_set - init css array and find the matching css_set
1097	* @old_cset: the css_set that we're using before the cgroup transition
1098	* @cgrp: the cgroup that we're moving into
1099	* @template: out param for the new set of csses, should be clear on entry
1100	*/
1101	static struct css_set *find_existing_css_set(struct css_set *old_cset,
1102	struct cgroup *cgrp,
1103	struct cgroup_subsys_state **template)
1104	{
1105	struct cgroup_root *root = cgrp->root;
1106	struct cgroup_subsys *ss;
1107	struct css_set *cset;
1108	unsigned long key;
1109	int i;
1110
1111	/*
1112	* Build the set of subsystem state objects that we want to see in the
1113	* new css_set. While subsystems can change globally, the entries here
1114	* won't change, so no need for locking.
1115	*/
1116	for_each_subsys(ss, i) {
1117	if (root->subsys_mask & (1UL << i)) {
1118	/*
1119	* @ss is in this hierarchy, so we want the
1120	* effective css from @cgrp.
1121	*/
1122	template[i] = cgroup_e_css_by_mask(cgrp, ss);
1123	} else {
1124	/*
1125	* @ss is not in this hierarchy, so we don't want
1126	* to change the css.
1127	*/
1128	template[i] = old_cset->subsys[i];
1129	}
1130	}
1131
1132	key = css_set_hash(css: template);
1133	hash_for_each_possible(css_set_table, cset, hlist, key) {
1134	if (!compare_css_sets(cset, old_cset, new_cgrp: cgrp, template))
1135	continue;
1136
1137	/* This css_set matches what we need */
1138	return cset;
1139	}
1140
1141	/* No existing cgroup group matched */
1142	return NULL;
1143	}
1144
1145	static void free_cgrp_cset_links(struct list_head *links_to_free)
1146	{
1147	struct cgrp_cset_link *link, *tmp_link;
1148
1149	list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1150	list_del(entry: &link->cset_link);
1151	kfree(objp: link);
1152	}
1153	}
1154
1155	/**
1156	* allocate_cgrp_cset_links - allocate cgrp_cset_links
1157	* @count: the number of links to allocate
1158	* @tmp_links: list_head the allocated links are put on
1159	*
1160	* Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1161	* through ->cset_link. Returns 0 on success or -errno.
1162	*/
1163	static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1164	{
1165	struct cgrp_cset_link *link;
1166	int i;
1167
1168	INIT_LIST_HEAD(list: tmp_links);
1169
1170	for (i = 0; i < count; i++) {
1171	link = kzalloc(sizeof(*link), GFP_KERNEL);
1172	if (!link) {
1173	free_cgrp_cset_links(links_to_free: tmp_links);
1174	return -ENOMEM;
1175	}
1176	list_add(new: &link->cset_link, head: tmp_links);
1177	}
1178	return 0;
1179	}
1180
1181	/**
1182	* link_css_set - a helper function to link a css_set to a cgroup
1183	* @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1184	* @cset: the css_set to be linked
1185	* @cgrp: the destination cgroup
1186	*/
1187	static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1188	struct cgroup *cgrp)
1189	{
1190	struct cgrp_cset_link *link;
1191
1192	BUG_ON(list_empty(tmp_links));
1193
1194	if (cgroup_on_dfl(cgrp))
1195	cset->dfl_cgrp = cgrp;
1196
1197	link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1198	link->cset = cset;
1199	link->cgrp = cgrp;
1200
1201	/*
1202	* Always add links to the tail of the lists so that the lists are
1203	* in chronological order.
1204	*/
1205	list_move_tail(list: &link->cset_link, head: &cgrp->cset_links);
1206	list_add_tail(new: &link->cgrp_link, head: &cset->cgrp_links);
1207
1208	if (cgroup_parent(cgrp))
1209	cgroup_get_live(cgrp);
1210	}
1211
1212	/**
1213	* find_css_set - return a new css_set with one cgroup updated
1214	* @old_cset: the baseline css_set
1215	* @cgrp: the cgroup to be updated
1216	*
1217	* Return a new css_set that's equivalent to @old_cset, but with @cgrp
1218	* substituted into the appropriate hierarchy.
1219	*/
1220	static struct css_set *find_css_set(struct css_set *old_cset,
1221	struct cgroup *cgrp)
1222	{
1223	struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1224	struct css_set *cset;
1225	struct list_head tmp_links;
1226	struct cgrp_cset_link *link;
1227	struct cgroup_subsys *ss;
1228	unsigned long key;
1229	int ssid;
1230
1231	lockdep_assert_held(&cgroup_mutex);
1232
1233	/* First see if we already have a cgroup group that matches
1234	* the desired set */
1235	spin_lock_irq(lock: &css_set_lock);
1236	cset = find_existing_css_set(old_cset, cgrp, template);
1237	if (cset)
1238	get_css_set(cset);
1239	spin_unlock_irq(lock: &css_set_lock);
1240
1241	if (cset)
1242	return cset;
1243
1244	cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1245	if (!cset)
1246	return NULL;
1247
1248	/* Allocate all the cgrp_cset_link objects that we'll need */
1249	if (allocate_cgrp_cset_links(count: cgroup_root_count, tmp_links: &tmp_links) < 0) {
1250	kfree(objp: cset);
1251	return NULL;
1252	}
1253
1254	refcount_set(r: &cset->refcount, n: 1);
1255	cset->dom_cset = cset;
1256	INIT_LIST_HEAD(list: &cset->tasks);
1257	INIT_LIST_HEAD(list: &cset->mg_tasks);
1258	INIT_LIST_HEAD(list: &cset->dying_tasks);
1259	INIT_LIST_HEAD(list: &cset->task_iters);
1260	INIT_LIST_HEAD(list: &cset->threaded_csets);
1261	INIT_HLIST_NODE(h: &cset->hlist);
1262	INIT_LIST_HEAD(list: &cset->cgrp_links);
1263	INIT_LIST_HEAD(list: &cset->mg_src_preload_node);
1264	INIT_LIST_HEAD(list: &cset->mg_dst_preload_node);
1265	INIT_LIST_HEAD(list: &cset->mg_node);
1266
1267	/* Copy the set of subsystem state objects generated in
1268	* find_existing_css_set() */
1269	memcpy(cset->subsys, template, sizeof(cset->subsys));
1270
1271	spin_lock_irq(lock: &css_set_lock);
1272	/* Add reference counts and links from the new css_set. */
1273	list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1274	struct cgroup *c = link->cgrp;
1275
1276	if (c->root == cgrp->root)
1277	c = cgrp;
1278	link_css_set(tmp_links: &tmp_links, cset, cgrp: c);
1279	}
1280
1281	BUG_ON(!list_empty(&tmp_links));
1282
1283	css_set_count++;
1284
1285	/* Add @cset to the hash table */
1286	key = css_set_hash(css: cset->subsys);
1287	hash_add(css_set_table, &cset->hlist, key);
1288
1289	for_each_subsys(ss, ssid) {
1290	struct cgroup_subsys_state *css = cset->subsys[ssid];
1291
1292	list_add_tail(new: &cset->e_cset_node[ssid],
1293	head: &css->cgroup->e_csets[ssid]);
1294	css_get(css);
1295	}
1296
1297	spin_unlock_irq(lock: &css_set_lock);
1298
1299	/*
1300	* If @cset should be threaded, look up the matching dom_cset and
1301	* link them up. We first fully initialize @cset then look for the
1302	* dom_cset. It's simpler this way and safe as @cset is guaranteed
1303	* to stay empty until we return.
1304	*/
1305	if (cgroup_is_threaded(cgrp: cset->dfl_cgrp)) {
1306	struct css_set *dcset;
1307
1308	dcset = find_css_set(old_cset: cset, cgrp: cset->dfl_cgrp->dom_cgrp);
1309	if (!dcset) {
1310	put_css_set(cset);
1311	return NULL;
1312	}
1313
1314	spin_lock_irq(lock: &css_set_lock);
1315	cset->dom_cset = dcset;
1316	list_add_tail(new: &cset->threaded_csets_node,
1317	head: &dcset->threaded_csets);
1318	spin_unlock_irq(lock: &css_set_lock);
1319	}
1320
1321	return cset;
1322	}
1323
1324	struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1325	{
1326	struct cgroup *root_cgrp = kernfs_root_to_node(root: kf_root)->priv;
1327
1328	return root_cgrp->root;
1329	}
1330
1331	void cgroup_favor_dynmods(struct cgroup_root *root, bool favor)
1332	{
1333	bool favoring = root->flags & CGRP_ROOT_FAVOR_DYNMODS;
1334
1335	/*
1336	* see the comment above CGRP_ROOT_FAVOR_DYNMODS definition.
1337	* favordynmods can flip while task is between
1338	* cgroup_threadgroup_change_begin() and end(), so down_write global
1339	* cgroup_threadgroup_rwsem to synchronize them.
1340	*
1341	* Once cgroup_enable_per_threadgroup_rwsem is enabled, holding
1342	* cgroup_threadgroup_rwsem doesn't exlude tasks between
1343	* cgroup_thread_group_change_begin() and end() and thus it's unsafe to
1344	* turn off. As the scenario is unlikely, simply disallow disabling once
1345	* enabled and print out a warning.
1346	*/
1347	percpu_down_write(&cgroup_threadgroup_rwsem);
1348	if (favor && !favoring) {
1349	cgroup_enable_per_threadgroup_rwsem = true;
1350	rcu_sync_enter(&cgroup_threadgroup_rwsem.rss);
1351	root->flags |= CGRP_ROOT_FAVOR_DYNMODS;
1352	} else if (!favor && favoring) {
1353	if (cgroup_enable_per_threadgroup_rwsem)
1354	pr_warn_once("cgroup favordynmods: per threadgroup rwsem mechanism can't be disabled\n");
1355	rcu_sync_exit(&cgroup_threadgroup_rwsem.rss);
1356	root->flags &= ~CGRP_ROOT_FAVOR_DYNMODS;
1357	}
1358	percpu_up_write(&cgroup_threadgroup_rwsem);
1359	}
1360
1361	static int cgroup_init_root_id(struct cgroup_root *root)
1362	{
1363	int id;
1364
1365	lockdep_assert_held(&cgroup_mutex);
1366
1367	id = idr_alloc_cyclic(&cgroup_hierarchy_idr, ptr: root, start: 0, end: 0, GFP_KERNEL);
1368	if (id < 0)
1369	return id;
1370
1371	root->hierarchy_id = id;
1372	return 0;
1373	}
1374
1375	static void cgroup_exit_root_id(struct cgroup_root *root)
1376	{
1377	lockdep_assert_held(&cgroup_mutex);
1378
1379	idr_remove(&cgroup_hierarchy_idr, id: root->hierarchy_id);
1380	}
1381
1382	void cgroup_free_root(struct cgroup_root *root)
1383	{
1384	kfree_rcu(root, rcu);
1385	}
1386
1387	static void cgroup_destroy_root(struct cgroup_root *root)
1388	{
1389	struct cgroup *cgrp = &root->cgrp;
1390	struct cgrp_cset_link *link, *tmp_link;
1391	int ret;
1392
1393	trace_cgroup_destroy_root(root);
1394
1395	cgroup_lock_and_drain_offline(cgrp: &cgrp_dfl_root.cgrp);
1396
1397	BUG_ON(atomic_read(&root->nr_cgrps));
1398	BUG_ON(!list_empty(&cgrp->self.children));
1399
1400	ret = blocking_notifier_call_chain(nh: &cgroup_lifetime_notifier,
1401	val: CGROUP_LIFETIME_OFFLINE, v: cgrp);
1402	WARN_ON_ONCE(notifier_to_errno(ret));
1403
1404	/* Rebind all subsystems back to the default hierarchy */
1405	WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1406
1407	/*
1408	* Release all the links from cset_links to this hierarchy's
1409	* root cgroup
1410	*/
1411	spin_lock_irq(lock: &css_set_lock);
1412
1413	list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1414	list_del(entry: &link->cset_link);
1415	list_del(entry: &link->cgrp_link);
1416	kfree(objp: link);
1417	}
1418
1419	spin_unlock_irq(lock: &css_set_lock);
1420
1421	WARN_ON_ONCE(list_empty(&root->root_list));
1422	list_del_rcu(entry: &root->root_list);
1423	cgroup_root_count--;
1424
1425	if (!have_favordynmods)
1426	cgroup_favor_dynmods(root, favor: false);
1427
1428	cgroup_exit_root_id(root);
1429
1430	cgroup_unlock();
1431
1432	kernfs_destroy_root(root: root->kf_root);
1433	cgroup_free_root(root);
1434	}
1435
1436	/*
1437	* Returned cgroup is without refcount but it's valid as long as cset pins it.
1438	*/
1439	static inline struct cgroup *__cset_cgroup_from_root(struct css_set *cset,
1440	struct cgroup_root *root)
1441	{
1442	struct cgroup *res_cgroup = NULL;
1443
1444	if (cset == &init_css_set) {
1445	res_cgroup = &root->cgrp;
1446	} else if (root == &cgrp_dfl_root) {
1447	res_cgroup = cset->dfl_cgrp;
1448	} else {
1449	struct cgrp_cset_link *link;
1450	lockdep_assert_held(&css_set_lock);
1451
1452	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1453	struct cgroup *c = link->cgrp;
1454
1455	if (c->root == root) {
1456	res_cgroup = c;
1457	break;
1458	}
1459	}
1460	}
1461
1462	/*
1463	* If cgroup_mutex is not held, the cgrp_cset_link will be freed
1464	* before we remove the cgroup root from the root_list. Consequently,
1465	* when accessing a cgroup root, the cset_link may have already been
1466	* freed, resulting in a NULL res_cgroup. However, by holding the
1467	* cgroup_mutex, we ensure that res_cgroup can't be NULL.
1468	* If we don't hold cgroup_mutex in the caller, we must do the NULL
1469	* check.
1470	*/
1471	return res_cgroup;
1472	}
1473
1474	/*
1475	* look up cgroup associated with current task's cgroup namespace on the
1476	* specified hierarchy
1477	*/
1478	static struct cgroup *
1479	current_cgns_cgroup_from_root(struct cgroup_root *root)
1480	{
1481	struct cgroup *res = NULL;
1482	struct css_set *cset;
1483
1484	lockdep_assert_held(&css_set_lock);
1485
1486	rcu_read_lock();
1487
1488	cset = current->nsproxy->cgroup_ns->root_cset;
1489	res = __cset_cgroup_from_root(cset, root);
1490
1491	rcu_read_unlock();
1492
1493	/*
1494	* The namespace_sem is held by current, so the root cgroup can't
1495	* be umounted. Therefore, we can ensure that the res is non-NULL.
1496	*/
1497	WARN_ON_ONCE(!res);
1498	return res;
1499	}
1500
1501	/*
1502	* Look up cgroup associated with current task's cgroup namespace on the default
1503	* hierarchy.
1504	*
1505	* Unlike current_cgns_cgroup_from_root(), this doesn't need locks:
1506	* - Internal rcu_read_lock is unnecessary because we don't dereference any rcu
1507	* pointers.
1508	* - css_set_lock is not needed because we just read cset->dfl_cgrp.
1509	* - As a bonus returned cgrp is pinned with the current because it cannot
1510	* switch cgroup_ns asynchronously.
1511	*/
1512	static struct cgroup *current_cgns_cgroup_dfl(void)
1513	{
1514	struct css_set *cset;
1515
1516	if (current->nsproxy) {
1517	cset = current->nsproxy->cgroup_ns->root_cset;
1518	return __cset_cgroup_from_root(cset, root: &cgrp_dfl_root);
1519	} else {
1520	/*
1521	* NOTE: This function may be called from bpf_cgroup_from_id()
1522	* on a task which has already passed exit_nsproxy_namespaces()
1523	* and nsproxy == NULL. Fall back to cgrp_dfl_root which will
1524	* make all cgroups visible for lookups.
1525	*/
1526	return &cgrp_dfl_root.cgrp;
1527	}
1528	}
1529
1530	/* look up cgroup associated with given css_set on the specified hierarchy */
1531	static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1532	struct cgroup_root *root)
1533	{
1534	lockdep_assert_held(&css_set_lock);
1535
1536	return __cset_cgroup_from_root(cset, root);
1537	}
1538
1539	/*
1540	* Return the cgroup for "task" from the given hierarchy. Must be
1541	* called with css_set_lock held to prevent task's groups from being modified.
1542	* Must be called with either cgroup_mutex or rcu read lock to prevent the
1543	* cgroup root from being destroyed.
1544	*/
1545	struct cgroup *task_cgroup_from_root(struct task_struct *task,
1546	struct cgroup_root *root)
1547	{
1548	/*
1549	* No need to lock the task - since we hold css_set_lock the
1550	* task can't change groups.
1551	*/
1552	return cset_cgroup_from_root(cset: task_css_set(task), root);
1553	}
1554
1555	/*
1556	* A task must hold cgroup_mutex to modify cgroups.
1557	*
1558	* Any task can increment and decrement the count field without lock.
1559	* So in general, code holding cgroup_mutex can't rely on the count
1560	* field not changing. However, if the count goes to zero, then only
1561	* cgroup_attach_task() can increment it again. Because a count of zero
1562	* means that no tasks are currently attached, therefore there is no
1563	* way a task attached to that cgroup can fork (the other way to
1564	* increment the count). So code holding cgroup_mutex can safely
1565	* assume that if the count is zero, it will stay zero. Similarly, if
1566	* a task holds cgroup_mutex on a cgroup with zero count, it
1567	* knows that the cgroup won't be removed, as cgroup_rmdir()
1568	* needs that mutex.
1569	*
1570	* A cgroup can only be deleted if both its 'count' of using tasks
1571	* is zero, and its list of 'children' cgroups is empty. Since all
1572	* tasks in the system use _some_ cgroup, and since there is always at
1573	* least one task in the system (init, pid == 1), therefore, root cgroup
1574	* always has either children cgroups and/or using tasks. So we don't
1575	* need a special hack to ensure that root cgroup cannot be deleted.
1576	*
1577	* P.S. One more locking exception. RCU is used to guard the
1578	* update of a tasks cgroup pointer by cgroup_attach_task()
1579	*/
1580
1581	static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1582
1583	static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1584	char *buf)
1585	{
1586	struct cgroup_subsys *ss = cft->ss;
1587
1588	if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1589	!(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1590	const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1591
1592	snprintf(buf, CGROUP_FILE_NAME_MAX, fmt: "%s%s.%s",
1593	dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1594	cft->name);
1595	} else {
1596	strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1597	}
1598	return buf;
1599	}
1600
1601	/**
1602	* cgroup_file_mode - deduce file mode of a control file
1603	* @cft: the control file in question
1604	*
1605	* S_IRUGO for read, S_IWUSR for write.
1606	*/
1607	static umode_t cgroup_file_mode(const struct cftype *cft)
1608	{
1609	umode_t mode = 0;
1610
1611	if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1612	mode |= S_IRUGO;
1613
1614	if (cft->write_u64 || cft->write_s64 || cft->write) {
1615	if (cft->flags & CFTYPE_WORLD_WRITABLE)
1616	mode |= S_IWUGO;
1617	else
1618	mode |= S_IWUSR;
1619	}
1620
1621	return mode;
1622	}
1623
1624	/**
1625	* cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1626	* @subtree_control: the new subtree_control mask to consider
1627	* @this_ss_mask: available subsystems
1628	*
1629	* On the default hierarchy, a subsystem may request other subsystems to be
1630	* enabled together through its ->depends_on mask. In such cases, more
1631	* subsystems than specified in "cgroup.subtree_control" may be enabled.
1632	*
1633	* This function calculates which subsystems need to be enabled if
1634	* @subtree_control is to be applied while restricted to @this_ss_mask.
1635	*/
1636	static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1637	{
1638	u16 cur_ss_mask = subtree_control;
1639	struct cgroup_subsys *ss;
1640	int ssid;
1641
1642	lockdep_assert_held(&cgroup_mutex);
1643
1644	cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1645
1646	while (true) {
1647	u16 new_ss_mask = cur_ss_mask;
1648
1649	do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1650	new_ss_mask |= ss->depends_on;
1651	} while_each_subsys_mask();
1652
1653	/*
1654	* Mask out subsystems which aren't available. This can
1655	* happen only if some depended-upon subsystems were bound
1656	* to non-default hierarchies.
1657	*/
1658	new_ss_mask &= this_ss_mask;
1659
1660	if (new_ss_mask == cur_ss_mask)
1661	break;
1662	cur_ss_mask = new_ss_mask;
1663	}
1664
1665	return cur_ss_mask;
1666	}
1667
1668	/**
1669	* cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1670	* @kn: the kernfs_node being serviced
1671	*
1672	* This helper undoes cgroup_kn_lock_live() and should be invoked before
1673	* the method finishes if locking succeeded. Note that once this function
1674	* returns the cgroup returned by cgroup_kn_lock_live() may become
1675	* inaccessible any time. If the caller intends to continue to access the
1676	* cgroup, it should pin it before invoking this function.
1677	*/
1678	void cgroup_kn_unlock(struct kernfs_node *kn)
1679	{
1680	struct cgroup *cgrp;
1681
1682	if (kernfs_type(kn) == KERNFS_DIR)
1683	cgrp = kn->priv;
1684	else
1685	cgrp = kn_priv(kn);
1686
1687	cgroup_unlock();
1688
1689	kernfs_unbreak_active_protection(kn);
1690	cgroup_put(cgrp);
1691	}
1692
1693	/**
1694	* cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1695	* @kn: the kernfs_node being serviced
1696	* @drain_offline: perform offline draining on the cgroup
1697	*
1698	* This helper is to be used by a cgroup kernfs method currently servicing
1699	* @kn. It breaks the active protection, performs cgroup locking and
1700	* verifies that the associated cgroup is alive. Returns the cgroup if
1701	* alive; otherwise, %NULL. A successful return should be undone by a
1702	* matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1703	* cgroup is drained of offlining csses before return.
1704	*
1705	* Any cgroup kernfs method implementation which requires locking the
1706	* associated cgroup should use this helper. It avoids nesting cgroup
1707	* locking under kernfs active protection and allows all kernfs operations
1708	* including self-removal.
1709	*/
1710	struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1711	{
1712	struct cgroup *cgrp;
1713
1714	if (kernfs_type(kn) == KERNFS_DIR)
1715	cgrp = kn->priv;
1716	else
1717	cgrp = kn_priv(kn);
1718
1719	/*
1720	* We're gonna grab cgroup_mutex which nests outside kernfs
1721	* active_ref. cgroup liveliness check alone provides enough
1722	* protection against removal. Ensure @cgrp stays accessible and
1723	* break the active_ref protection.
1724	*/
1725	if (!cgroup_tryget(cgrp))
1726	return NULL;
1727	kernfs_break_active_protection(kn);
1728
1729	if (drain_offline)
1730	cgroup_lock_and_drain_offline(cgrp);
1731	else
1732	cgroup_lock();
1733
1734	if (!cgroup_is_dead(cgrp))
1735	return cgrp;
1736
1737	cgroup_kn_unlock(kn);
1738	return NULL;
1739	}
1740
1741	static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1742	{
1743	char name[CGROUP_FILE_NAME_MAX];
1744
1745	lockdep_assert_held(&cgroup_mutex);
1746
1747	if (cft->file_offset) {
1748	struct cgroup_subsys_state *css = cgroup_css(cgrp, ss: cft->ss);
1749	struct cgroup_file *cfile = (void *)css + cft->file_offset;
1750
1751	spin_lock_irq(lock: &cgroup_file_kn_lock);
1752	cfile->kn = NULL;
1753	spin_unlock_irq(lock: &cgroup_file_kn_lock);
1754
1755	timer_delete_sync(timer: &cfile->notify_timer);
1756	}
1757
1758	kernfs_remove_by_name(parent: cgrp->kn, name: cgroup_file_name(cgrp, cft, buf: name));
1759	}
1760
1761	/**
1762	* css_clear_dir - remove subsys files in a cgroup directory
1763	* @css: target css
1764	*/
1765	static void css_clear_dir(struct cgroup_subsys_state *css)
1766	{
1767	struct cgroup *cgrp = css->cgroup;
1768	struct cftype *cfts;
1769
1770	if (!(css->flags & CSS_VISIBLE))
1771	return;
1772
1773	css->flags &= ~CSS_VISIBLE;
1774
1775	if (css_is_self(css)) {
1776	if (cgroup_on_dfl(cgrp)) {
1777	cgroup_addrm_files(css, cgrp,
1778	cfts: cgroup_base_files, is_add: false);
1779	if (cgroup_psi_enabled())
1780	cgroup_addrm_files(css, cgrp,
1781	cfts: cgroup_psi_files, is_add: false);
1782	} else {
1783	cgroup_addrm_files(css, cgrp,
1784	cfts: cgroup1_base_files, is_add: false);
1785	}
1786	} else {
1787	list_for_each_entry(cfts, &css->ss->cfts, node)
1788	cgroup_addrm_files(css, cgrp, cfts, is_add: false);
1789	}
1790	}
1791
1792	/**
1793	* css_populate_dir - create subsys files in a cgroup directory
1794	* @css: target css
1795	*
1796	* On failure, no file is added.
1797	*/
1798	static int css_populate_dir(struct cgroup_subsys_state *css)
1799	{
1800	struct cgroup *cgrp = css->cgroup;
1801	struct cftype *cfts, *failed_cfts;
1802	int ret;
1803
1804	if (css->flags & CSS_VISIBLE)
1805	return 0;
1806
1807	if (css_is_self(css)) {
1808	if (cgroup_on_dfl(cgrp)) {
1809	ret = cgroup_addrm_files(css, cgrp,
1810	cfts: cgroup_base_files, is_add: true);
1811	if (ret < 0)
1812	return ret;
1813
1814	if (cgroup_psi_enabled()) {
1815	ret = cgroup_addrm_files(css, cgrp,
1816	cfts: cgroup_psi_files, is_add: true);
1817	if (ret < 0) {
1818	cgroup_addrm_files(css, cgrp,
1819	cfts: cgroup_base_files, is_add: false);
1820	return ret;
1821	}
1822	}
1823	} else {
1824	ret = cgroup_addrm_files(css, cgrp,
1825	cfts: cgroup1_base_files, is_add: true);
1826	if (ret < 0)
1827	return ret;
1828	}
1829	} else {
1830	list_for_each_entry(cfts, &css->ss->cfts, node) {
1831	ret = cgroup_addrm_files(css, cgrp, cfts, is_add: true);
1832	if (ret < 0) {
1833	failed_cfts = cfts;
1834	goto err;
1835	}
1836	}
1837	}
1838
1839	css->flags |= CSS_VISIBLE;
1840
1841	return 0;
1842	err:
1843	list_for_each_entry(cfts, &css->ss->cfts, node) {
1844	if (cfts == failed_cfts)
1845	break;
1846	cgroup_addrm_files(css, cgrp, cfts, is_add: false);
1847	}
1848	return ret;
1849	}
1850
1851	int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1852	{
1853	struct cgroup *dcgrp = &dst_root->cgrp;
1854	struct cgroup_subsys *ss;
1855	int ssid, ret;
1856	u16 dfl_disable_ss_mask = 0;
1857
1858	lockdep_assert_held(&cgroup_mutex);
1859
1860	do_each_subsys_mask(ss, ssid, ss_mask) {
1861	/*
1862	* If @ss has non-root csses attached to it, can't move.
1863	* If @ss is an implicit controller, it is exempt from this
1864	* rule and can be stolen.
1865	*/
1866	if (css_next_child(NULL, parent: cgroup_css(cgrp: &ss->root->cgrp, ss)) &&
1867	!ss->implicit_on_dfl)
1868	return -EBUSY;
1869
1870	/* can't move between two non-dummy roots either */
1871	if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1872	return -EBUSY;
1873
1874	/*
1875	* Collect ssid's that need to be disabled from default
1876	* hierarchy.
1877	*/
1878	if (ss->root == &cgrp_dfl_root)
1879	dfl_disable_ss_mask |= 1 << ssid;
1880
1881	} while_each_subsys_mask();
1882
1883	if (dfl_disable_ss_mask) {
1884	struct cgroup *scgrp = &cgrp_dfl_root.cgrp;
1885
1886	/*
1887	* Controllers from default hierarchy that need to be rebound
1888	* are all disabled together in one go.
1889	*/
1890	cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask;
1891	WARN_ON(cgroup_apply_control(scgrp));
1892	cgroup_finalize_control(cgrp: scgrp, ret: 0);
1893	}
1894
1895	do_each_subsys_mask(ss, ssid, ss_mask) {
1896	struct cgroup_root *src_root = ss->root;
1897	struct cgroup *scgrp = &src_root->cgrp;
1898	struct cgroup_subsys_state *css = cgroup_css(cgrp: scgrp, ss);
1899	struct css_set *cset, *cset_pos;
1900	struct css_task_iter *it;
1901
1902	WARN_ON(!css || cgroup_css(dcgrp, ss));
1903
1904	if (src_root != &cgrp_dfl_root) {
1905	/* disable from the source */
1906	src_root->subsys_mask &= ~(1 << ssid);
1907	WARN_ON(cgroup_apply_control(scgrp));
1908	cgroup_finalize_control(cgrp: scgrp, ret: 0);
1909	}
1910
1911	/* rebind */
1912	RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1913	rcu_assign_pointer(dcgrp->subsys[ssid], css);
1914	ss->root = dst_root;
1915
1916	spin_lock_irq(lock: &css_set_lock);
1917	css->cgroup = dcgrp;
1918	WARN_ON(!list_empty(&dcgrp->e_csets[ss->id]));
1919	list_for_each_entry_safe(cset, cset_pos, &scgrp->e_csets[ss->id],
1920	e_cset_node[ss->id]) {
1921	list_move_tail(list: &cset->e_cset_node[ss->id],
1922	head: &dcgrp->e_csets[ss->id]);
1923	/*
1924	* all css_sets of scgrp together in same order to dcgrp,
1925	* patch in-flight iterators to preserve correct iteration.
1926	* since the iterator is always advanced right away and
1927	* finished when it->cset_pos meets it->cset_head, so only
1928	* update it->cset_head is enough here.
1929	*/
1930	list_for_each_entry(it, &cset->task_iters, iters_node)
1931	if (it->cset_head == &scgrp->e_csets[ss->id])
1932	it->cset_head = &dcgrp->e_csets[ss->id];
1933	}
1934	spin_unlock_irq(lock: &css_set_lock);
1935
1936	/* default hierarchy doesn't enable controllers by default */
1937	dst_root->subsys_mask |= 1 << ssid;
1938	if (dst_root == &cgrp_dfl_root) {
1939	static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1940	} else {
1941	dcgrp->subtree_control |= 1 << ssid;
1942	static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1943	}
1944
1945	ret = cgroup_apply_control(cgrp: dcgrp);
1946	if (ret)
1947	pr_warn("partial failure to rebind %s controller (err=%d)\n",
1948	ss->name, ret);
1949
1950	if (ss->bind)
1951	ss->bind(css);
1952	} while_each_subsys_mask();
1953
1954	kernfs_activate(kn: dcgrp->kn);
1955	return 0;
1956	}
1957
1958	int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1959	struct kernfs_root *kf_root)
1960	{
1961	int len = 0;
1962	char *buf = NULL;
1963	struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1964	struct cgroup *ns_cgroup;
1965
1966	buf = kmalloc(PATH_MAX, GFP_KERNEL);
1967	if (!buf)
1968	return -ENOMEM;
1969
1970	spin_lock_irq(lock: &css_set_lock);
1971	ns_cgroup = current_cgns_cgroup_from_root(root: kf_cgroot);
1972	len = kernfs_path_from_node(kn_to: kf_node, kn_from: ns_cgroup->kn, buf, PATH_MAX);
1973	spin_unlock_irq(lock: &css_set_lock);
1974
1975	if (len == -E2BIG)
1976	len = -ERANGE;
1977	else if (len > 0) {
1978	seq_escape(m: sf, s: buf, esc: " \t\n\\");
1979	len = 0;
1980	}
1981	kfree(objp: buf);
1982	return len;
1983	}
1984
1985	enum cgroup2_param {
1986	Opt_nsdelegate,
1987	Opt_favordynmods,
1988	Opt_memory_localevents,
1989	Opt_memory_recursiveprot,
1990	Opt_memory_hugetlb_accounting,
1991	Opt_pids_localevents,
1992	nr__cgroup2_params
1993	};
1994
1995	static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1996	fsparam_flag("nsdelegate", Opt_nsdelegate),
1997	fsparam_flag("favordynmods", Opt_favordynmods),
1998	fsparam_flag("memory_localevents", Opt_memory_localevents),
1999	fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot),
2000	fsparam_flag("memory_hugetlb_accounting", Opt_memory_hugetlb_accounting),
2001	fsparam_flag("pids_localevents", Opt_pids_localevents),
2002	{}
2003	};
2004
2005	static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
2006	{
2007	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2008	struct fs_parse_result result;
2009	int opt;
2010
2011	opt = fs_parse(fc, desc: cgroup2_fs_parameters, param, result: &result);
2012	if (opt < 0)
2013	return opt;
2014
2015	switch (opt) {
2016	case Opt_nsdelegate:
2017	ctx->flags |= CGRP_ROOT_NS_DELEGATE;
2018	return 0;
2019	case Opt_favordynmods:
2020	ctx->flags |= CGRP_ROOT_FAVOR_DYNMODS;
2021	return 0;
2022	case Opt_memory_localevents:
2023	ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
2024	return 0;
2025	case Opt_memory_recursiveprot:
2026	ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
2027	return 0;
2028	case Opt_memory_hugetlb_accounting:
2029	ctx->flags |= CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING;
2030	return 0;
2031	case Opt_pids_localevents:
2032	ctx->flags |= CGRP_ROOT_PIDS_LOCAL_EVENTS;
2033	return 0;
2034	}
2035	return -EINVAL;
2036	}
2037
2038	struct cgroup_of_peak *of_peak(struct kernfs_open_file *of)
2039	{
2040	struct cgroup_file_ctx *ctx = of->priv;
2041
2042	return &ctx->peak;
2043	}
2044
2045	static void apply_cgroup_root_flags(unsigned int root_flags)
2046	{
2047	if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
2048	if (root_flags & CGRP_ROOT_NS_DELEGATE)
2049	cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
2050	else
2051	cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
2052
2053	cgroup_favor_dynmods(root: &cgrp_dfl_root,
2054	favor: root_flags & CGRP_ROOT_FAVOR_DYNMODS);
2055
2056	if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
2057	cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
2058	else
2059	cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
2060
2061	if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
2062	cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
2063	else
2064	cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
2065
2066	if (root_flags & CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING)
2067	cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING;
2068	else
2069	cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING;
2070
2071	if (root_flags & CGRP_ROOT_PIDS_LOCAL_EVENTS)
2072	cgrp_dfl_root.flags |= CGRP_ROOT_PIDS_LOCAL_EVENTS;
2073	else
2074	cgrp_dfl_root.flags &= ~CGRP_ROOT_PIDS_LOCAL_EVENTS;
2075	}
2076	}
2077
2078	static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
2079	{
2080	if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
2081	seq_puts(m: seq, s: ",nsdelegate");
2082	if (cgrp_dfl_root.flags & CGRP_ROOT_FAVOR_DYNMODS)
2083	seq_puts(m: seq, s: ",favordynmods");
2084	if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
2085	seq_puts(m: seq, s: ",memory_localevents");
2086	if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
2087	seq_puts(m: seq, s: ",memory_recursiveprot");
2088	if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING)
2089	seq_puts(m: seq, s: ",memory_hugetlb_accounting");
2090	if (cgrp_dfl_root.flags & CGRP_ROOT_PIDS_LOCAL_EVENTS)
2091	seq_puts(m: seq, s: ",pids_localevents");
2092	return 0;
2093	}
2094
2095	static int cgroup_reconfigure(struct fs_context *fc)
2096	{
2097	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2098
2099	apply_cgroup_root_flags(root_flags: ctx->flags);
2100	return 0;
2101	}
2102
2103	static void init_cgroup_housekeeping(struct cgroup *cgrp)
2104	{
2105	struct cgroup_subsys *ss;
2106	int ssid;
2107
2108	INIT_LIST_HEAD(list: &cgrp->self.sibling);
2109	INIT_LIST_HEAD(list: &cgrp->self.children);
2110	INIT_LIST_HEAD(list: &cgrp->cset_links);
2111	INIT_LIST_HEAD(list: &cgrp->pidlists);
2112	mutex_init(&cgrp->pidlist_mutex);
2113	cgrp->self.cgroup = cgrp;
2114	cgrp->self.flags |= CSS_ONLINE;
2115	cgrp->dom_cgrp = cgrp;
2116	cgrp->max_descendants = INT_MAX;
2117	cgrp->max_depth = INT_MAX;
2118	prev_cputime_init(prev: &cgrp->prev_cputime);
2119
2120	for_each_subsys(ss, ssid)
2121	INIT_LIST_HEAD(list: &cgrp->e_csets[ssid]);
2122
2123	#ifdef CONFIG_CGROUP_BPF
2124	for (int i = 0; i < ARRAY_SIZE(cgrp->bpf.revisions); i++)
2125	cgrp->bpf.revisions[i] = 1;
2126	#endif
2127
2128	init_waitqueue_head(&cgrp->offline_waitq);
2129	INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
2130	}
2131
2132	void init_cgroup_root(struct cgroup_fs_context *ctx)
2133	{
2134	struct cgroup_root *root = ctx->root;
2135	struct cgroup *cgrp = &root->cgrp;
2136
2137	INIT_LIST_HEAD_RCU(list: &root->root_list);
2138	atomic_set(v: &root->nr_cgrps, i: 1);
2139	cgrp->root = root;
2140	init_cgroup_housekeeping(cgrp);
2141
2142	/* DYNMODS must be modified through cgroup_favor_dynmods() */
2143	root->flags = ctx->flags & ~CGRP_ROOT_FAVOR_DYNMODS;
2144	if (ctx->release_agent)
2145	strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
2146	if (ctx->name)
2147	strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
2148	if (ctx->cpuset_clone_children)
2149	set_bit(nr: CGRP_CPUSET_CLONE_CHILDREN, addr: &root->cgrp.flags);
2150	}
2151
2152	int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
2153	{
2154	LIST_HEAD(tmp_links);
2155	struct cgroup *root_cgrp = &root->cgrp;
2156	struct kernfs_syscall_ops *kf_sops;
2157	struct css_set *cset;
2158	int i, ret;
2159
2160	lockdep_assert_held(&cgroup_mutex);
2161
2162	ret = percpu_ref_init(ref: &root_cgrp->self.refcnt, release: css_release,
2163	flags: 0, GFP_KERNEL);
2164	if (ret)
2165	goto out;
2166
2167	/*
2168	* We're accessing css_set_count without locking css_set_lock here,
2169	* but that's OK - it can only be increased by someone holding
2170	* cgroup_lock, and that's us. Later rebinding may disable
2171	* controllers on the default hierarchy and thus create new csets,
2172	* which can't be more than the existing ones. Allocate 2x.
2173	*/
2174	ret = allocate_cgrp_cset_links(count: 2 * css_set_count, tmp_links: &tmp_links);
2175	if (ret)
2176	goto cancel_ref;
2177
2178	ret = cgroup_init_root_id(root);
2179	if (ret)
2180	goto cancel_ref;
2181
2182	kf_sops = root == &cgrp_dfl_root ?
2183	&cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
2184
2185	root->kf_root = kernfs_create_root(scops: kf_sops,
2186	flags: KERNFS_ROOT_CREATE_DEACTIVATED |
2187	KERNFS_ROOT_SUPPORT_EXPORTOP |
2188	KERNFS_ROOT_SUPPORT_USER_XATTR |
2189	KERNFS_ROOT_INVARIANT_PARENT,
2190	priv: root_cgrp);
2191	if (IS_ERR(ptr: root->kf_root)) {
2192	ret = PTR_ERR(ptr: root->kf_root);
2193	goto exit_root_id;
2194	}
2195	root_cgrp->kn = kernfs_root_to_node(root: root->kf_root);
2196	WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
2197	root_cgrp->ancestors[0] = root_cgrp;
2198
2199	ret = css_populate_dir(css: &root_cgrp->self);
2200	if (ret)
2201	goto destroy_root;
2202
2203	ret = css_rstat_init(css: &root_cgrp->self);
2204	if (ret)
2205	goto destroy_root;
2206
2207	ret = rebind_subsystems(dst_root: root, ss_mask);
2208	if (ret)
2209	goto exit_stats;
2210
2211	ret = blocking_notifier_call_chain(nh: &cgroup_lifetime_notifier,
2212	val: CGROUP_LIFETIME_ONLINE, v: root_cgrp);
2213	WARN_ON_ONCE(notifier_to_errno(ret));
2214
2215	trace_cgroup_setup_root(root);
2216
2217	/*
2218	* There must be no failure case after here, since rebinding takes
2219	* care of subsystems' refcounts, which are explicitly dropped in
2220	* the failure exit path.
2221	*/
2222	list_add_rcu(new: &root->root_list, head: &cgroup_roots);
2223	cgroup_root_count++;
2224
2225	/*
2226	* Link the root cgroup in this hierarchy into all the css_set
2227	* objects.
2228	*/
2229	spin_lock_irq(lock: &css_set_lock);
2230	hash_for_each(css_set_table, i, cset, hlist) {
2231	link_css_set(tmp_links: &tmp_links, cset, cgrp: root_cgrp);
2232	if (css_set_populated(cset))
2233	cgroup_update_populated(cgrp: root_cgrp, populated: true);
2234	}
2235	spin_unlock_irq(lock: &css_set_lock);
2236
2237	BUG_ON(!list_empty(&root_cgrp->self.children));
2238	BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2239
2240	ret = 0;
2241	goto out;
2242
2243	exit_stats:
2244	css_rstat_exit(css: &root_cgrp->self);
2245	destroy_root:
2246	kernfs_destroy_root(root: root->kf_root);
2247	root->kf_root = NULL;
2248	exit_root_id:
2249	cgroup_exit_root_id(root);
2250	cancel_ref:
2251	percpu_ref_exit(ref: &root_cgrp->self.refcnt);
2252	out:
2253	free_cgrp_cset_links(links_to_free: &tmp_links);
2254	return ret;
2255	}
2256
2257	int cgroup_do_get_tree(struct fs_context *fc)
2258	{
2259	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2260	int ret;
2261
2262	ctx->kfc.root = ctx->root->kf_root;
2263	if (fc->fs_type == &cgroup2_fs_type)
2264	ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2265	else
2266	ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2267	ret = kernfs_get_tree(fc);
2268
2269	/*
2270	* In non-init cgroup namespace, instead of root cgroup's dentry,
2271	* we return the dentry corresponding to the cgroupns->root_cgrp.
2272	*/
2273	if (!ret && ctx->ns != &init_cgroup_ns) {
2274	struct dentry *nsdentry;
2275	struct super_block *sb = fc->root->d_sb;
2276	struct cgroup *cgrp;
2277
2278	cgroup_lock();
2279	spin_lock_irq(lock: &css_set_lock);
2280
2281	cgrp = cset_cgroup_from_root(cset: ctx->ns->root_cset, root: ctx->root);
2282
2283	spin_unlock_irq(lock: &css_set_lock);
2284	cgroup_unlock();
2285
2286	nsdentry = kernfs_node_dentry(kn: cgrp->kn, sb);
2287	dput(fc->root);
2288	if (IS_ERR(ptr: nsdentry)) {
2289	deactivate_locked_super(sb);
2290	ret = PTR_ERR(ptr: nsdentry);
2291	nsdentry = NULL;
2292	}
2293	fc->root = nsdentry;
2294	}
2295
2296	if (!ctx->kfc.new_sb_created)
2297	cgroup_put(cgrp: &ctx->root->cgrp);
2298
2299	return ret;
2300	}
2301
2302	/*
2303	* Destroy a cgroup filesystem context.
2304	*/
2305	static void cgroup_fs_context_free(struct fs_context *fc)
2306	{
2307	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2308
2309	kfree(objp: ctx->name);
2310	kfree(objp: ctx->release_agent);
2311	put_cgroup_ns(ns: ctx->ns);
2312	kernfs_free_fs_context(fc);
2313	kfree(objp: ctx);
2314	}
2315
2316	static int cgroup_get_tree(struct fs_context *fc)
2317	{
2318	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2319	int ret;
2320
2321	WRITE_ONCE(cgrp_dfl_visible, true);
2322	cgroup_get_live(cgrp: &cgrp_dfl_root.cgrp);
2323	ctx->root = &cgrp_dfl_root;
2324
2325	ret = cgroup_do_get_tree(fc);
2326	if (!ret)
2327	apply_cgroup_root_flags(root_flags: ctx->flags);
2328	return ret;
2329	}
2330
2331	static const struct fs_context_operations cgroup_fs_context_ops = {
2332	.free = cgroup_fs_context_free,
2333	.parse_param = cgroup2_parse_param,
2334	.get_tree = cgroup_get_tree,
2335	.reconfigure = cgroup_reconfigure,
2336	};
2337
2338	static const struct fs_context_operations cgroup1_fs_context_ops = {
2339	.free = cgroup_fs_context_free,
2340	.parse_param = cgroup1_parse_param,
2341	.get_tree = cgroup1_get_tree,
2342	.reconfigure = cgroup1_reconfigure,
2343	};
2344
2345	/*
2346	* Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2347	* we select the namespace we're going to use.
2348	*/
2349	static int cgroup_init_fs_context(struct fs_context *fc)
2350	{
2351	struct cgroup_fs_context *ctx;
2352
2353	ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2354	if (!ctx)
2355	return -ENOMEM;
2356
2357	ctx->ns = current->nsproxy->cgroup_ns;
2358	get_cgroup_ns(ns: ctx->ns);
2359	fc->fs_private = &ctx->kfc;
2360	if (fc->fs_type == &cgroup2_fs_type)
2361	fc->ops = &cgroup_fs_context_ops;
2362	else
2363	fc->ops = &cgroup1_fs_context_ops;
2364	put_user_ns(ns: fc->user_ns);
2365	fc->user_ns = get_user_ns(ns: ctx->ns->user_ns);
2366	fc->global = true;
2367
2368	if (have_favordynmods)
2369	ctx->flags |= CGRP_ROOT_FAVOR_DYNMODS;
2370
2371	return 0;
2372	}
2373
2374	static void cgroup_kill_sb(struct super_block *sb)
2375	{
2376	struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2377	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2378
2379	/*
2380	* If @root doesn't have any children, start killing it.
2381	* This prevents new mounts by disabling percpu_ref_tryget_live().
2382	*
2383	* And don't kill the default root.
2384	*/
2385	if (list_empty(head: &root->cgrp.self.children) && root != &cgrp_dfl_root &&
2386	!percpu_ref_is_dying(ref: &root->cgrp.self.refcnt))
2387	percpu_ref_kill(ref: &root->cgrp.self.refcnt);
2388	cgroup_put(cgrp: &root->cgrp);
2389	kernfs_kill_sb(sb);
2390	}
2391
2392	struct file_system_type cgroup_fs_type = {
2393	.name = "cgroup",
2394	.init_fs_context = cgroup_init_fs_context,
2395	.parameters = cgroup1_fs_parameters,
2396	.kill_sb = cgroup_kill_sb,
2397	.fs_flags = FS_USERNS_MOUNT,
2398	};
2399
2400	static struct file_system_type cgroup2_fs_type = {
2401	.name = "cgroup2",
2402	.init_fs_context = cgroup_init_fs_context,
2403	.parameters = cgroup2_fs_parameters,
2404	.kill_sb = cgroup_kill_sb,
2405	.fs_flags = FS_USERNS_MOUNT,
2406	};
2407
2408	#ifdef CONFIG_CPUSETS_V1
2409	enum cpuset_param {
2410	Opt_cpuset_v2_mode,
2411	};
2412
2413	static const struct fs_parameter_spec cpuset_fs_parameters[] = {
2414	fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode),
2415	{}
2416	};
2417
2418	static int cpuset_parse_param(struct fs_context *fc, struct fs_parameter *param)
2419	{
2420	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2421	struct fs_parse_result result;
2422	int opt;
2423
2424	opt = fs_parse(fc, desc: cpuset_fs_parameters, param, result: &result);
2425	if (opt < 0)
2426	return opt;
2427
2428	switch (opt) {
2429	case Opt_cpuset_v2_mode:
2430	ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
2431	return 0;
2432	}
2433	return -EINVAL;
2434	}
2435
2436	static const struct fs_context_operations cpuset_fs_context_ops = {
2437	.get_tree = cgroup1_get_tree,
2438	.free = cgroup_fs_context_free,
2439	.parse_param = cpuset_parse_param,
2440	};
2441
2442	/*
2443	* This is ugly, but preserves the userspace API for existing cpuset
2444	* users. If someone tries to mount the "cpuset" filesystem, we
2445	* silently switch it to mount "cgroup" instead
2446	*/
2447	static int cpuset_init_fs_context(struct fs_context *fc)
2448	{
2449	char *agent = kstrdup(s: "/sbin/cpuset_release_agent", GFP_USER);
2450	struct cgroup_fs_context *ctx;
2451	int err;
2452
2453	err = cgroup_init_fs_context(fc);
2454	if (err) {
2455	kfree(objp: agent);
2456	return err;
2457	}
2458
2459	fc->ops = &cpuset_fs_context_ops;
2460
2461	ctx = cgroup_fc2context(fc);
2462	ctx->subsys_mask = 1 << cpuset_cgrp_id;
2463	ctx->flags |= CGRP_ROOT_NOPREFIX;
2464	ctx->release_agent = agent;
2465
2466	get_filesystem(fs: &cgroup_fs_type);
2467	put_filesystem(fs: fc->fs_type);
2468	fc->fs_type = &cgroup_fs_type;
2469
2470	return 0;
2471	}
2472
2473	static struct file_system_type cpuset_fs_type = {
2474	.name = "cpuset",
2475	.init_fs_context = cpuset_init_fs_context,
2476	.parameters = cpuset_fs_parameters,
2477	.fs_flags = FS_USERNS_MOUNT,
2478	};
2479	#endif
2480
2481	int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2482	struct cgroup_namespace *ns)
2483	{
2484	struct cgroup *root = cset_cgroup_from_root(cset: ns->root_cset, root: cgrp->root);
2485
2486	return kernfs_path_from_node(kn_to: cgrp->kn, kn_from: root->kn, buf, buflen);
2487	}
2488
2489	int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2490	struct cgroup_namespace *ns)
2491	{
2492	int ret;
2493
2494	cgroup_lock();
2495	spin_lock_irq(lock: &css_set_lock);
2496
2497	ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2498
2499	spin_unlock_irq(lock: &css_set_lock);
2500	cgroup_unlock();
2501
2502	return ret;
2503	}
2504	EXPORT_SYMBOL_GPL(cgroup_path_ns);
2505
2506	/**
2507	* cgroup_attach_lock - Lock for ->attach()
2508	* @lock_mode: whether acquire and acquire which rwsem
2509	* @tsk: thread group to lock
2510	*
2511	* cgroup migration sometimes needs to stabilize threadgroups against forks and
2512	* exits by write-locking cgroup_threadgroup_rwsem. However, some ->attach()
2513	* implementations (e.g. cpuset), also need to disable CPU hotplug.
2514	* Unfortunately, letting ->attach() operations acquire cpus_read_lock() can
2515	* lead to deadlocks.
2516	*
2517	* Bringing up a CPU may involve creating and destroying tasks which requires
2518	* read-locking threadgroup_rwsem, so threadgroup_rwsem nests inside
2519	* cpus_read_lock(). If we call an ->attach() which acquires the cpus lock while
2520	* write-locking threadgroup_rwsem, the locking order is reversed and we end up
2521	* waiting for an on-going CPU hotplug operation which in turn is waiting for
2522	* the threadgroup_rwsem to be released to create new tasks. For more details:
2523	*
2524	* http://lkml.kernel.org/r/20220711174629.uehfmqegcwn2lqzu@wubuntu
2525	*
2526	* Resolve the situation by always acquiring cpus_read_lock() before optionally
2527	* write-locking cgroup_threadgroup_rwsem. This allows ->attach() to assume that
2528	* CPU hotplug is disabled on entry.
2529	*
2530	* When favordynmods is enabled, take per threadgroup rwsem to reduce overhead
2531	* on dynamic cgroup modifications. see the comment above
2532	* CGRP_ROOT_FAVOR_DYNMODS definition.
2533	*
2534	* tsk is not NULL only when writing to cgroup.procs.
2535	*/
2536	void cgroup_attach_lock(enum cgroup_attach_lock_mode lock_mode,
2537	struct task_struct *tsk)
2538	{
2539	cpus_read_lock();
2540
2541	switch (lock_mode) {
2542	case CGRP_ATTACH_LOCK_NONE:
2543	break;
2544	case CGRP_ATTACH_LOCK_GLOBAL:
2545	percpu_down_write(&cgroup_threadgroup_rwsem);
2546	break;
2547	case CGRP_ATTACH_LOCK_PER_THREADGROUP:
2548	down_write(sem: &tsk->signal->cgroup_threadgroup_rwsem);
2549	break;
2550	default:
2551	pr_warn("cgroup: Unexpected attach lock mode.");
2552	break;
2553	}
2554	}
2555
2556	/**
2557	* cgroup_attach_unlock - Undo cgroup_attach_lock()
2558	* @lock_mode: whether release and release which rwsem
2559	* @tsk: thread group to lock
2560	*/
2561	void cgroup_attach_unlock(enum cgroup_attach_lock_mode lock_mode,
2562	struct task_struct *tsk)
2563	{
2564	switch (lock_mode) {
2565	case CGRP_ATTACH_LOCK_NONE:
2566	break;
2567	case CGRP_ATTACH_LOCK_GLOBAL:
2568	percpu_up_write(&cgroup_threadgroup_rwsem);
2569	break;
2570	case CGRP_ATTACH_LOCK_PER_THREADGROUP:
2571	up_write(sem: &tsk->signal->cgroup_threadgroup_rwsem);
2572	break;
2573	default:
2574	pr_warn("cgroup: Unexpected attach lock mode.");
2575	break;
2576	}
2577
2578	cpus_read_unlock();
2579	}
2580
2581	/**
2582	* cgroup_migrate_add_task - add a migration target task to a migration context
2583	* @task: target task
2584	* @mgctx: target migration context
2585	*
2586	* Add @task, which is a migration target, to @mgctx->tset. This function
2587	* becomes noop if @task doesn't need to be migrated. @task's css_set
2588	* should have been added as a migration source and @task->cg_list will be
2589	* moved from the css_set's tasks list to mg_tasks one.
2590	*/
2591	static void cgroup_migrate_add_task(struct task_struct *task,
2592	struct cgroup_mgctx *mgctx)
2593	{
2594	struct css_set *cset;
2595
2596	lockdep_assert_held(&css_set_lock);
2597
2598	/* @task either already exited or can't exit until the end */
2599	if (task->flags & PF_EXITING)
2600	return;
2601
2602	/* cgroup_threadgroup_rwsem protects racing against forks */
2603	WARN_ON_ONCE(list_empty(&task->cg_list));
2604
2605	cset = task_css_set(task);
2606	if (!cset->mg_src_cgrp)
2607	return;
2608
2609	mgctx->tset.nr_tasks++;
2610
2611	list_move_tail(list: &task->cg_list, head: &cset->mg_tasks);
2612	if (list_empty(head: &cset->mg_node))
2613	list_add_tail(new: &cset->mg_node,
2614	head: &mgctx->tset.src_csets);
2615	if (list_empty(head: &cset->mg_dst_cset->mg_node))
2616	list_add_tail(new: &cset->mg_dst_cset->mg_node,
2617	head: &mgctx->tset.dst_csets);
2618	}
2619
2620	/**
2621	* cgroup_taskset_first - reset taskset and return the first task
2622	* @tset: taskset of interest
2623	* @dst_cssp: output variable for the destination css
2624	*
2625	* @tset iteration is initialized and the first task is returned.
2626	*/
2627	struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2628	struct cgroup_subsys_state **dst_cssp)
2629	{
2630	tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2631	tset->cur_task = NULL;
2632
2633	return cgroup_taskset_next(tset, dst_cssp);
2634	}
2635
2636	/**
2637	* cgroup_taskset_next - iterate to the next task in taskset
2638	* @tset: taskset of interest
2639	* @dst_cssp: output variable for the destination css
2640	*
2641	* Return the next task in @tset. Iteration must have been initialized
2642	* with cgroup_taskset_first().
2643	*/
2644	struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2645	struct cgroup_subsys_state **dst_cssp)
2646	{
2647	struct css_set *cset = tset->cur_cset;
2648	struct task_struct *task = tset->cur_task;
2649
2650	while (CGROUP_HAS_SUBSYS_CONFIG && &cset->mg_node != tset->csets) {
2651	if (!task)
2652	task = list_first_entry(&cset->mg_tasks,
2653	struct task_struct, cg_list);
2654	else
2655	task = list_next_entry(task, cg_list);
2656
2657	if (&task->cg_list != &cset->mg_tasks) {
2658	tset->cur_cset = cset;
2659	tset->cur_task = task;
2660
2661	/*
2662	* This function may be called both before and
2663	* after cgroup_migrate_execute(). The two cases
2664	* can be distinguished by looking at whether @cset
2665	* has its ->mg_dst_cset set.
2666	*/
2667	if (cset->mg_dst_cset)
2668	*dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2669	else
2670	*dst_cssp = cset->subsys[tset->ssid];
2671
2672	return task;
2673	}
2674
2675	cset = list_next_entry(cset, mg_node);
2676	task = NULL;
2677	}
2678
2679	return NULL;
2680	}
2681
2682	/**
2683	* cgroup_migrate_execute - migrate a taskset
2684	* @mgctx: migration context
2685	*
2686	* Migrate tasks in @mgctx as setup by migration preparation functions.
2687	* This function fails iff one of the ->can_attach callbacks fails and
2688	* guarantees that either all or none of the tasks in @mgctx are migrated.
2689	* @mgctx is consumed regardless of success.
2690	*/
2691	static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2692	{
2693	struct cgroup_taskset *tset = &mgctx->tset;
2694	struct cgroup_subsys *ss;
2695	struct task_struct *task, *tmp_task;
2696	struct css_set *cset, *tmp_cset;
2697	int ssid, failed_ssid, ret;
2698
2699	/* check that we can legitimately attach to the cgroup */
2700	if (tset->nr_tasks) {
2701	do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2702	if (ss->can_attach) {
2703	tset->ssid = ssid;
2704	ret = ss->can_attach(tset);
2705	if (ret) {
2706	failed_ssid = ssid;
2707	goto out_cancel_attach;
2708	}
2709	}
2710	} while_each_subsys_mask();
2711	}
2712
2713	/*
2714	* Now that we're guaranteed success, proceed to move all tasks to
2715	* the new cgroup. There are no failure cases after here, so this
2716	* is the commit point.
2717	*/
2718	spin_lock_irq(lock: &css_set_lock);
2719	list_for_each_entry(cset, &tset->src_csets, mg_node) {
2720	list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2721	struct css_set *from_cset = task_css_set(task);
2722	struct css_set *to_cset = cset->mg_dst_cset;
2723
2724	get_css_set(cset: to_cset);
2725	to_cset->nr_tasks++;
2726	css_set_move_task(task, from_cset, to_cset, use_mg_tasks: true);
2727	from_cset->nr_tasks--;
2728	/*
2729	* If the source or destination cgroup is frozen,
2730	* the task might require to change its state.
2731	*/
2732	cgroup_freezer_migrate_task(task, src: from_cset->dfl_cgrp,
2733	dst: to_cset->dfl_cgrp);
2734	put_css_set_locked(cset: from_cset);
2735
2736	}
2737	}
2738	spin_unlock_irq(lock: &css_set_lock);
2739
2740	/*
2741	* Migration is committed, all target tasks are now on dst_csets.
2742	* Nothing is sensitive to fork() after this point. Notify
2743	* controllers that migration is complete.
2744	*/
2745	tset->csets = &tset->dst_csets;
2746
2747	if (tset->nr_tasks) {
2748	do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2749	if (ss->attach) {
2750	tset->ssid = ssid;
2751	ss->attach(tset);
2752	}
2753	} while_each_subsys_mask();
2754	}
2755
2756	ret = 0;
2757	goto out_release_tset;
2758
2759	out_cancel_attach:
2760	if (tset->nr_tasks) {
2761	do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2762	if (ssid == failed_ssid)
2763	break;
2764	if (ss->cancel_attach) {
2765	tset->ssid = ssid;
2766	ss->cancel_attach(tset);
2767	}
2768	} while_each_subsys_mask();
2769	}
2770	out_release_tset:
2771	spin_lock_irq(lock: &css_set_lock);
2772	list_splice_init(list: &tset->dst_csets, head: &tset->src_csets);
2773	list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2774	list_splice_tail_init(list: &cset->mg_tasks, head: &cset->tasks);
2775	list_del_init(entry: &cset->mg_node);
2776	}
2777	spin_unlock_irq(lock: &css_set_lock);
2778
2779	/*
2780	* Re-initialize the cgroup_taskset structure in case it is reused
2781	* again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2782	* iteration.
2783	*/
2784	tset->nr_tasks = 0;
2785	tset->csets = &tset->src_csets;
2786	return ret;
2787	}
2788
2789	/**
2790	* cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2791	* @dst_cgrp: destination cgroup to test
2792	*
2793	* On the default hierarchy, except for the mixable, (possible) thread root
2794	* and threaded cgroups, subtree_control must be zero for migration
2795	* destination cgroups with tasks so that child cgroups don't compete
2796	* against tasks.
2797	*/
2798	int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2799	{
2800	/* v1 doesn't have any restriction */
2801	if (!cgroup_on_dfl(cgrp: dst_cgrp))
2802	return 0;
2803
2804	/* verify @dst_cgrp can host resources */
2805	if (!cgroup_is_valid_domain(cgrp: dst_cgrp->dom_cgrp))
2806	return -EOPNOTSUPP;
2807
2808	/*
2809	* If @dst_cgrp is already or can become a thread root or is
2810	* threaded, it doesn't matter.
2811	*/
2812	if (cgroup_can_be_thread_root(cgrp: dst_cgrp) || cgroup_is_threaded(cgrp: dst_cgrp))
2813	return 0;
2814
2815	/* apply no-internal-process constraint */
2816	if (dst_cgrp->subtree_control)
2817	return -EBUSY;
2818
2819	return 0;
2820	}
2821
2822	/**
2823	* cgroup_migrate_finish - cleanup after attach
2824	* @mgctx: migration context
2825	*
2826	* Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2827	* those functions for details.
2828	*/
2829	void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2830	{
2831	struct css_set *cset, *tmp_cset;
2832
2833	lockdep_assert_held(&cgroup_mutex);
2834
2835	spin_lock_irq(lock: &css_set_lock);
2836
2837	list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_src_csets,
2838	mg_src_preload_node) {
2839	cset->mg_src_cgrp = NULL;
2840	cset->mg_dst_cgrp = NULL;
2841	cset->mg_dst_cset = NULL;
2842	list_del_init(entry: &cset->mg_src_preload_node);
2843	put_css_set_locked(cset);
2844	}
2845
2846	list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_dst_csets,
2847	mg_dst_preload_node) {
2848	cset->mg_src_cgrp = NULL;
2849	cset->mg_dst_cgrp = NULL;
2850	cset->mg_dst_cset = NULL;
2851	list_del_init(entry: &cset->mg_dst_preload_node);
2852	put_css_set_locked(cset);
2853	}
2854
2855	spin_unlock_irq(lock: &css_set_lock);
2856	}
2857
2858	/**
2859	* cgroup_migrate_add_src - add a migration source css_set
2860	* @src_cset: the source css_set to add
2861	* @dst_cgrp: the destination cgroup
2862	* @mgctx: migration context
2863	*
2864	* Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2865	* @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2866	* up by cgroup_migrate_finish().
2867	*
2868	* This function may be called without holding cgroup_threadgroup_rwsem
2869	* even if the target is a process. Threads may be created and destroyed
2870	* but as long as cgroup_mutex is not dropped, no new css_set can be put
2871	* into play and the preloaded css_sets are guaranteed to cover all
2872	* migrations.
2873	*/
2874	void cgroup_migrate_add_src(struct css_set *src_cset,
2875	struct cgroup *dst_cgrp,
2876	struct cgroup_mgctx *mgctx)
2877	{
2878	struct cgroup *src_cgrp;
2879
2880	lockdep_assert_held(&cgroup_mutex);
2881	lockdep_assert_held(&css_set_lock);
2882
2883	/*
2884	* If ->dead, @src_set is associated with one or more dead cgroups
2885	* and doesn't contain any migratable tasks. Ignore it early so
2886	* that the rest of migration path doesn't get confused by it.
2887	*/
2888	if (src_cset->dead)
2889	return;
2890
2891	if (!list_empty(head: &src_cset->mg_src_preload_node))
2892	return;
2893
2894	src_cgrp = cset_cgroup_from_root(cset: src_cset, root: dst_cgrp->root);
2895
2896	WARN_ON(src_cset->mg_src_cgrp);
2897	WARN_ON(src_cset->mg_dst_cgrp);
2898	WARN_ON(!list_empty(&src_cset->mg_tasks));
2899	WARN_ON(!list_empty(&src_cset->mg_node));
2900
2901	src_cset->mg_src_cgrp = src_cgrp;
2902	src_cset->mg_dst_cgrp = dst_cgrp;
2903	get_css_set(cset: src_cset);
2904	list_add_tail(new: &src_cset->mg_src_preload_node, head: &mgctx->preloaded_src_csets);
2905	}
2906
2907	/**
2908	* cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2909	* @mgctx: migration context
2910	*
2911	* Tasks are about to be moved and all the source css_sets have been
2912	* preloaded to @mgctx->preloaded_src_csets. This function looks up and
2913	* pins all destination css_sets, links each to its source, and append them
2914	* to @mgctx->preloaded_dst_csets.
2915	*
2916	* This function must be called after cgroup_migrate_add_src() has been
2917	* called on each migration source css_set. After migration is performed
2918	* using cgroup_migrate(), cgroup_migrate_finish() must be called on
2919	* @mgctx.
2920	*/
2921	int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2922	{
2923	struct css_set *src_cset, *tmp_cset;
2924
2925	lockdep_assert_held(&cgroup_mutex);
2926
2927	/* look up the dst cset for each src cset and link it to src */
2928	list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2929	mg_src_preload_node) {
2930	struct css_set *dst_cset;
2931	struct cgroup_subsys *ss;
2932	int ssid;
2933
2934	dst_cset = find_css_set(old_cset: src_cset, cgrp: src_cset->mg_dst_cgrp);
2935	if (!dst_cset)
2936	return -ENOMEM;
2937
2938	WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2939
2940	/*
2941	* If src cset equals dst, it's noop. Drop the src.
2942	* cgroup_migrate() will skip the cset too. Note that we
2943	* can't handle src == dst as some nodes are used by both.
2944	*/
2945	if (src_cset == dst_cset) {
2946	src_cset->mg_src_cgrp = NULL;
2947	src_cset->mg_dst_cgrp = NULL;
2948	list_del_init(entry: &src_cset->mg_src_preload_node);
2949	put_css_set(cset: src_cset);
2950	put_css_set(cset: dst_cset);
2951	continue;
2952	}
2953
2954	src_cset->mg_dst_cset = dst_cset;
2955
2956	if (list_empty(head: &dst_cset->mg_dst_preload_node))
2957	list_add_tail(new: &dst_cset->mg_dst_preload_node,
2958	head: &mgctx->preloaded_dst_csets);
2959	else
2960	put_css_set(cset: dst_cset);
2961
2962	for_each_subsys(ss, ssid)
2963	if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2964	mgctx->ss_mask |= 1 << ssid;
2965	}
2966
2967	return 0;
2968	}
2969
2970	/**
2971	* cgroup_migrate - migrate a process or task to a cgroup
2972	* @leader: the leader of the process or the task to migrate
2973	* @threadgroup: whether @leader points to the whole process or a single task
2974	* @mgctx: migration context
2975	*
2976	* Migrate a process or task denoted by @leader. If migrating a process,
2977	* the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2978	* responsible for invoking cgroup_migrate_add_src() and
2979	* cgroup_migrate_prepare_dst() on the targets before invoking this
2980	* function and following up with cgroup_migrate_finish().
2981	*
2982	* As long as a controller's ->can_attach() doesn't fail, this function is
2983	* guaranteed to succeed. This means that, excluding ->can_attach()
2984	* failure, when migrating multiple targets, the success or failure can be
2985	* decided for all targets by invoking group_migrate_prepare_dst() before
2986	* actually starting migrating.
2987	*/
2988	int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2989	struct cgroup_mgctx *mgctx)
2990	{
2991	struct task_struct *task;
2992
2993	/*
2994	* The following thread iteration should be inside an RCU critical
2995	* section to prevent tasks from being freed while taking the snapshot.
2996	* spin_lock_irq() implies RCU critical section here.
2997	*/
2998	spin_lock_irq(lock: &css_set_lock);
2999	task = leader;
3000	do {
3001	cgroup_migrate_add_task(task, mgctx);
3002	if (!threadgroup)
3003	break;
3004	} while_each_thread(leader, task);
3005	spin_unlock_irq(lock: &css_set_lock);
3006
3007	return cgroup_migrate_execute(mgctx);
3008	}
3009
3010	/**
3011	* cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
3012	* @dst_cgrp: the cgroup to attach to
3013	* @leader: the task or the leader of the threadgroup to be attached
3014	* @threadgroup: attach the whole threadgroup?
3015	*
3016	* Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
3017	*/
3018	int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
3019	bool threadgroup)
3020	{
3021	DEFINE_CGROUP_MGCTX(mgctx);
3022	struct task_struct *task;
3023	int ret = 0;
3024
3025	/* look up all src csets */
3026	spin_lock_irq(lock: &css_set_lock);
3027	task = leader;
3028	do {
3029	cgroup_migrate_add_src(src_cset: task_css_set(task), dst_cgrp, mgctx: &mgctx);
3030	if (!threadgroup)
3031	break;
3032	} while_each_thread(leader, task);
3033	spin_unlock_irq(lock: &css_set_lock);
3034
3035	/* prepare dst csets and commit */
3036	ret = cgroup_migrate_prepare_dst(mgctx: &mgctx);
3037	if (!ret)
3038	ret = cgroup_migrate(leader, threadgroup, mgctx: &mgctx);
3039
3040	cgroup_migrate_finish(mgctx: &mgctx);
3041
3042	if (!ret)
3043	TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
3044
3045	return ret;
3046	}
3047
3048	struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
3049	enum cgroup_attach_lock_mode *lock_mode)
3050	{
3051	struct task_struct *tsk;
3052	pid_t pid;
3053
3054	if (kstrtoint(s: strstrip(str: buf), base: 0, res: &pid) || pid < 0)
3055	return ERR_PTR(error: -EINVAL);
3056
3057	retry_find_task:
3058	rcu_read_lock();
3059	if (pid) {
3060	tsk = find_task_by_vpid(nr: pid);
3061	if (!tsk) {
3062	tsk = ERR_PTR(error: -ESRCH);
3063	goto out_unlock_rcu;
3064	}
3065	} else {
3066	tsk = current;
3067	}
3068
3069	if (threadgroup)
3070	tsk = tsk->group_leader;
3071
3072	/*
3073	* kthreads may acquire PF_NO_SETAFFINITY during initialization.
3074	* If userland migrates such a kthread to a non-root cgroup, it can
3075	* become trapped in a cpuset, or RT kthread may be born in a
3076	* cgroup with no rt_runtime allocated. Just say no.
3077	*/
3078	if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
3079	tsk = ERR_PTR(error: -EINVAL);
3080	goto out_unlock_rcu;
3081	}
3082	get_task_struct(t: tsk);
3083	rcu_read_unlock();
3084
3085	/*
3086	* If we migrate a single thread, we don't care about threadgroup
3087	* stability. If the thread is `current`, it won't exit(2) under our
3088	* hands or change PID through exec(2). We exclude
3089	* cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write callers
3090	* by cgroup_mutex. Therefore, we can skip the global lock.
3091	*/
3092	lockdep_assert_held(&cgroup_mutex);
3093
3094	if (pid || threadgroup) {
3095	if (cgroup_enable_per_threadgroup_rwsem)
3096	*lock_mode = CGRP_ATTACH_LOCK_PER_THREADGROUP;
3097	else
3098	*lock_mode = CGRP_ATTACH_LOCK_GLOBAL;
3099	} else {
3100	*lock_mode = CGRP_ATTACH_LOCK_NONE;
3101	}
3102
3103	cgroup_attach_lock(lock_mode: *lock_mode, tsk);
3104
3105	if (threadgroup) {
3106	if (!thread_group_leader(p: tsk)) {
3107	/*
3108	* A race with de_thread from another thread's exec()
3109	* may strip us of our leadership. If this happens,
3110	* throw this task away and try again.
3111	*/
3112	cgroup_attach_unlock(lock_mode: *lock_mode, tsk);
3113	put_task_struct(t: tsk);
3114	goto retry_find_task;
3115	}
3116	}
3117
3118	return tsk;
3119
3120	out_unlock_rcu:
3121	rcu_read_unlock();
3122	return tsk;
3123	}
3124
3125	void cgroup_procs_write_finish(struct task_struct *task,
3126	enum cgroup_attach_lock_mode lock_mode)
3127	{
3128	cgroup_attach_unlock(lock_mode, tsk: task);
3129
3130	/* release reference from cgroup_procs_write_start() */
3131	put_task_struct(t: task);
3132	}
3133
3134	static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
3135	{
3136	struct cgroup_subsys *ss;
3137	bool printed = false;
3138	int ssid;
3139
3140	do_each_subsys_mask(ss, ssid, ss_mask) {
3141	if (printed)
3142	seq_putc(m: seq, c: ' ');
3143	seq_puts(m: seq, s: ss->name);
3144	printed = true;
3145	} while_each_subsys_mask();
3146	if (printed)
3147	seq_putc(m: seq, c: '\n');
3148	}
3149
3150	/* show controllers which are enabled from the parent */
3151	static int cgroup_controllers_show(struct seq_file *seq, void *v)
3152	{
3153	struct cgroup *cgrp = seq_css(seq)->cgroup;
3154
3155	cgroup_print_ss_mask(seq, ss_mask: cgroup_control(cgrp));
3156	return 0;
3157	}
3158
3159	/* show controllers which are enabled for a given cgroup's children */
3160	static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
3161	{
3162	struct cgroup *cgrp = seq_css(seq)->cgroup;
3163
3164	cgroup_print_ss_mask(seq, ss_mask: cgrp->subtree_control);
3165	return 0;
3166	}
3167
3168	/**
3169	* cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
3170	* @cgrp: root of the subtree to update csses for
3171	*
3172	* @cgrp's control masks have changed and its subtree's css associations
3173	* need to be updated accordingly. This function looks up all css_sets
3174	* which are attached to the subtree, creates the matching updated css_sets
3175	* and migrates the tasks to the new ones.
3176	*/
3177	static int cgroup_update_dfl_csses(struct cgroup *cgrp)
3178	{
3179	DEFINE_CGROUP_MGCTX(mgctx);
3180	struct cgroup_subsys_state *d_css;
3181	struct cgroup *dsct;
3182	struct css_set *src_cset;
3183	enum cgroup_attach_lock_mode lock_mode;
3184	bool has_tasks;
3185	int ret;
3186
3187	lockdep_assert_held(&cgroup_mutex);
3188
3189	/* look up all csses currently attached to @cgrp's subtree */
3190	spin_lock_irq(lock: &css_set_lock);
3191	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3192	struct cgrp_cset_link *link;
3193
3194	/*
3195	* As cgroup_update_dfl_csses() is only called by
3196	* cgroup_apply_control(). The csses associated with the
3197	* given cgrp will not be affected by changes made to
3198	* its subtree_control file. We can skip them.
3199	*/
3200	if (dsct == cgrp)
3201	continue;
3202
3203	list_for_each_entry(link, &dsct->cset_links, cset_link)
3204	cgroup_migrate_add_src(src_cset: link->cset, dst_cgrp: dsct, mgctx: &mgctx);
3205	}
3206	spin_unlock_irq(lock: &css_set_lock);
3207
3208	/*
3209	* We need to write-lock threadgroup_rwsem while migrating tasks.
3210	* However, if there are no source csets for @cgrp, changing its
3211	* controllers isn't gonna produce any task migrations and the
3212	* write-locking can be skipped safely.
3213	*/
3214	has_tasks = !list_empty(head: &mgctx.preloaded_src_csets);
3215
3216	if (has_tasks)
3217	lock_mode = CGRP_ATTACH_LOCK_GLOBAL;
3218	else
3219	lock_mode = CGRP_ATTACH_LOCK_NONE;
3220
3221	cgroup_attach_lock(lock_mode, NULL);
3222
3223	/* NULL dst indicates self on default hierarchy */
3224	ret = cgroup_migrate_prepare_dst(mgctx: &mgctx);
3225	if (ret)
3226	goto out_finish;
3227
3228	spin_lock_irq(lock: &css_set_lock);
3229	list_for_each_entry(src_cset, &mgctx.preloaded_src_csets,
3230	mg_src_preload_node) {
3231	struct task_struct *task, *ntask;
3232
3233	/* all tasks in src_csets need to be migrated */
3234	list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
3235	cgroup_migrate_add_task(task, mgctx: &mgctx);
3236	}
3237	spin_unlock_irq(lock: &css_set_lock);
3238
3239	ret = cgroup_migrate_execute(mgctx: &mgctx);
3240	out_finish:
3241	cgroup_migrate_finish(mgctx: &mgctx);
3242	cgroup_attach_unlock(lock_mode, NULL);
3243	return ret;
3244	}
3245
3246	/**
3247	* cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3248	* @cgrp: root of the target subtree
3249	*
3250	* Because css offlining is asynchronous, userland may try to re-enable a
3251	* controller while the previous css is still around. This function grabs
3252	* cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3253	*/
3254	void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
3255	__acquires(&cgroup_mutex)
3256	{
3257	struct cgroup *dsct;
3258	struct cgroup_subsys_state *d_css;
3259	struct cgroup_subsys *ss;
3260	int ssid;
3261
3262	restart:
3263	cgroup_lock();
3264
3265	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3266	for_each_subsys(ss, ssid) {
3267	struct cgroup_subsys_state *css = cgroup_css(cgrp: dsct, ss);
3268	DEFINE_WAIT(wait);
3269
3270	if (!css || !percpu_ref_is_dying(ref: &css->refcnt))
3271	continue;
3272
3273	cgroup_get_live(cgrp: dsct);
3274	prepare_to_wait(wq_head: &dsct->offline_waitq, wq_entry: &wait,
3275	TASK_UNINTERRUPTIBLE);
3276
3277	cgroup_unlock();
3278	schedule();
3279	finish_wait(wq_head: &dsct->offline_waitq, wq_entry: &wait);
3280
3281	cgroup_put(cgrp: dsct);
3282	goto restart;
3283	}
3284	}
3285	}
3286
3287	/**
3288	* cgroup_save_control - save control masks and dom_cgrp of a subtree
3289	* @cgrp: root of the target subtree
3290	*
3291	* Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3292	* respective old_ prefixed fields for @cgrp's subtree including @cgrp
3293	* itself.
3294	*/
3295	static void cgroup_save_control(struct cgroup *cgrp)
3296	{
3297	struct cgroup *dsct;
3298	struct cgroup_subsys_state *d_css;
3299
3300	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3301	dsct->old_subtree_control = dsct->subtree_control;
3302	dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3303	dsct->old_dom_cgrp = dsct->dom_cgrp;
3304	}
3305	}
3306
3307	/**
3308	* cgroup_propagate_control - refresh control masks of a subtree
3309	* @cgrp: root of the target subtree
3310	*
3311	* For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3312	* ->subtree_control and propagate controller availability through the
3313	* subtree so that descendants don't have unavailable controllers enabled.
3314	*/
3315	static void cgroup_propagate_control(struct cgroup *cgrp)
3316	{
3317	struct cgroup *dsct;
3318	struct cgroup_subsys_state *d_css;
3319
3320	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3321	dsct->subtree_control &= cgroup_control(cgrp: dsct);
3322	dsct->subtree_ss_mask =
3323	cgroup_calc_subtree_ss_mask(subtree_control: dsct->subtree_control,
3324	this_ss_mask: cgroup_ss_mask(cgrp: dsct));
3325	}
3326	}
3327
3328	/**
3329	* cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3330	* @cgrp: root of the target subtree
3331	*
3332	* Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3333	* respective old_ prefixed fields for @cgrp's subtree including @cgrp
3334	* itself.
3335	*/
3336	static void cgroup_restore_control(struct cgroup *cgrp)
3337	{
3338	struct cgroup *dsct;
3339	struct cgroup_subsys_state *d_css;
3340
3341	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3342	dsct->subtree_control = dsct->old_subtree_control;
3343	dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3344	dsct->dom_cgrp = dsct->old_dom_cgrp;
3345	}
3346	}
3347
3348	static bool css_visible(struct cgroup_subsys_state *css)
3349	{
3350	struct cgroup_subsys *ss = css->ss;
3351	struct cgroup *cgrp = css->cgroup;
3352
3353	if (cgroup_control(cgrp) & (1 << ss->id))
3354	return true;
3355	if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3356	return false;
3357	return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3358	}
3359
3360	/**
3361	* cgroup_apply_control_enable - enable or show csses according to control
3362	* @cgrp: root of the target subtree
3363	*
3364	* Walk @cgrp's subtree and create new csses or make the existing ones
3365	* visible. A css is created invisible if it's being implicitly enabled
3366	* through dependency. An invisible css is made visible when the userland
3367	* explicitly enables it.
3368	*
3369	* Returns 0 on success, -errno on failure. On failure, csses which have
3370	* been processed already aren't cleaned up. The caller is responsible for
3371	* cleaning up with cgroup_apply_control_disable().
3372	*/
3373	static int cgroup_apply_control_enable(struct cgroup *cgrp)
3374	{
3375	struct cgroup *dsct;
3376	struct cgroup_subsys_state *d_css;
3377	struct cgroup_subsys *ss;
3378	int ssid, ret;
3379
3380	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3381	for_each_subsys(ss, ssid) {
3382	struct cgroup_subsys_state *css = cgroup_css(cgrp: dsct, ss);
3383
3384	if (!(cgroup_ss_mask(cgrp: dsct) & (1 << ss->id)))
3385	continue;
3386
3387	if (!css) {
3388	css = css_create(cgrp: dsct, ss);
3389	if (IS_ERR(ptr: css))
3390	return PTR_ERR(ptr: css);
3391	}
3392
3393	WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3394
3395	if (css_visible(css)) {
3396	ret = css_populate_dir(css);
3397	if (ret)
3398	return ret;
3399	}
3400	}
3401	}
3402
3403	return 0;
3404	}
3405
3406	/**
3407	* cgroup_apply_control_disable - kill or hide csses according to control
3408	* @cgrp: root of the target subtree
3409	*
3410	* Walk @cgrp's subtree and kill and hide csses so that they match
3411	* cgroup_ss_mask() and cgroup_visible_mask().
3412	*
3413	* A css is hidden when the userland requests it to be disabled while other
3414	* subsystems are still depending on it. The css must not actively control
3415	* resources and be in the vanilla state if it's made visible again later.
3416	* Controllers which may be depended upon should provide ->css_reset() for
3417	* this purpose.
3418	*/
3419	static void cgroup_apply_control_disable(struct cgroup *cgrp)
3420	{
3421	struct cgroup *dsct;
3422	struct cgroup_subsys_state *d_css;
3423	struct cgroup_subsys *ss;
3424	int ssid;
3425
3426	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3427	for_each_subsys(ss, ssid) {
3428	struct cgroup_subsys_state *css = cgroup_css(cgrp: dsct, ss);
3429
3430	if (!css)
3431	continue;
3432
3433	WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3434
3435	if (css->parent &&
3436	!(cgroup_ss_mask(cgrp: dsct) & (1 << ss->id))) {
3437	kill_css(css);
3438	} else if (!css_visible(css)) {
3439	css_clear_dir(css);
3440	if (ss->css_reset)
3441	ss->css_reset(css);
3442	}
3443	}
3444	}
3445	}
3446
3447	/**
3448	* cgroup_apply_control - apply control mask updates to the subtree
3449	* @cgrp: root of the target subtree
3450	*
3451	* subsystems can be enabled and disabled in a subtree using the following
3452	* steps.
3453	*
3454	* 1. Call cgroup_save_control() to stash the current state.
3455	* 2. Update ->subtree_control masks in the subtree as desired.
3456	* 3. Call cgroup_apply_control() to apply the changes.
3457	* 4. Optionally perform other related operations.
3458	* 5. Call cgroup_finalize_control() to finish up.
3459	*
3460	* This function implements step 3 and propagates the mask changes
3461	* throughout @cgrp's subtree, updates csses accordingly and perform
3462	* process migrations.
3463	*/
3464	static int cgroup_apply_control(struct cgroup *cgrp)
3465	{
3466	int ret;
3467
3468	cgroup_propagate_control(cgrp);
3469
3470	ret = cgroup_apply_control_enable(cgrp);
3471	if (ret)
3472	return ret;
3473
3474	/*
3475	* At this point, cgroup_e_css_by_mask() results reflect the new csses
3476	* making the following cgroup_update_dfl_csses() properly update
3477	* css associations of all tasks in the subtree.
3478	*/
3479	return cgroup_update_dfl_csses(cgrp);
3480	}
3481
3482	/**
3483	* cgroup_finalize_control - finalize control mask update
3484	* @cgrp: root of the target subtree
3485	* @ret: the result of the update
3486	*
3487	* Finalize control mask update. See cgroup_apply_control() for more info.
3488	*/
3489	static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3490	{
3491	if (ret) {
3492	cgroup_restore_control(cgrp);
3493	cgroup_propagate_control(cgrp);
3494	}
3495
3496	cgroup_apply_control_disable(cgrp);
3497	}
3498
3499	static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3500	{
3501	u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3502
3503	/* if nothing is getting enabled, nothing to worry about */
3504	if (!enable)
3505	return 0;
3506
3507	/* can @cgrp host any resources? */
3508	if (!cgroup_is_valid_domain(cgrp: cgrp->dom_cgrp))
3509	return -EOPNOTSUPP;
3510
3511	/* mixables don't care */
3512	if (cgroup_is_mixable(cgrp))
3513	return 0;
3514
3515	if (domain_enable) {
3516	/* can't enable domain controllers inside a thread subtree */
3517	if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3518	return -EOPNOTSUPP;
3519	} else {
3520	/*
3521	* Threaded controllers can handle internal competitions
3522	* and are always allowed inside a (prospective) thread
3523	* subtree.
3524	*/
3525	if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3526	return 0;
3527	}
3528
3529	/*
3530	* Controllers can't be enabled for a cgroup with tasks to avoid
3531	* child cgroups competing against tasks.
3532	*/
3533	if (cgroup_has_tasks(cgrp))
3534	return -EBUSY;
3535
3536	return 0;
3537	}
3538
3539	/* change the enabled child controllers for a cgroup in the default hierarchy */
3540	static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3541	char *buf, size_t nbytes,
3542	loff_t off)
3543	{
3544	u16 enable = 0, disable = 0;
3545	struct cgroup *cgrp, *child;
3546	struct cgroup_subsys *ss;
3547	char *tok;
3548	int ssid, ret;
3549
3550	/*
3551	* Parse input - space separated list of subsystem names prefixed
3552	* with either + or -.
3553	*/
3554	buf = strstrip(str: buf);
3555	while ((tok = strsep(&buf, " "))) {
3556	if (tok[0] == '\0')
3557	continue;
3558	do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3559	if (!cgroup_ssid_enabled(ssid) ||
3560	strcmp(tok + 1, ss->name))
3561	continue;
3562
3563	if (*tok == '+') {
3564	enable |= 1 << ssid;
3565	disable &= ~(1 << ssid);
3566	} else if (*tok == '-') {
3567	disable |= 1 << ssid;
3568	enable &= ~(1 << ssid);
3569	} else {
3570	return -EINVAL;
3571	}
3572	break;
3573	} while_each_subsys_mask();
3574	if (ssid == CGROUP_SUBSYS_COUNT)
3575	return -EINVAL;
3576	}
3577
3578	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: true);
3579	if (!cgrp)
3580	return -ENODEV;
3581
3582	for_each_subsys(ss, ssid) {
3583	if (enable & (1 << ssid)) {
3584	if (cgrp->subtree_control & (1 << ssid)) {
3585	enable &= ~(1 << ssid);
3586	continue;
3587	}
3588
3589	if (!(cgroup_control(cgrp) & (1 << ssid))) {
3590	ret = -ENOENT;
3591	goto out_unlock;
3592	}
3593	} else if (disable & (1 << ssid)) {
3594	if (!(cgrp->subtree_control & (1 << ssid))) {
3595	disable &= ~(1 << ssid);
3596	continue;
3597	}
3598
3599	/* a child has it enabled? */
3600	cgroup_for_each_live_child(child, cgrp) {
3601	if (child->subtree_control & (1 << ssid)) {
3602	ret = -EBUSY;
3603	goto out_unlock;
3604	}
3605	}
3606	}
3607	}
3608
3609	if (!enable && !disable) {
3610	ret = 0;
3611	goto out_unlock;
3612	}
3613
3614	ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3615	if (ret)
3616	goto out_unlock;
3617
3618	/* save and update control masks and prepare csses */
3619	cgroup_save_control(cgrp);
3620
3621	cgrp->subtree_control |= enable;
3622	cgrp->subtree_control &= ~disable;
3623
3624	ret = cgroup_apply_control(cgrp);
3625	cgroup_finalize_control(cgrp, ret);
3626	if (ret)
3627	goto out_unlock;
3628
3629	kernfs_activate(kn: cgrp->kn);
3630	out_unlock:
3631	cgroup_kn_unlock(kn: of->kn);
3632	return ret ?: nbytes;
3633	}
3634
3635	/**
3636	* cgroup_enable_threaded - make @cgrp threaded
3637	* @cgrp: the target cgroup
3638	*
3639	* Called when "threaded" is written to the cgroup.type interface file and
3640	* tries to make @cgrp threaded and join the parent's resource domain.
3641	* This function is never called on the root cgroup as cgroup.type doesn't
3642	* exist on it.
3643	*/
3644	static int cgroup_enable_threaded(struct cgroup *cgrp)
3645	{
3646	struct cgroup *parent = cgroup_parent(cgrp);
3647	struct cgroup *dom_cgrp = parent->dom_cgrp;
3648	struct cgroup *dsct;
3649	struct cgroup_subsys_state *d_css;
3650	int ret;
3651
3652	lockdep_assert_held(&cgroup_mutex);
3653
3654	/* noop if already threaded */
3655	if (cgroup_is_threaded(cgrp))
3656	return 0;
3657
3658	/*
3659	* If @cgroup is populated or has domain controllers enabled, it
3660	* can't be switched. While the below cgroup_can_be_thread_root()
3661	* test can catch the same conditions, that's only when @parent is
3662	* not mixable, so let's check it explicitly.
3663	*/
3664	if (cgroup_is_populated(cgrp) ||
3665	cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3666	return -EOPNOTSUPP;
3667
3668	/* we're joining the parent's domain, ensure its validity */
3669	if (!cgroup_is_valid_domain(cgrp: dom_cgrp) ||
3670	!cgroup_can_be_thread_root(cgrp: dom_cgrp))
3671	return -EOPNOTSUPP;
3672
3673	/*
3674	* The following shouldn't cause actual migrations and should
3675	* always succeed.
3676	*/
3677	cgroup_save_control(cgrp);
3678
3679	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3680	if (dsct == cgrp || cgroup_is_threaded(cgrp: dsct))
3681	dsct->dom_cgrp = dom_cgrp;
3682
3683	ret = cgroup_apply_control(cgrp);
3684	if (!ret)
3685	parent->nr_threaded_children++;
3686
3687	cgroup_finalize_control(cgrp, ret);
3688	return ret;
3689	}
3690
3691	static int cgroup_type_show(struct seq_file *seq, void *v)
3692	{
3693	struct cgroup *cgrp = seq_css(seq)->cgroup;
3694
3695	if (cgroup_is_threaded(cgrp))
3696	seq_puts(m: seq, s: "threaded\n");
3697	else if (!cgroup_is_valid_domain(cgrp))
3698	seq_puts(m: seq, s: "domain invalid\n");
3699	else if (cgroup_is_thread_root(cgrp))
3700	seq_puts(m: seq, s: "domain threaded\n");
3701	else
3702	seq_puts(m: seq, s: "domain\n");
3703
3704	return 0;
3705	}
3706
3707	static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3708	size_t nbytes, loff_t off)
3709	{
3710	struct cgroup *cgrp;
3711	int ret;
3712
3713	/* only switching to threaded mode is supported */
3714	if (strcmp(strstrip(str: buf), "threaded"))
3715	return -EINVAL;
3716
3717	/* drain dying csses before we re-apply (threaded) subtree control */
3718	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: true);
3719	if (!cgrp)
3720	return -ENOENT;
3721
3722	/* threaded can only be enabled */
3723	ret = cgroup_enable_threaded(cgrp);
3724
3725	cgroup_kn_unlock(kn: of->kn);
3726	return ret ?: nbytes;
3727	}
3728
3729	static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3730	{
3731	struct cgroup *cgrp = seq_css(seq)->cgroup;
3732	int descendants = READ_ONCE(cgrp->max_descendants);
3733
3734	if (descendants == INT_MAX)
3735	seq_puts(m: seq, s: "max\n");
3736	else
3737	seq_printf(m: seq, fmt: "%d\n", descendants);
3738
3739	return 0;
3740	}
3741
3742	static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3743	char *buf, size_t nbytes, loff_t off)
3744	{
3745	struct cgroup *cgrp;
3746	int descendants;
3747	ssize_t ret;
3748
3749	buf = strstrip(str: buf);
3750	if (!strcmp(buf, "max")) {
3751	descendants = INT_MAX;
3752	} else {
3753	ret = kstrtoint(s: buf, base: 0, res: &descendants);
3754	if (ret)
3755	return ret;
3756	}
3757
3758	if (descendants < 0)
3759	return -ERANGE;
3760
3761	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: false);
3762	if (!cgrp)
3763	return -ENOENT;
3764
3765	cgrp->max_descendants = descendants;
3766
3767	cgroup_kn_unlock(kn: of->kn);
3768
3769	return nbytes;
3770	}
3771
3772	static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3773	{
3774	struct cgroup *cgrp = seq_css(seq)->cgroup;
3775	int depth = READ_ONCE(cgrp->max_depth);
3776
3777	if (depth == INT_MAX)
3778	seq_puts(m: seq, s: "max\n");
3779	else
3780	seq_printf(m: seq, fmt: "%d\n", depth);
3781
3782	return 0;
3783	}
3784
3785	static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3786	char *buf, size_t nbytes, loff_t off)
3787	{
3788	struct cgroup *cgrp;
3789	ssize_t ret;
3790	int depth;
3791
3792	buf = strstrip(str: buf);
3793	if (!strcmp(buf, "max")) {
3794	depth = INT_MAX;
3795	} else {
3796	ret = kstrtoint(s: buf, base: 0, res: &depth);
3797	if (ret)
3798	return ret;
3799	}
3800
3801	if (depth < 0)
3802	return -ERANGE;
3803
3804	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: false);
3805	if (!cgrp)
3806	return -ENOENT;
3807
3808	cgrp->max_depth = depth;
3809
3810	cgroup_kn_unlock(kn: of->kn);
3811
3812	return nbytes;
3813	}
3814
3815	static int cgroup_events_show(struct seq_file *seq, void *v)
3816	{
3817	struct cgroup *cgrp = seq_css(seq)->cgroup;
3818
3819	seq_printf(m: seq, fmt: "populated %d\n", cgroup_is_populated(cgrp));
3820	seq_printf(m: seq, fmt: "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3821
3822	return 0;
3823	}
3824
3825	static int cgroup_stat_show(struct seq_file *seq, void *v)
3826	{
3827	struct cgroup *cgroup = seq_css(seq)->cgroup;
3828	struct cgroup_subsys_state *css;
3829	int dying_cnt[CGROUP_SUBSYS_COUNT];
3830	int ssid;
3831
3832	seq_printf(m: seq, fmt: "nr_descendants %d\n",
3833	cgroup->nr_descendants);
3834
3835	/*
3836	* Show the number of live and dying csses associated with each of
3837	* non-inhibited cgroup subsystems that is bound to cgroup v2.
3838	*
3839	* Without proper lock protection, racing is possible. So the
3840	* numbers may not be consistent when that happens.
3841	*/
3842	rcu_read_lock();
3843	for (ssid = 0; ssid < CGROUP_SUBSYS_COUNT; ssid++) {
3844	dying_cnt[ssid] = -1;
3845	if ((BIT(ssid) & cgrp_dfl_inhibit_ss_mask) ||
3846	(cgroup_subsys[ssid]->root != &cgrp_dfl_root))
3847	continue;
3848	css = rcu_dereference_raw(cgroup->subsys[ssid]);
3849	dying_cnt[ssid] = cgroup->nr_dying_subsys[ssid];
3850	seq_printf(m: seq, fmt: "nr_subsys_%s %d\n", cgroup_subsys[ssid]->name,
3851	css ? (css->nr_descendants + 1) : 0);
3852	}
3853
3854	seq_printf(m: seq, fmt: "nr_dying_descendants %d\n",
3855	cgroup->nr_dying_descendants);
3856	for (ssid = 0; ssid < CGROUP_SUBSYS_COUNT; ssid++) {
3857	if (dying_cnt[ssid] >= 0)
3858	seq_printf(m: seq, fmt: "nr_dying_subsys_%s %d\n",
3859	cgroup_subsys[ssid]->name, dying_cnt[ssid]);
3860	}
3861	rcu_read_unlock();
3862	return 0;
3863	}
3864
3865	static int cgroup_core_local_stat_show(struct seq_file *seq, void *v)
3866	{
3867	struct cgroup *cgrp = seq_css(seq)->cgroup;
3868	unsigned int sequence;
3869	u64 freeze_time;
3870
3871	do {
3872	sequence = read_seqcount_begin(&cgrp->freezer.freeze_seq);
3873	freeze_time = cgrp->freezer.frozen_nsec;
3874	/* Add in current freezer interval if the cgroup is freezing. */
3875	if (test_bit(CGRP_FREEZE, &cgrp->flags))
3876	freeze_time += (ktime_get_ns() -
3877	cgrp->freezer.freeze_start_nsec);
3878	} while (read_seqcount_retry(&cgrp->freezer.freeze_seq, sequence));
3879
3880	do_div(freeze_time, NSEC_PER_USEC);
3881	seq_printf(m: seq, fmt: "frozen_usec %llu\n", freeze_time);
3882
3883	return 0;
3884	}
3885
3886	#ifdef CONFIG_CGROUP_SCHED
3887	/**
3888	* cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
3889	* @cgrp: the cgroup of interest
3890	* @ss: the subsystem of interest
3891	*
3892	* Find and get @cgrp's css associated with @ss. If the css doesn't exist
3893	* or is offline, %NULL is returned.
3894	*/
3895	static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
3896	struct cgroup_subsys *ss)
3897	{
3898	struct cgroup_subsys_state *css;
3899
3900	rcu_read_lock();
3901	css = cgroup_css(cgrp, ss);
3902	if (css && !css_tryget_online(css))
3903	css = NULL;
3904	rcu_read_unlock();
3905
3906	return css;
3907	}
3908
3909	static int cgroup_extra_stat_show(struct seq_file *seq, int ssid)
3910	{
3911	struct cgroup *cgrp = seq_css(seq)->cgroup;
3912	struct cgroup_subsys *ss = cgroup_subsys[ssid];
3913	struct cgroup_subsys_state *css;
3914	int ret;
3915
3916	if (!ss->css_extra_stat_show)
3917	return 0;
3918
3919	css = cgroup_tryget_css(cgrp, ss);
3920	if (!css)
3921	return 0;
3922
3923	ret = ss->css_extra_stat_show(seq, css);
3924	css_put(css);
3925	return ret;
3926	}
3927
3928	static int cgroup_local_stat_show(struct seq_file *seq,
3929	struct cgroup *cgrp, int ssid)
3930	{
3931	struct cgroup_subsys *ss = cgroup_subsys[ssid];
3932	struct cgroup_subsys_state *css;
3933	int ret;
3934
3935	if (!ss->css_local_stat_show)
3936	return 0;
3937
3938	css = cgroup_tryget_css(cgrp, ss);
3939	if (!css)
3940	return 0;
3941
3942	ret = ss->css_local_stat_show(seq, css);
3943	css_put(css);
3944	return ret;
3945	}
3946	#endif
3947
3948	static int cpu_stat_show(struct seq_file *seq, void *v)
3949	{
3950	int ret = 0;
3951
3952	cgroup_base_stat_cputime_show(seq);
3953	#ifdef CONFIG_CGROUP_SCHED
3954	ret = cgroup_extra_stat_show(seq, ssid: cpu_cgrp_id);
3955	#endif
3956	return ret;
3957	}
3958
3959	static int cpu_local_stat_show(struct seq_file *seq, void *v)
3960	{
3961	struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3962	int ret = 0;
3963
3964	#ifdef CONFIG_CGROUP_SCHED
3965	ret = cgroup_local_stat_show(seq, cgrp, ssid: cpu_cgrp_id);
3966	#endif
3967	return ret;
3968	}
3969
3970	#ifdef CONFIG_PSI
3971	static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3972	{
3973	struct cgroup *cgrp = seq_css(seq)->cgroup;
3974	struct psi_group *psi = cgroup_psi(cgrp);
3975
3976	return psi_show(s: seq, group: psi, res: PSI_IO);
3977	}
3978	static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3979	{
3980	struct cgroup *cgrp = seq_css(seq)->cgroup;
3981	struct psi_group *psi = cgroup_psi(cgrp);
3982
3983	return psi_show(s: seq, group: psi, res: PSI_MEM);
3984	}
3985	static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3986	{
3987	struct cgroup *cgrp = seq_css(seq)->cgroup;
3988	struct psi_group *psi = cgroup_psi(cgrp);
3989
3990	return psi_show(s: seq, group: psi, res: PSI_CPU);
3991	}
3992
3993	static ssize_t pressure_write(struct kernfs_open_file *of, char *buf,
3994	size_t nbytes, enum psi_res res)
3995	{
3996	struct cgroup_file_ctx *ctx = of->priv;
3997	struct psi_trigger *new;
3998	struct cgroup *cgrp;
3999	struct psi_group *psi;
4000
4001	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: false);
4002	if (!cgrp)
4003	return -ENODEV;
4004
4005	cgroup_get(cgrp);
4006	cgroup_kn_unlock(kn: of->kn);
4007
4008	/* Allow only one trigger per file descriptor */
4009	if (ctx->psi.trigger) {
4010	cgroup_put(cgrp);
4011	return -EBUSY;
4012	}
4013
4014	psi = cgroup_psi(cgrp);
4015	new = psi_trigger_create(group: psi, buf, res, file: of->file, of);
4016	if (IS_ERR(ptr: new)) {
4017	cgroup_put(cgrp);
4018	return PTR_ERR(ptr: new);
4019	}
4020
4021	smp_store_release(&ctx->psi.trigger, new);
4022	cgroup_put(cgrp);
4023
4024	return nbytes;
4025	}
4026
4027	static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
4028	char *buf, size_t nbytes,
4029	loff_t off)
4030	{
4031	return pressure_write(of, buf, nbytes, res: PSI_IO);
4032	}
4033
4034	static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
4035	char *buf, size_t nbytes,
4036	loff_t off)
4037	{
4038	return pressure_write(of, buf, nbytes, res: PSI_MEM);
4039	}
4040
4041	static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
4042	char *buf, size_t nbytes,
4043	loff_t off)
4044	{
4045	return pressure_write(of, buf, nbytes, res: PSI_CPU);
4046	}
4047
4048	#ifdef CONFIG_IRQ_TIME_ACCOUNTING
4049	static int cgroup_irq_pressure_show(struct seq_file *seq, void *v)
4050	{
4051	struct cgroup *cgrp = seq_css(seq)->cgroup;
4052	struct psi_group *psi = cgroup_psi(cgrp);
4053
4054	return psi_show(s: seq, group: psi, res: PSI_IRQ);
4055	}
4056
4057	static ssize_t cgroup_irq_pressure_write(struct kernfs_open_file *of,
4058	char *buf, size_t nbytes,
4059	loff_t off)
4060	{
4061	return pressure_write(of, buf, nbytes, res: PSI_IRQ);
4062	}
4063	#endif
4064
4065	static int cgroup_pressure_show(struct seq_file *seq, void *v)
4066	{
4067	struct cgroup *cgrp = seq_css(seq)->cgroup;
4068	struct psi_group *psi = cgroup_psi(cgrp);
4069
4070	seq_printf(m: seq, fmt: "%d\n", psi->enabled);
4071
4072	return 0;
4073	}
4074
4075	static ssize_t cgroup_pressure_write(struct kernfs_open_file *of,
4076	char *buf, size_t nbytes,
4077	loff_t off)
4078	{
4079	ssize_t ret;
4080	int enable;
4081	struct cgroup *cgrp;
4082	struct psi_group *psi;
4083
4084	ret = kstrtoint(s: strstrip(str: buf), base: 0, res: &enable);
4085	if (ret)
4086	return ret;
4087
4088	if (enable < 0 || enable > 1)
4089	return -ERANGE;
4090
4091	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: false);
4092	if (!cgrp)
4093	return -ENOENT;
4094
4095	psi = cgroup_psi(cgrp);
4096	if (psi->enabled != enable) {
4097	int i;
4098
4099	/* show or hide {cpu,memory,io,irq}.pressure files */
4100	for (i = 0; i < NR_PSI_RESOURCES; i++)
4101	cgroup_file_show(cfile: &cgrp->psi_files[i], show: enable);
4102
4103	psi->enabled = enable;
4104	if (enable)
4105	psi_cgroup_restart(group: psi);
4106	}
4107
4108	cgroup_kn_unlock(kn: of->kn);
4109
4110	return nbytes;
4111	}
4112
4113	static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
4114	poll_table *pt)
4115	{
4116	struct cgroup_file_ctx *ctx = of->priv;
4117
4118	return psi_trigger_poll(trigger_ptr: &ctx->psi.trigger, file: of->file, wait: pt);
4119	}
4120
4121	static void cgroup_pressure_release(struct kernfs_open_file *of)
4122	{
4123	struct cgroup_file_ctx *ctx = of->priv;
4124
4125	psi_trigger_destroy(t: ctx->psi.trigger);
4126	}
4127
4128	bool cgroup_psi_enabled(void)
4129	{
4130	if (static_branch_likely(&psi_disabled))
4131	return false;
4132
4133	return (cgroup_feature_disable_mask & (1 << OPT_FEATURE_PRESSURE)) == 0;
4134	}
4135
4136	#else /* CONFIG_PSI */
4137	bool cgroup_psi_enabled(void)
4138	{
4139	return false;
4140	}
4141
4142	#endif /* CONFIG_PSI */
4143
4144	static int cgroup_freeze_show(struct seq_file *seq, void *v)
4145	{
4146	struct cgroup *cgrp = seq_css(seq)->cgroup;
4147
4148	seq_printf(m: seq, fmt: "%d\n", cgrp->freezer.freeze);
4149
4150	return 0;
4151	}
4152
4153	static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
4154	char *buf, size_t nbytes, loff_t off)
4155	{
4156	struct cgroup *cgrp;
4157	ssize_t ret;
4158	int freeze;
4159
4160	ret = kstrtoint(s: strstrip(str: buf), base: 0, res: &freeze);
4161	if (ret)
4162	return ret;
4163
4164	if (freeze < 0 || freeze > 1)
4165	return -ERANGE;
4166
4167	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: false);
4168	if (!cgrp)
4169	return -ENOENT;
4170
4171	cgroup_freeze(cgrp, freeze);
4172
4173	cgroup_kn_unlock(kn: of->kn);
4174
4175	return nbytes;
4176	}
4177
4178	static void __cgroup_kill(struct cgroup *cgrp)
4179	{
4180	struct css_task_iter it;
4181	struct task_struct *task;
4182
4183	lockdep_assert_held(&cgroup_mutex);
4184
4185	spin_lock_irq(lock: &css_set_lock);
4186	cgrp->kill_seq++;
4187	spin_unlock_irq(lock: &css_set_lock);
4188
4189	css_task_iter_start(css: &cgrp->self, flags: CSS_TASK_ITER_PROCS | CSS_TASK_ITER_THREADED, it: &it);
4190	while ((task = css_task_iter_next(it: &it))) {
4191	/* Ignore kernel threads here. */
4192	if (task->flags & PF_KTHREAD)
4193	continue;
4194
4195	/* Skip tasks that are already dying. */
4196	if (__fatal_signal_pending(p: task))
4197	continue;
4198
4199	send_sig(SIGKILL, task, 0);
4200	}
4201	css_task_iter_end(it: &it);
4202	}
4203
4204	static void cgroup_kill(struct cgroup *cgrp)
4205	{
4206	struct cgroup_subsys_state *css;
4207	struct cgroup *dsct;
4208
4209	lockdep_assert_held(&cgroup_mutex);
4210
4211	cgroup_for_each_live_descendant_pre(dsct, css, cgrp)
4212	__cgroup_kill(cgrp: dsct);
4213	}
4214
4215	static ssize_t cgroup_kill_write(struct kernfs_open_file *of, char *buf,
4216	size_t nbytes, loff_t off)
4217	{
4218	ssize_t ret = 0;
4219	int kill;
4220	struct cgroup *cgrp;
4221
4222	ret = kstrtoint(s: strstrip(str: buf), base: 0, res: &kill);
4223	if (ret)
4224	return ret;
4225
4226	if (kill != 1)
4227	return -ERANGE;
4228
4229	cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: false);
4230	if (!cgrp)
4231	return -ENOENT;
4232
4233	/*
4234	* Killing is a process directed operation, i.e. the whole thread-group
4235	* is taken down so act like we do for cgroup.procs and only make this
4236	* writable in non-threaded cgroups.
4237	*/
4238	if (cgroup_is_threaded(cgrp))
4239	ret = -EOPNOTSUPP;
4240	else
4241	cgroup_kill(cgrp);
4242
4243	cgroup_kn_unlock(kn: of->kn);
4244
4245	return ret ?: nbytes;
4246	}
4247
4248	static int cgroup_file_open(struct kernfs_open_file *of)
4249	{
4250	struct cftype *cft = of_cft(of);
4251	struct cgroup_file_ctx *ctx;
4252	int ret;
4253
4254	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
4255	if (!ctx)
4256	return -ENOMEM;
4257
4258	ctx->ns = current->nsproxy->cgroup_ns;
4259	get_cgroup_ns(ns: ctx->ns);
4260	of->priv = ctx;
4261
4262	if (!cft->open)
4263	return 0;
4264
4265	ret = cft->open(of);
4266	if (ret) {
4267	put_cgroup_ns(ns: ctx->ns);
4268	kfree(objp: ctx);
4269	}
4270	return ret;
4271	}
4272
4273	static void cgroup_file_release(struct kernfs_open_file *of)
4274	{
4275	struct cftype *cft = of_cft(of);
4276	struct cgroup_file_ctx *ctx = of->priv;
4277
4278	if (cft->release)
4279	cft->release(of);
4280	put_cgroup_ns(ns: ctx->ns);
4281	kfree(objp: ctx);
4282	of->priv = NULL;
4283	}
4284
4285	static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
4286	size_t nbytes, loff_t off)
4287	{
4288	struct cgroup_file_ctx *ctx = of->priv;
4289	struct cgroup *cgrp = kn_priv(kn: of->kn);
4290	struct cftype *cft = of_cft(of);
4291	struct cgroup_subsys_state *css;
4292	int ret;
4293
4294	if (!nbytes)
4295	return 0;
4296
4297	/*
4298	* If namespaces are delegation boundaries, disallow writes to
4299	* files in an non-init namespace root from inside the namespace
4300	* except for the files explicitly marked delegatable -
4301	* eg. cgroup.procs, cgroup.threads and cgroup.subtree_control.
4302	*/
4303	if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
4304	!(cft->flags & CFTYPE_NS_DELEGATABLE) &&
4305	ctx->ns != &init_cgroup_ns && ctx->ns->root_cset->dfl_cgrp == cgrp)
4306	return -EPERM;
4307
4308	if (cft->write)
4309	return cft->write(of, buf, nbytes, off);
4310
4311	/*
4312	* kernfs guarantees that a file isn't deleted with operations in
4313	* flight, which means that the matching css is and stays alive and
4314	* doesn't need to be pinned. The RCU locking is not necessary
4315	* either. It's just for the convenience of using cgroup_css().
4316	*/
4317	rcu_read_lock();
4318	css = cgroup_css(cgrp, ss: cft->ss);
4319	rcu_read_unlock();
4320
4321	if (cft->write_u64) {
4322	unsigned long long v;
4323	ret = kstrtoull(s: buf, base: 0, res: &v);
4324	if (!ret)
4325	ret = cft->write_u64(css, cft, v);
4326	} else if (cft->write_s64) {
4327	long long v;
4328	ret = kstrtoll(s: buf, base: 0, res: &v);
4329	if (!ret)
4330	ret = cft->write_s64(css, cft, v);
4331	} else {
4332	ret = -EINVAL;
4333	}
4334
4335	return ret ?: nbytes;
4336	}
4337
4338	static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
4339	{
4340	struct cftype *cft = of_cft(of);
4341
4342	if (cft->poll)
4343	return cft->poll(of, pt);
4344
4345	return kernfs_generic_poll(of, pt);
4346	}
4347
4348	static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
4349	{
4350	return seq_cft(seq)->seq_start(seq, ppos);
4351	}
4352
4353	static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
4354	{
4355	return seq_cft(seq)->seq_next(seq, v, ppos);
4356	}
4357
4358	static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
4359	{
4360	if (seq_cft(seq)->seq_stop)
4361	seq_cft(seq)->seq_stop(seq, v);
4362	}
4363
4364	static int cgroup_seqfile_show(struct seq_file *m, void *arg)
4365	{
4366	struct cftype *cft = seq_cft(seq: m);
4367	struct cgroup_subsys_state *css = seq_css(seq: m);
4368
4369	if (cft->seq_show)
4370	return cft->seq_show(m, arg);
4371
4372	if (cft->read_u64)
4373	seq_printf(m, fmt: "%llu\n", cft->read_u64(css, cft));
4374	else if (cft->read_s64)
4375	seq_printf(m, fmt: "%lld\n", cft->read_s64(css, cft));
4376	else
4377	return -EINVAL;
4378	return 0;
4379	}
4380
4381	static struct kernfs_ops cgroup_kf_single_ops = {
4382	.atomic_write_len = PAGE_SIZE,
4383	.open = cgroup_file_open,
4384	.release = cgroup_file_release,
4385	.write = cgroup_file_write,
4386	.poll = cgroup_file_poll,
4387	.seq_show = cgroup_seqfile_show,
4388	};
4389
4390	static struct kernfs_ops cgroup_kf_ops = {
4391	.atomic_write_len = PAGE_SIZE,
4392	.open = cgroup_file_open,
4393	.release = cgroup_file_release,
4394	.write = cgroup_file_write,
4395	.poll = cgroup_file_poll,
4396	.seq_start = cgroup_seqfile_start,
4397	.seq_next = cgroup_seqfile_next,
4398	.seq_stop = cgroup_seqfile_stop,
4399	.seq_show = cgroup_seqfile_show,
4400	};
4401
4402	static void cgroup_file_notify_timer(struct timer_list *timer)
4403	{
4404	cgroup_file_notify(container_of(timer, struct cgroup_file,
4405	notify_timer));
4406	}
4407
4408	static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
4409	struct cftype *cft)
4410	{
4411	char name[CGROUP_FILE_NAME_MAX];
4412	struct kernfs_node *kn;
4413	struct lock_class_key *key = NULL;
4414
4415	#ifdef CONFIG_DEBUG_LOCK_ALLOC
4416	key = &cft->lockdep_key;
4417	#endif
4418	kn = __kernfs_create_file(parent: cgrp->kn, name: cgroup_file_name(cgrp, cft, buf: name),
4419	mode: cgroup_file_mode(cft),
4420	current_fsuid(), current_fsgid(),
4421	size: 0, ops: cft->kf_ops, priv: cft,
4422	NULL, key);
4423	if (IS_ERR(ptr: kn))
4424	return PTR_ERR(ptr: kn);
4425
4426	if (cft->file_offset) {
4427	struct cgroup_file *cfile = (void *)css + cft->file_offset;
4428
4429	timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
4430
4431	spin_lock_irq(lock: &cgroup_file_kn_lock);
4432	cfile->kn = kn;
4433	spin_unlock_irq(lock: &cgroup_file_kn_lock);
4434	}
4435
4436	return 0;
4437	}
4438
4439	/**
4440	* cgroup_addrm_files - add or remove files to a cgroup directory
4441	* @css: the target css
4442	* @cgrp: the target cgroup (usually css->cgroup)
4443	* @cfts: array of cftypes to be added
4444	* @is_add: whether to add or remove
4445	*
4446	* Depending on @is_add, add or remove files defined by @cfts on @cgrp.
4447	* For removals, this function never fails.
4448	*/
4449	static int cgroup_addrm_files(struct cgroup_subsys_state *css,
4450	struct cgroup *cgrp, struct cftype cfts[],
4451	bool is_add)
4452	{
4453	struct cftype *cft, *cft_end = NULL;
4454	int ret = 0;
4455
4456	lockdep_assert_held(&cgroup_mutex);
4457
4458	restart:
4459	for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
4460	/* does cft->flags tell us to skip this file on @cgrp? */
4461	if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
4462	continue;
4463	if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
4464	continue;
4465	if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
4466	continue;
4467	if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
4468	continue;
4469	if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
4470	continue;
4471	if (is_add) {
4472	ret = cgroup_add_file(css, cgrp, cft);
4473	if (ret) {
4474	pr_warn("%s: failed to add %s, err=%d\n",
4475	__func__, cft->name, ret);
4476	cft_end = cft;
4477	is_add = false;
4478	goto restart;
4479	}
4480	} else {
4481	cgroup_rm_file(cgrp, cft);
4482	}
4483	}
4484	return ret;
4485	}
4486
4487	static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
4488	{
4489	struct cgroup_subsys *ss = cfts[0].ss;
4490	struct cgroup *root = &ss->root->cgrp;
4491	struct cgroup_subsys_state *css;
4492	int ret = 0;
4493
4494	lockdep_assert_held(&cgroup_mutex);
4495
4496	/* add/rm files for all cgroups created before */
4497	css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
4498	struct cgroup *cgrp = css->cgroup;
4499
4500	if (!(css->flags & CSS_VISIBLE))
4501	continue;
4502
4503	ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
4504	if (ret)
4505	break;
4506	}
4507
4508	if (is_add && !ret)
4509	kernfs_activate(kn: root->kn);
4510	return ret;
4511	}
4512
4513	static void cgroup_exit_cftypes(struct cftype *cfts)
4514	{
4515	struct cftype *cft;
4516
4517	for (cft = cfts; cft->name[0] != '\0'; cft++) {
4518	/* free copy for custom atomic_write_len, see init_cftypes() */
4519	if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
4520	kfree(objp: cft->kf_ops);
4521	cft->kf_ops = NULL;
4522	cft->ss = NULL;
4523
4524	/* revert flags set by cgroup core while adding @cfts */
4525	cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL |
4526	__CFTYPE_ADDED);
4527	}
4528	}
4529
4530	static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4531	{
4532	struct cftype *cft;
4533	int ret = 0;
4534
4535	for (cft = cfts; cft->name[0] != '\0'; cft++) {
4536	struct kernfs_ops *kf_ops;
4537
4538	WARN_ON(cft->ss || cft->kf_ops);
4539
4540	if (cft->flags & __CFTYPE_ADDED) {
4541	ret = -EBUSY;
4542	break;
4543	}
4544
4545	if (cft->seq_start)
4546	kf_ops = &cgroup_kf_ops;
4547	else
4548	kf_ops = &cgroup_kf_single_ops;
4549
4550	/*
4551	* Ugh... if @cft wants a custom max_write_len, we need to
4552	* make a copy of kf_ops to set its atomic_write_len.
4553	*/
4554	if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
4555	kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
4556	if (!kf_ops) {
4557	ret = -ENOMEM;
4558	break;
4559	}
4560	kf_ops->atomic_write_len = cft->max_write_len;
4561	}
4562
4563	cft->kf_ops = kf_ops;
4564	cft->ss = ss;
4565	cft->flags |= __CFTYPE_ADDED;
4566	}
4567
4568	if (ret)
4569	cgroup_exit_cftypes(cfts);
4570	return ret;
4571	}
4572
4573	static void cgroup_rm_cftypes_locked(struct cftype *cfts)
4574	{
4575	lockdep_assert_held(&cgroup_mutex);
4576
4577	list_del(entry: &cfts->node);
4578	cgroup_apply_cftypes(cfts, is_add: false);
4579	cgroup_exit_cftypes(cfts);
4580	}
4581
4582	/**
4583	* cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4584	* @cfts: zero-length name terminated array of cftypes
4585	*
4586	* Unregister @cfts. Files described by @cfts are removed from all
4587	* existing cgroups and all future cgroups won't have them either. This
4588	* function can be called anytime whether @cfts' subsys is attached or not.
4589	*
4590	* Returns 0 on successful unregistration, -ENOENT if @cfts is not
4591	* registered.
4592	*/
4593	int cgroup_rm_cftypes(struct cftype *cfts)
4594	{
4595	if (!cfts || cfts[0].name[0] == '\0')
4596	return 0;
4597
4598	if (!(cfts[0].flags & __CFTYPE_ADDED))
4599	return -ENOENT;
4600
4601	cgroup_lock();
4602	cgroup_rm_cftypes_locked(cfts);
4603	cgroup_unlock();
4604	return 0;
4605	}
4606
4607	/**
4608	* cgroup_add_cftypes - add an array of cftypes to a subsystem
4609	* @ss: target cgroup subsystem
4610	* @cfts: zero-length name terminated array of cftypes
4611	*
4612	* Register @cfts to @ss. Files described by @cfts are created for all
4613	* existing cgroups to which @ss is attached and all future cgroups will
4614	* have them too. This function can be called anytime whether @ss is
4615	* attached or not.
4616	*
4617	* Returns 0 on successful registration, -errno on failure. Note that this
4618	* function currently returns 0 as long as @cfts registration is successful
4619	* even if some file creation attempts on existing cgroups fail.
4620	*/
4621	int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4622	{
4623	int ret;
4624
4625	if (!cgroup_ssid_enabled(ssid: ss->id))
4626	return 0;
4627
4628	if (!cfts || cfts[0].name[0] == '\0')
4629	return 0;
4630
4631	ret = cgroup_init_cftypes(ss, cfts);
4632	if (ret)
4633	return ret;
4634
4635	cgroup_lock();
4636
4637	list_add_tail(new: &cfts->node, head: &ss->cfts);
4638	ret = cgroup_apply_cftypes(cfts, is_add: true);
4639	if (ret)
4640	cgroup_rm_cftypes_locked(cfts);
4641
4642	cgroup_unlock();
4643	return ret;
4644	}
4645
4646	/**
4647	* cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4648	* @ss: target cgroup subsystem
4649	* @cfts: zero-length name terminated array of cftypes
4650	*
4651	* Similar to cgroup_add_cftypes() but the added files are only used for
4652	* the default hierarchy.
4653	*/
4654	int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4655	{
4656	struct cftype *cft;
4657
4658	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4659	cft->flags |= __CFTYPE_ONLY_ON_DFL;
4660	return cgroup_add_cftypes(ss, cfts);
4661	}
4662
4663	/**
4664	* cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4665	* @ss: target cgroup subsystem
4666	* @cfts: zero-length name terminated array of cftypes
4667	*
4668	* Similar to cgroup_add_cftypes() but the added files are only used for
4669	* the legacy hierarchies.
4670	*/
4671	int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4672	{
4673	struct cftype *cft;
4674
4675	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4676	cft->flags |= __CFTYPE_NOT_ON_DFL;
4677	return cgroup_add_cftypes(ss, cfts);
4678	}
4679
4680	/**
4681	* cgroup_file_notify - generate a file modified event for a cgroup_file
4682	* @cfile: target cgroup_file
4683	*
4684	* @cfile must have been obtained by setting cftype->file_offset.
4685	*/
4686	void cgroup_file_notify(struct cgroup_file *cfile)
4687	{
4688	unsigned long flags;
4689
4690	spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4691	if (cfile->kn) {
4692	unsigned long last = cfile->notified_at;
4693	unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4694
4695	if (time_in_range(jiffies, last, next)) {
4696	timer_reduce(timer: &cfile->notify_timer, expires: next);
4697	} else {
4698	kernfs_notify(kn: cfile->kn);
4699	cfile->notified_at = jiffies;
4700	}
4701	}
4702	spin_unlock_irqrestore(lock: &cgroup_file_kn_lock, flags);
4703	}
4704	EXPORT_SYMBOL_GPL(cgroup_file_notify);
4705
4706	/**
4707	* cgroup_file_show - show or hide a hidden cgroup file
4708	* @cfile: target cgroup_file obtained by setting cftype->file_offset
4709	* @show: whether to show or hide
4710	*/
4711	void cgroup_file_show(struct cgroup_file *cfile, bool show)
4712	{
4713	struct kernfs_node *kn;
4714
4715	spin_lock_irq(lock: &cgroup_file_kn_lock);
4716	kn = cfile->kn;
4717	kernfs_get(kn);
4718	spin_unlock_irq(lock: &cgroup_file_kn_lock);
4719
4720	if (kn)
4721	kernfs_show(kn, show);
4722
4723	kernfs_put(kn);
4724	}
4725
4726	/**
4727	* css_next_child - find the next child of a given css
4728	* @pos: the current position (%NULL to initiate traversal)
4729	* @parent: css whose children to walk
4730	*
4731	* This function returns the next child of @parent and should be called
4732	* under either cgroup_mutex or RCU read lock. The only requirement is
4733	* that @parent and @pos are accessible. The next sibling is guaranteed to
4734	* be returned regardless of their states.
4735	*
4736	* If a subsystem synchronizes ->css_online() and the start of iteration, a
4737	* css which finished ->css_online() is guaranteed to be visible in the
4738	* future iterations and will stay visible until the last reference is put.
4739	* A css which hasn't finished ->css_online() or already finished
4740	* ->css_offline() may show up during traversal. It's each subsystem's
4741	* responsibility to synchronize against on/offlining.
4742	*/
4743	struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4744	struct cgroup_subsys_state *parent)
4745	{
4746	struct cgroup_subsys_state *next;
4747
4748	cgroup_assert_mutex_or_rcu_locked();
4749
4750	/*
4751	* @pos could already have been unlinked from the sibling list.
4752	* Once a cgroup is removed, its ->sibling.next is no longer
4753	* updated when its next sibling changes. CSS_RELEASED is set when
4754	* @pos is taken off list, at which time its next pointer is valid,
4755	* and, as releases are serialized, the one pointed to by the next
4756	* pointer is guaranteed to not have started release yet. This
4757	* implies that if we observe !CSS_RELEASED on @pos in this RCU
4758	* critical section, the one pointed to by its next pointer is
4759	* guaranteed to not have finished its RCU grace period even if we
4760	* have dropped rcu_read_lock() in-between iterations.
4761	*
4762	* If @pos has CSS_RELEASED set, its next pointer can't be
4763	* dereferenced; however, as each css is given a monotonically
4764	* increasing unique serial number and always appended to the
4765	* sibling list, the next one can be found by walking the parent's
4766	* children until the first css with higher serial number than
4767	* @pos's. While this path can be slower, it happens iff iteration
4768	* races against release and the race window is very small.
4769	*/
4770	if (!pos) {
4771	next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4772	} else if (likely(!(pos->flags & CSS_RELEASED))) {
4773	next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4774	} else {
4775	list_for_each_entry_rcu(next, &parent->children, sibling,
4776	lockdep_is_held(&cgroup_mutex))
4777	if (next->serial_nr > pos->serial_nr)
4778	break;
4779	}
4780
4781	/*
4782	* @next, if not pointing to the head, can be dereferenced and is
4783	* the next sibling.
4784	*/
4785	if (&next->sibling != &parent->children)
4786	return next;
4787	return NULL;
4788	}
4789
4790	/**
4791	* css_next_descendant_pre - find the next descendant for pre-order walk
4792	* @pos: the current position (%NULL to initiate traversal)
4793	* @root: css whose descendants to walk
4794	*
4795	* To be used by css_for_each_descendant_pre(). Find the next descendant
4796	* to visit for pre-order traversal of @root's descendants. @root is
4797	* included in the iteration and the first node to be visited.
4798	*
4799	* While this function requires cgroup_mutex or RCU read locking, it
4800	* doesn't require the whole traversal to be contained in a single critical
4801	* section. Additionally, it isn't necessary to hold onto a reference to @pos.
4802	* This function will return the correct next descendant as long as both @pos
4803	* and @root are accessible and @pos is a descendant of @root.
4804	*
4805	* If a subsystem synchronizes ->css_online() and the start of iteration, a
4806	* css which finished ->css_online() is guaranteed to be visible in the
4807	* future iterations and will stay visible until the last reference is put.
4808	* A css which hasn't finished ->css_online() or already finished
4809	* ->css_offline() may show up during traversal. It's each subsystem's
4810	* responsibility to synchronize against on/offlining.
4811	*/
4812	struct cgroup_subsys_state *
4813	css_next_descendant_pre(struct cgroup_subsys_state *pos,
4814	struct cgroup_subsys_state *root)
4815	{
4816	struct cgroup_subsys_state *next;
4817
4818	cgroup_assert_mutex_or_rcu_locked();
4819
4820	/* if first iteration, visit @root */
4821	if (!pos)
4822	return root;
4823
4824	/* visit the first child if exists */
4825	next = css_next_child(NULL, parent: pos);
4826	if (next)
4827	return next;
4828
4829	/* no child, visit my or the closest ancestor's next sibling */
4830	while (pos != root) {
4831	next = css_next_child(pos, parent: pos->parent);
4832	if (next)
4833	return next;
4834	pos = pos->parent;
4835	}
4836
4837	return NULL;
4838	}
4839	EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4840
4841	/**
4842	* css_rightmost_descendant - return the rightmost descendant of a css
4843	* @pos: css of interest
4844	*
4845	* Return the rightmost descendant of @pos. If there's no descendant, @pos
4846	* is returned. This can be used during pre-order traversal to skip
4847	* subtree of @pos.
4848	*
4849	* While this function requires cgroup_mutex or RCU read locking, it
4850	* doesn't require the whole traversal to be contained in a single critical
4851	* section. Additionally, it isn't necessary to hold onto a reference to @pos.
4852	* This function will return the correct rightmost descendant as long as @pos
4853	* is accessible.
4854	*/
4855	struct cgroup_subsys_state *
4856	css_rightmost_descendant(struct cgroup_subsys_state *pos)
4857	{
4858	struct cgroup_subsys_state *last, *tmp;
4859
4860	cgroup_assert_mutex_or_rcu_locked();
4861
4862	do {
4863	last = pos;
4864	/* ->prev isn't RCU safe, walk ->next till the end */
4865	pos = NULL;
4866	css_for_each_child(tmp, last)
4867	pos = tmp;
4868	} while (pos);
4869
4870	return last;
4871	}
4872
4873	static struct cgroup_subsys_state *
4874	css_leftmost_descendant(struct cgroup_subsys_state *pos)
4875	{
4876	struct cgroup_subsys_state *last;
4877
4878	do {
4879	last = pos;
4880	pos = css_next_child(NULL, parent: pos);
4881	} while (pos);
4882
4883	return last;
4884	}
4885
4886	/**
4887	* css_next_descendant_post - find the next descendant for post-order walk
4888	* @pos: the current position (%NULL to initiate traversal)
4889	* @root: css whose descendants to walk
4890	*
4891	* To be used by css_for_each_descendant_post(). Find the next descendant
4892	* to visit for post-order traversal of @root's descendants. @root is
4893	* included in the iteration and the last node to be visited.
4894	*
4895	* While this function requires cgroup_mutex or RCU read locking, it
4896	* doesn't require the whole traversal to be contained in a single critical
4897	* section. Additionally, it isn't necessary to hold onto a reference to @pos.
4898	* This function will return the correct next descendant as long as both @pos
4899	* and @cgroup are accessible and @pos is a descendant of @cgroup.
4900	*
4901	* If a subsystem synchronizes ->css_online() and the start of iteration, a
4902	* css which finished ->css_online() is guaranteed to be visible in the
4903	* future iterations and will stay visible until the last reference is put.
4904	* A css which hasn't finished ->css_online() or already finished
4905	* ->css_offline() may show up during traversal. It's each subsystem's
4906	* responsibility to synchronize against on/offlining.
4907	*/
4908	struct cgroup_subsys_state *
4909	css_next_descendant_post(struct cgroup_subsys_state *pos,
4910	struct cgroup_subsys_state *root)
4911	{
4912	struct cgroup_subsys_state *next;
4913
4914	cgroup_assert_mutex_or_rcu_locked();
4915
4916	/* if first iteration, visit leftmost descendant which may be @root */
4917	if (!pos)
4918	return css_leftmost_descendant(pos: root);
4919
4920	/* if we visited @root, we're done */
4921	if (pos == root)
4922	return NULL;
4923
4924	/* if there's an unvisited sibling, visit its leftmost descendant */
4925	next = css_next_child(pos, parent: pos->parent);
4926	if (next)
4927	return css_leftmost_descendant(pos: next);
4928
4929	/* no sibling left, visit parent */
4930	return pos->parent;
4931	}
4932
4933	/**
4934	* css_has_online_children - does a css have online children
4935	* @css: the target css
4936	*
4937	* Returns %true if @css has any online children; otherwise, %false. This
4938	* function can be called from any context but the caller is responsible
4939	* for synchronizing against on/offlining as necessary.
4940	*/
4941	bool css_has_online_children(struct cgroup_subsys_state *css)
4942	{
4943	struct cgroup_subsys_state *child;
4944	bool ret = false;
4945
4946	rcu_read_lock();
4947	css_for_each_child(child, css) {
4948	if (child->flags & CSS_ONLINE) {
4949	ret = true;
4950	break;
4951	}
4952	}
4953	rcu_read_unlock();
4954	return ret;
4955	}
4956
4957	static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4958	{
4959	struct list_head *l;
4960	struct cgrp_cset_link *link;
4961	struct css_set *cset;
4962
4963	lockdep_assert_held(&css_set_lock);
4964
4965	/* find the next threaded cset */
4966	if (it->tcset_pos) {
4967	l = it->tcset_pos->next;
4968
4969	if (l != it->tcset_head) {
4970	it->tcset_pos = l;
4971	return container_of(l, struct css_set,
4972	threaded_csets_node);
4973	}
4974
4975	it->tcset_pos = NULL;
4976	}
4977
4978	/* find the next cset */
4979	l = it->cset_pos;
4980	l = l->next;
4981	if (l == it->cset_head) {
4982	it->cset_pos = NULL;
4983	return NULL;
4984	}
4985
4986	if (it->ss) {
4987	cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4988	} else {
4989	link = list_entry(l, struct cgrp_cset_link, cset_link);
4990	cset = link->cset;
4991	}
4992
4993	it->cset_pos = l;
4994
4995	/* initialize threaded css_set walking */
4996	if (it->flags & CSS_TASK_ITER_THREADED) {
4997	if (it->cur_dcset)
4998	put_css_set_locked(cset: it->cur_dcset);
4999	it->cur_dcset = cset;
5000	get_css_set(cset);
5001
5002	it->tcset_head = &cset->threaded_csets;
5003	it->tcset_pos = &cset->threaded_csets;
5004	}
5005
5006	return cset;
5007	}
5008
5009	/**
5010	* css_task_iter_advance_css_set - advance a task iterator to the next css_set
5011	* @it: the iterator to advance
5012	*
5013	* Advance @it to the next css_set to walk.
5014	*/
5015	static void css_task_iter_advance_css_set(struct css_task_iter *it)
5016	{
5017	struct css_set *cset;
5018
5019	lockdep_assert_held(&css_set_lock);
5020
5021	/* Advance to the next non-empty css_set and find first non-empty tasks list*/
5022	while ((cset = css_task_iter_next_css_set(it))) {
5023	if (!list_empty(head: &cset->tasks)) {
5024	it->cur_tasks_head = &cset->tasks;
5025	break;
5026	} else if (!list_empty(head: &cset->mg_tasks)) {
5027	it->cur_tasks_head = &cset->mg_tasks;
5028	break;
5029	} else if (!list_empty(head: &cset->dying_tasks)) {
5030	it->cur_tasks_head = &cset->dying_tasks;
5031	break;
5032	}
5033	}
5034	if (!cset) {
5035	it->task_pos = NULL;
5036	return;
5037	}
5038	it->task_pos = it->cur_tasks_head->next;
5039
5040	/*
5041	* We don't keep css_sets locked across iteration steps and thus
5042	* need to take steps to ensure that iteration can be resumed after
5043	* the lock is re-acquired. Iteration is performed at two levels -
5044	* css_sets and tasks in them.
5045	*
5046	* Once created, a css_set never leaves its cgroup lists, so a
5047	* pinned css_set is guaranteed to stay put and we can resume
5048	* iteration afterwards.
5049	*
5050	* Tasks may leave @cset across iteration steps. This is resolved
5051	* by registering each iterator with the css_set currently being
5052	* walked and making css_set_move_task() advance iterators whose
5053	* next task is leaving.
5054	*/
5055	if (it->cur_cset) {
5056	list_del(entry: &it->iters_node);
5057	put_css_set_locked(cset: it->cur_cset);
5058	}
5059	get_css_set(cset);
5060	it->cur_cset = cset;
5061	list_add(new: &it->iters_node, head: &cset->task_iters);
5062	}
5063
5064	static void css_task_iter_skip(struct css_task_iter *it,
5065	struct task_struct *task)
5066	{
5067	lockdep_assert_held(&css_set_lock);
5068
5069	if (it->task_pos == &task->cg_list) {
5070	it->task_pos = it->task_pos->next;
5071	it->flags |= CSS_TASK_ITER_SKIPPED;
5072	}
5073	}
5074
5075	static void css_task_iter_advance(struct css_task_iter *it)
5076	{
5077	struct task_struct *task;
5078
5079	lockdep_assert_held(&css_set_lock);
5080	repeat:
5081	if (it->task_pos) {
5082	/*
5083	* Advance iterator to find next entry. We go through cset
5084	* tasks, mg_tasks and dying_tasks, when consumed we move onto
5085	* the next cset.
5086	*/
5087	if (it->flags & CSS_TASK_ITER_SKIPPED)
5088	it->flags &= ~CSS_TASK_ITER_SKIPPED;
5089	else
5090	it->task_pos = it->task_pos->next;
5091
5092	if (it->task_pos == &it->cur_cset->tasks) {
5093	it->cur_tasks_head = &it->cur_cset->mg_tasks;
5094	it->task_pos = it->cur_tasks_head->next;
5095	}
5096	if (it->task_pos == &it->cur_cset->mg_tasks) {
5097	it->cur_tasks_head = &it->cur_cset->dying_tasks;
5098	it->task_pos = it->cur_tasks_head->next;
5099	}
5100	if (it->task_pos == &it->cur_cset->dying_tasks)
5101	css_task_iter_advance_css_set(it);
5102	} else {
5103	/* called from start, proceed to the first cset */
5104	css_task_iter_advance_css_set(it);
5105	}
5106
5107	if (!it->task_pos)
5108	return;
5109
5110	task = list_entry(it->task_pos, struct task_struct, cg_list);
5111
5112	if (it->flags & CSS_TASK_ITER_PROCS) {
5113	/* if PROCS, skip over tasks which aren't group leaders */
5114	if (!thread_group_leader(p: task))
5115	goto repeat;
5116
5117	/* and dying leaders w/o live member threads */
5118	if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
5119	!atomic_read(v: &task->signal->live))
5120	goto repeat;
5121	} else {
5122	/* skip all dying ones */
5123	if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
5124	goto repeat;
5125	}
5126	}
5127
5128	/**
5129	* css_task_iter_start - initiate task iteration
5130	* @css: the css to walk tasks of
5131	* @flags: CSS_TASK_ITER_* flags
5132	* @it: the task iterator to use
5133	*
5134	* Initiate iteration through the tasks of @css. The caller can call
5135	* css_task_iter_next() to walk through the tasks until the function
5136	* returns NULL. On completion of iteration, css_task_iter_end() must be
5137	* called.
5138	*/
5139	void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
5140	struct css_task_iter *it)
5141	{
5142	unsigned long irqflags;
5143
5144	memset(it, 0, sizeof(*it));
5145
5146	spin_lock_irqsave(&css_set_lock, irqflags);
5147
5148	it->ss = css->ss;
5149	it->flags = flags;
5150
5151	if (CGROUP_HAS_SUBSYS_CONFIG && it->ss)
5152	it->cset_pos = &css->cgroup->e_csets[css->ss->id];
5153	else
5154	it->cset_pos = &css->cgroup->cset_links;
5155
5156	it->cset_head = it->cset_pos;
5157
5158	css_task_iter_advance(it);
5159
5160	spin_unlock_irqrestore(lock: &css_set_lock, flags: irqflags);
5161	}
5162
5163	/**
5164	* css_task_iter_next - return the next task for the iterator
5165	* @it: the task iterator being iterated
5166	*
5167	* The "next" function for task iteration. @it should have been
5168	* initialized via css_task_iter_start(). Returns NULL when the iteration
5169	* reaches the end.
5170	*/
5171	struct task_struct *css_task_iter_next(struct css_task_iter *it)
5172	{
5173	unsigned long irqflags;
5174
5175	if (it->cur_task) {
5176	put_task_struct(t: it->cur_task);
5177	it->cur_task = NULL;
5178	}
5179
5180	spin_lock_irqsave(&css_set_lock, irqflags);
5181
5182	/* @it may be half-advanced by skips, finish advancing */
5183	if (it->flags & CSS_TASK_ITER_SKIPPED)
5184	css_task_iter_advance(it);
5185
5186	if (it->task_pos) {
5187	it->cur_task = list_entry(it->task_pos, struct task_struct,
5188	cg_list);
5189	get_task_struct(t: it->cur_task);
5190	css_task_iter_advance(it);
5191	}
5192
5193	spin_unlock_irqrestore(lock: &css_set_lock, flags: irqflags);
5194
5195	return it->cur_task;
5196	}
5197
5198	/**
5199	* css_task_iter_end - finish task iteration
5200	* @it: the task iterator to finish
5201	*
5202	* Finish task iteration started by css_task_iter_start().
5203	*/
5204	void css_task_iter_end(struct css_task_iter *it)
5205	{
5206	unsigned long irqflags;
5207
5208	if (it->cur_cset) {
5209	spin_lock_irqsave(&css_set_lock, irqflags);
5210	list_del(entry: &it->iters_node);
5211	put_css_set_locked(cset: it->cur_cset);
5212	spin_unlock_irqrestore(lock: &css_set_lock, flags: irqflags);
5213	}
5214
5215	if (it->cur_dcset)
5216	put_css_set(cset: it->cur_dcset);
5217
5218	if (it->cur_task)
5219	put_task_struct(t: it->cur_task);
5220	}
5221
5222	static void cgroup_procs_release(struct kernfs_open_file *of)
5223	{
5224	struct cgroup_file_ctx *ctx = of->priv;
5225
5226	if (ctx->procs.started)
5227	css_task_iter_end(it: &ctx->procs.iter);
5228	}
5229
5230	static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
5231	{
5232	struct kernfs_open_file *of = s->private;
5233	struct cgroup_file_ctx *ctx = of->priv;
5234
5235	if (pos)
5236	(*pos)++;
5237
5238	return css_task_iter_next(it: &ctx->procs.iter);
5239	}
5240
5241	static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
5242	unsigned int iter_flags)
5243	{
5244	struct kernfs_open_file *of = s->private;
5245	struct cgroup *cgrp = seq_css(seq: s)->cgroup;
5246	struct cgroup_file_ctx *ctx = of->priv;
5247	struct css_task_iter *it = &ctx->procs.iter;
5248
5249	/*
5250	* When a seq_file is seeked, it's always traversed sequentially
5251	* from position 0, so we can simply keep iterating on !0 *pos.
5252	*/
5253	if (!ctx->procs.started) {
5254	if (WARN_ON_ONCE((*pos)))
5255	return ERR_PTR(error: -EINVAL);
5256	css_task_iter_start(css: &cgrp->self, flags: iter_flags, it);
5257	ctx->procs.started = true;
5258	} else if (!(*pos)) {
5259	css_task_iter_end(it);
5260	css_task_iter_start(css: &cgrp->self, flags: iter_flags, it);
5261	} else
5262	return it->cur_task;
5263
5264	return cgroup_procs_next(s, NULL, NULL);
5265	}
5266
5267	static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
5268	{
5269	struct cgroup *cgrp = seq_css(seq: s)->cgroup;
5270
5271	/*
5272	* All processes of a threaded subtree belong to the domain cgroup
5273	* of the subtree. Only threads can be distributed across the
5274	* subtree. Reject reads on cgroup.procs in the subtree proper.
5275	* They're always empty anyway.
5276	*/
5277	if (cgroup_is_threaded(cgrp))
5278	return ERR_PTR(error: -EOPNOTSUPP);
5279
5280	return __cgroup_procs_start(s, pos, iter_flags: CSS_TASK_ITER_PROCS |
5281	CSS_TASK_ITER_THREADED);
5282	}
5283
5284	static int cgroup_procs_show(struct seq_file *s, void *v)
5285	{
5286	seq_printf(m: s, fmt: "%d\n", task_pid_vnr(tsk: v));
5287	return 0;
5288	}
5289
5290	static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
5291	{
5292	int ret;
5293	struct inode *inode;
5294
5295	lockdep_assert_held(&cgroup_mutex);
5296
5297	inode = kernfs_get_inode(sb, kn: cgrp->procs_file.kn);
5298	if (!inode)
5299	return -ENOMEM;
5300
5301	ret = inode_permission(&nop_mnt_idmap, inode, MAY_WRITE);
5302	iput(inode);
5303	return ret;
5304	}
5305
5306	static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
5307	struct cgroup *dst_cgrp,
5308	struct super_block *sb,
5309	struct cgroup_namespace *ns)
5310	{
5311	struct cgroup *com_cgrp = src_cgrp;
5312	int ret;
5313
5314	lockdep_assert_held(&cgroup_mutex);
5315
5316	/* find the common ancestor */
5317	while (!cgroup_is_descendant(cgrp: dst_cgrp, ancestor: com_cgrp))
5318	com_cgrp = cgroup_parent(cgrp: com_cgrp);
5319
5320	/* %current should be authorized to migrate to the common ancestor */
5321	ret = cgroup_may_write(cgrp: com_cgrp, sb);
5322	if (ret)
5323	return ret;
5324
5325	/*
5326	* If namespaces are delegation boundaries, %current must be able
5327	* to see both source and destination cgroups from its namespace.
5328	*/
5329	if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
5330	(!cgroup_is_descendant(cgrp: src_cgrp, ancestor: ns->root_cset->dfl_cgrp) ||
5331	!cgroup_is_descendant(cgrp: dst_cgrp, ancestor: ns->root_cset->dfl_cgrp)))
5332	return -ENOENT;
5333
5334	return 0;
5335	}
5336
5337	static int cgroup_attach_permissions(struct cgroup *src_cgrp,
5338	struct cgroup *dst_cgrp,
5339	struct super_block *sb, bool threadgroup,
5340	struct cgroup_namespace *ns)
5341	{
5342	int ret = 0;
5343
5344	ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb, ns);
5345	if (ret)
5346	return ret;
5347
5348	ret = cgroup_migrate_vet_dst(dst_cgrp);
5349	if (ret)
5350	return ret;
5351
5352	if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
5353	ret = -EOPNOTSUPP;
5354
5355	return ret;
5356	}
5357
5358	static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
5359	bool threadgroup)
5360	{
5361	struct cgroup_file_ctx *ctx = of->priv;
5362	struct cgroup *src_cgrp, *dst_cgrp;
5363	struct task_struct *task;
5364	ssize_t ret;
5365	enum cgroup_attach_lock_mode lock_mode;
5366
5367	dst_cgrp = cgroup_kn_lock_live(kn: of->kn, drain_offline: false);
5368	if (!dst_cgrp)
5369	return -ENODEV;
5370
5371	task = cgroup_procs_write_start(buf, threadgroup, lock_mode: &lock_mode);
5372	ret = PTR_ERR_OR_ZERO(ptr: task);
5373	if (ret)
5374	goto out_unlock;
5375
5376	/* find the source cgroup */
5377	spin_lock_irq(lock: &css_set_lock);
5378	src_cgrp = task_cgroup_from_root(task, root: &cgrp_dfl_root);
5379	spin_unlock_irq(lock: &css_set_lock);
5380
5381	/*
5382	* Process and thread migrations follow same delegation rule. Check
5383	* permissions using the credentials from file open to protect against
5384	* inherited fd attacks.
5385	*/
5386	scoped_with_creds(of->file->f_cred)
5387	ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
5388	sb: of->file->f_path.dentry->d_sb,
5389	threadgroup, ns: ctx->ns);
5390	if (ret)
5391	goto out_finish;
5392
5393	ret = cgroup_attach_task(dst_cgrp, leader: task, threadgroup);
5394
5395	out_finish:
5396	cgroup_procs_write_finish(task, lock_mode);
5397	out_unlock:
5398	cgroup_kn_unlock(kn: of->kn);
5399
5400	return ret;
5401	}
5402
5403	static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
5404	char *buf, size_t nbytes, loff_t off)
5405	{
5406	return __cgroup_procs_write(of, buf, threadgroup: true) ?: nbytes;
5407	}
5408
5409	static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
5410	{
5411	return __cgroup_procs_start(s, pos, iter_flags: 0);
5412	}
5413
5414	static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
5415	char *buf, size_t nbytes, loff_t off)
5416	{
5417	return __cgroup_procs_write(of, buf, threadgroup: false) ?: nbytes;
5418	}
5419
5420	/* cgroup core interface files for the default hierarchy */
5421	static struct cftype cgroup_base_files[] = {
5422	{
5423	.name = "cgroup.type",
5424	.flags = CFTYPE_NOT_ON_ROOT,
5425	.seq_show = cgroup_type_show,
5426	.write = cgroup_type_write,
5427	},
5428	{
5429	.name = "cgroup.procs",
5430	.flags = CFTYPE_NS_DELEGATABLE,
5431	.file_offset = offsetof(struct cgroup, procs_file),
5432	.release = cgroup_procs_release,
5433	.seq_start = cgroup_procs_start,
5434	.seq_next = cgroup_procs_next,
5435	.seq_show = cgroup_procs_show,
5436	.write = cgroup_procs_write,
5437	},
5438	{
5439	.name = "cgroup.threads",
5440	.flags = CFTYPE_NS_DELEGATABLE,
5441	.release = cgroup_procs_release,
5442	.seq_start = cgroup_threads_start,
5443	.seq_next = cgroup_procs_next,
5444	.seq_show = cgroup_procs_show,
5445	.write = cgroup_threads_write,
5446	},
5447	{
5448	.name = "cgroup.controllers",
5449	.seq_show = cgroup_controllers_show,
5450	},
5451	{
5452	.name = "cgroup.subtree_control",
5453	.flags = CFTYPE_NS_DELEGATABLE,
5454	.seq_show = cgroup_subtree_control_show,
5455	.write = cgroup_subtree_control_write,
5456	},
5457	{
5458	.name = "cgroup.events",
5459	.flags = CFTYPE_NOT_ON_ROOT,
5460	.file_offset = offsetof(struct cgroup, events_file),
5461	.seq_show = cgroup_events_show,
5462	},
5463	{
5464	.name = "cgroup.max.descendants",
5465	.seq_show = cgroup_max_descendants_show,
5466	.write = cgroup_max_descendants_write,
5467	},
5468	{
5469	.name = "cgroup.max.depth",
5470	.seq_show = cgroup_max_depth_show,
5471	.write = cgroup_max_depth_write,
5472	},
5473	{
5474	.name = "cgroup.stat",
5475	.seq_show = cgroup_stat_show,
5476	},
5477	{
5478	.name = "cgroup.stat.local",
5479	.flags = CFTYPE_NOT_ON_ROOT,
5480	.seq_show = cgroup_core_local_stat_show,
5481	},
5482	{
5483	.name = "cgroup.freeze",
5484	.flags = CFTYPE_NOT_ON_ROOT,
5485	.seq_show = cgroup_freeze_show,
5486	.write = cgroup_freeze_write,
5487	},
5488	{
5489	.name = "cgroup.kill",
5490	.flags = CFTYPE_NOT_ON_ROOT,
5491	.write = cgroup_kill_write,
5492	},
5493	{
5494	.name = "cpu.stat",
5495	.seq_show = cpu_stat_show,
5496	},
5497	{
5498	.name = "cpu.stat.local",
5499	.seq_show = cpu_local_stat_show,
5500	},
5501	{ } /* terminate */
5502	};
5503
5504	static struct cftype cgroup_psi_files[] = {
5505	#ifdef CONFIG_PSI
5506	{
5507	.name = "io.pressure",
5508	.file_offset = offsetof(struct cgroup, psi_files[PSI_IO]),
5509	.seq_show = cgroup_io_pressure_show,
5510	.write = cgroup_io_pressure_write,
5511	.poll = cgroup_pressure_poll,
5512	.release = cgroup_pressure_release,
5513	},
5514	{
5515	.name = "memory.pressure",
5516	.file_offset = offsetof(struct cgroup, psi_files[PSI_MEM]),
5517	.seq_show = cgroup_memory_pressure_show,
5518	.write = cgroup_memory_pressure_write,
5519	.poll = cgroup_pressure_poll,
5520	.release = cgroup_pressure_release,
5521	},
5522	{
5523	.name = "cpu.pressure",
5524	.file_offset = offsetof(struct cgroup, psi_files[PSI_CPU]),
5525	.seq_show = cgroup_cpu_pressure_show,
5526	.write = cgroup_cpu_pressure_write,
5527	.poll = cgroup_pressure_poll,
5528	.release = cgroup_pressure_release,
5529	},
5530	#ifdef CONFIG_IRQ_TIME_ACCOUNTING
5531	{
5532	.name = "irq.pressure",
5533	.file_offset = offsetof(struct cgroup, psi_files[PSI_IRQ]),
5534	.seq_show = cgroup_irq_pressure_show,
5535	.write = cgroup_irq_pressure_write,
5536	.poll = cgroup_pressure_poll,
5537	.release = cgroup_pressure_release,
5538	},
5539	#endif
5540	{
5541	.name = "cgroup.pressure",
5542	.seq_show = cgroup_pressure_show,
5543	.write = cgroup_pressure_write,
5544	},
5545	#endif /* CONFIG_PSI */
5546	{ } /* terminate */
5547	};
5548
5549	/*
5550	* css destruction is four-stage process.
5551	*
5552	* 1. Destruction starts. Killing of the percpu_ref is initiated.
5553	* Implemented in kill_css().
5554	*
5555	* 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5556	* and thus css_tryget_online() is guaranteed to fail, the css can be
5557	* offlined by invoking offline_css(). After offlining, the base ref is
5558	* put. Implemented in css_killed_work_fn().
5559	*
5560	* 3. When the percpu_ref reaches zero, the only possible remaining
5561	* accessors are inside RCU read sections. css_release() schedules the
5562	* RCU callback.
5563	*
5564	* 4. After the grace period, the css can be freed. Implemented in
5565	* css_free_rwork_fn().
5566	*
5567	* It is actually hairier because both step 2 and 4 require process context
5568	* and thus involve punting to css->destroy_work adding two additional
5569	* steps to the already complex sequence.
5570	*/
5571	static void css_free_rwork_fn(struct work_struct *work)
5572	{
5573	struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
5574	struct cgroup_subsys_state, destroy_rwork);
5575	struct cgroup_subsys *ss = css->ss;
5576	struct cgroup *cgrp = css->cgroup;
5577
5578	percpu_ref_exit(ref: &css->refcnt);
5579	css_rstat_exit(css);
5580
5581	if (!css_is_self(css)) {
5582	/* css free path */
5583	struct cgroup_subsys_state *parent = css->parent;
5584	int id = css->id;
5585
5586	ss->css_free(css);
5587	cgroup_idr_remove(idr: &ss->css_idr, id);
5588	cgroup_put(cgrp);
5589
5590	if (parent)
5591	css_put(parent);
5592	} else {
5593	/* cgroup free path */
5594	atomic_dec(v: &cgrp->root->nr_cgrps);
5595	if (!cgroup_on_dfl(cgrp))
5596	cgroup1_pidlist_destroy_all(cgrp);
5597	cancel_work_sync(work: &cgrp->release_agent_work);
5598	bpf_cgrp_storage_free(cgroup: cgrp);
5599
5600	if (cgroup_parent(cgrp)) {
5601	/*
5602	* We get a ref to the parent, and put the ref when
5603	* this cgroup is being freed, so it's guaranteed
5604	* that the parent won't be destroyed before its
5605	* children.
5606	*/
5607	cgroup_put(cgrp: cgroup_parent(cgrp));
5608	kernfs_put(kn: cgrp->kn);
5609	psi_cgroup_free(cgrp);
5610	kfree(objp: cgrp);
5611	} else {
5612	/*
5613	* This is root cgroup's refcnt reaching zero,
5614	* which indicates that the root should be
5615	* released.
5616	*/
5617	cgroup_destroy_root(root: cgrp->root);
5618	}
5619	}
5620	}
5621
5622	static void css_release_work_fn(struct work_struct *work)
5623	{
5624	struct cgroup_subsys_state *css =
5625	container_of(work, struct cgroup_subsys_state, destroy_work);
5626	struct cgroup_subsys *ss = css->ss;
5627	struct cgroup *cgrp = css->cgroup;
5628
5629	cgroup_lock();
5630
5631	css->flags |= CSS_RELEASED;
5632	list_del_rcu(entry: &css->sibling);
5633
5634	if (!css_is_self(css)) {
5635	struct cgroup *parent_cgrp;
5636
5637	css_rstat_flush(css);
5638
5639	cgroup_idr_replace(idr: &ss->css_idr, NULL, id: css->id);
5640	if (ss->css_released)
5641	ss->css_released(css);
5642
5643	cgrp->nr_dying_subsys[ss->id]--;
5644	/*
5645	* When a css is released and ready to be freed, its
5646	* nr_descendants must be zero. However, the corresponding
5647	* cgrp->nr_dying_subsys[ss->id] may not be 0 if a subsystem
5648	* is activated and deactivated multiple times with one or
5649	* more of its previous activation leaving behind dying csses.
5650	*/
5651	WARN_ON_ONCE(css->nr_descendants);
5652	parent_cgrp = cgroup_parent(cgrp);
5653	while (parent_cgrp) {
5654	parent_cgrp->nr_dying_subsys[ss->id]--;
5655	parent_cgrp = cgroup_parent(cgrp: parent_cgrp);
5656	}
5657	} else {
5658	struct cgroup *tcgrp;
5659
5660	/* cgroup release path */
5661	TRACE_CGROUP_PATH(release, cgrp);
5662
5663	css_rstat_flush(css: &cgrp->self);
5664
5665	spin_lock_irq(lock: &css_set_lock);
5666	for (tcgrp = cgroup_parent(cgrp); tcgrp;
5667	tcgrp = cgroup_parent(cgrp: tcgrp))
5668	tcgrp->nr_dying_descendants--;
5669	spin_unlock_irq(lock: &css_set_lock);
5670
5671	/*
5672	* There are two control paths which try to determine
5673	* cgroup from dentry without going through kernfs -
5674	* cgroupstats_build() and css_tryget_online_from_dir().
5675	* Those are supported by RCU protecting clearing of
5676	* cgrp->kn->priv backpointer.
5677	*/
5678	if (cgrp->kn)
5679	RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5680	NULL);
5681	}
5682
5683	cgroup_unlock();
5684
5685	INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5686	queue_rcu_work(wq: cgroup_free_wq, rwork: &css->destroy_rwork);
5687	}
5688
5689	static void css_release(struct percpu_ref *ref)
5690	{
5691	struct cgroup_subsys_state *css =
5692	container_of(ref, struct cgroup_subsys_state, refcnt);
5693
5694	INIT_WORK(&css->destroy_work, css_release_work_fn);
5695	queue_work(wq: cgroup_release_wq, work: &css->destroy_work);
5696	}
5697
5698	static void init_and_link_css(struct cgroup_subsys_state *css,
5699	struct cgroup_subsys *ss, struct cgroup *cgrp)
5700	{
5701	lockdep_assert_held(&cgroup_mutex);
5702
5703	cgroup_get_live(cgrp);
5704
5705	memset(css, 0, sizeof(*css));
5706	css->cgroup = cgrp;
5707	css->ss = ss;
5708	css->id = -1;
5709	INIT_LIST_HEAD(list: &css->sibling);
5710	INIT_LIST_HEAD(list: &css->children);
5711	css->serial_nr = css_serial_nr_next++;
5712	atomic_set(v: &css->online_cnt, i: 0);
5713
5714	if (cgroup_parent(cgrp)) {
5715	css->parent = cgroup_css(cgrp: cgroup_parent(cgrp), ss);
5716	css_get(css->parent);
5717	}
5718
5719	BUG_ON(cgroup_css(cgrp, ss));
5720	}
5721
5722	/* invoke ->css_online() on a new CSS and mark it online if successful */
5723	static int online_css(struct cgroup_subsys_state *css)
5724	{
5725	struct cgroup_subsys *ss = css->ss;
5726	int ret = 0;
5727
5728	lockdep_assert_held(&cgroup_mutex);
5729
5730	if (ss->css_online)
5731	ret = ss->css_online(css);
5732	if (!ret) {
5733	css->flags |= CSS_ONLINE;
5734	rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5735
5736	atomic_inc(v: &css->online_cnt);
5737	if (css->parent) {
5738	atomic_inc(v: &css->parent->online_cnt);
5739	while ((css = css->parent))
5740	css->nr_descendants++;
5741	}
5742	}
5743	return ret;
5744	}
5745
5746	/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5747	static void offline_css(struct cgroup_subsys_state *css)
5748	{
5749	struct cgroup_subsys *ss = css->ss;
5750
5751	lockdep_assert_held(&cgroup_mutex);
5752
5753	if (!(css->flags & CSS_ONLINE))
5754	return;
5755
5756	if (ss->css_offline)
5757	ss->css_offline(css);
5758
5759	css->flags &= ~CSS_ONLINE;
5760	RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5761
5762	wake_up_all(&css->cgroup->offline_waitq);
5763
5764	css->cgroup->nr_dying_subsys[ss->id]++;
5765	/*
5766	* Parent css and cgroup cannot be freed until after the freeing
5767	* of child css, see css_free_rwork_fn().
5768	*/
5769	while ((css = css->parent)) {
5770	css->nr_descendants--;
5771	css->cgroup->nr_dying_subsys[ss->id]++;
5772	}
5773	}
5774
5775	/**
5776	* css_create - create a cgroup_subsys_state
5777	* @cgrp: the cgroup new css will be associated with
5778	* @ss: the subsys of new css
5779	*
5780	* Create a new css associated with @cgrp - @ss pair. On success, the new
5781	* css is online and installed in @cgrp. This function doesn't create the
5782	* interface files. Returns 0 on success, -errno on failure.
5783	*/
5784	static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5785	struct cgroup_subsys *ss)
5786	{
5787	struct cgroup *parent = cgroup_parent(cgrp);
5788	struct cgroup_subsys_state *parent_css = cgroup_css(cgrp: parent, ss);
5789	struct cgroup_subsys_state *css;
5790	int err;
5791
5792	lockdep_assert_held(&cgroup_mutex);
5793
5794	css = ss->css_alloc(parent_css);
5795	if (!css)
5796	css = ERR_PTR(error: -ENOMEM);
5797	if (IS_ERR(ptr: css))
5798	return css;
5799
5800	init_and_link_css(css, ss, cgrp);
5801
5802	err = percpu_ref_init(ref: &css->refcnt, release: css_release, flags: 0, GFP_KERNEL);
5803	if (err)
5804	goto err_free_css;
5805
5806	err = cgroup_idr_alloc(idr: &ss->css_idr, NULL, start: 2, end: 0, GFP_KERNEL);
5807	if (err < 0)
5808	goto err_free_css;
5809	css->id = err;
5810
5811	err = css_rstat_init(css);
5812	if (err)
5813	goto err_free_css;
5814
5815	/* @css is ready to be brought online now, make it visible */
5816	list_add_tail_rcu(new: &css->sibling, head: &parent_css->children);
5817	cgroup_idr_replace(idr: &ss->css_idr, ptr: css, id: css->id);
5818
5819	err = online_css(css);
5820	if (err)
5821	goto err_list_del;
5822
5823	return css;
5824
5825	err_list_del:
5826	list_del_rcu(entry: &css->sibling);
5827	err_free_css:
5828	INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5829	queue_rcu_work(wq: cgroup_free_wq, rwork: &css->destroy_rwork);
5830	return ERR_PTR(error: err);
5831	}
5832
5833	/*
5834	* The returned cgroup is fully initialized including its control mask, but
5835	* it doesn't have the control mask applied.
5836	*/
5837	static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5838	umode_t mode)
5839	{
5840	struct cgroup_root *root = parent->root;
5841	struct cgroup *cgrp, *tcgrp;
5842	struct kernfs_node *kn;
5843	int i, level = parent->level + 1;
5844	int ret;
5845
5846	/* allocate the cgroup and its ID, 0 is reserved for the root */
5847	cgrp = kzalloc(struct_size(cgrp, _low_ancestors, level), GFP_KERNEL);
5848	if (!cgrp)
5849	return ERR_PTR(error: -ENOMEM);
5850
5851	ret = percpu_ref_init(ref: &cgrp->self.refcnt, release: css_release, flags: 0, GFP_KERNEL);
5852	if (ret)
5853	goto out_free_cgrp;
5854
5855	/* create the directory */
5856	kn = kernfs_create_dir_ns(parent: parent->kn, name, mode,
5857	current_fsuid(), current_fsgid(),
5858	priv: cgrp, NULL);
5859	if (IS_ERR(ptr: kn)) {
5860	ret = PTR_ERR(ptr: kn);
5861	goto out_cancel_ref;
5862	}
5863	cgrp->kn = kn;
5864
5865	init_cgroup_housekeeping(cgrp);
5866
5867	cgrp->self.parent = &parent->self;
5868	cgrp->root = root;
5869	cgrp->level = level;
5870
5871	/*
5872	* Now that init_cgroup_housekeeping() has been called and cgrp->self
5873	* is setup, it is safe to perform rstat initialization on it.
5874	*/
5875	ret = css_rstat_init(css: &cgrp->self);
5876	if (ret)
5877	goto out_kernfs_remove;
5878
5879	ret = psi_cgroup_alloc(cgrp);
5880	if (ret)
5881	goto out_stat_exit;
5882
5883	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(cgrp: tcgrp))
5884	cgrp->ancestors[tcgrp->level] = tcgrp;
5885
5886	/*
5887	* New cgroup inherits effective freeze counter, and
5888	* if the parent has to be frozen, the child has too.
5889	*/
5890	cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5891	seqcount_spinlock_init(&cgrp->freezer.freeze_seq, &css_set_lock);
5892	if (cgrp->freezer.e_freeze) {
5893	/*
5894	* Set the CGRP_FREEZE flag, so when a process will be
5895	* attached to the child cgroup, it will become frozen.
5896	* At this point the new cgroup is unpopulated, so we can
5897	* consider it frozen immediately.
5898	*/
5899	set_bit(nr: CGRP_FREEZE, addr: &cgrp->flags);
5900	cgrp->freezer.freeze_start_nsec = ktime_get_ns();
5901	set_bit(nr: CGRP_FROZEN, addr: &cgrp->flags);
5902	}
5903
5904	if (notify_on_release(cgrp: parent))
5905	set_bit(nr: CGRP_NOTIFY_ON_RELEASE, addr: &cgrp->flags);
5906
5907	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5908	set_bit(nr: CGRP_CPUSET_CLONE_CHILDREN, addr: &cgrp->flags);
5909
5910	cgrp->self.serial_nr = css_serial_nr_next++;
5911
5912	ret = blocking_notifier_call_chain_robust(nh: &cgroup_lifetime_notifier,
5913	val_up: CGROUP_LIFETIME_ONLINE,
5914	val_down: CGROUP_LIFETIME_OFFLINE, v: cgrp);
5915	ret = notifier_to_errno(ret);
5916	if (ret)
5917	goto out_psi_free;
5918
5919	/* allocation complete, commit to creation */
5920	spin_lock_irq(lock: &css_set_lock);
5921	for (i = 0; i < level; i++) {
5922	tcgrp = cgrp->ancestors[i];
5923	tcgrp->nr_descendants++;
5924
5925	/*
5926	* If the new cgroup is frozen, all ancestor cgroups get a new
5927	* frozen descendant, but their state can't change because of
5928	* this.
5929	*/
5930	if (cgrp->freezer.e_freeze)
5931	tcgrp->freezer.nr_frozen_descendants++;
5932	}
5933	spin_unlock_irq(lock: &css_set_lock);
5934
5935	list_add_tail_rcu(new: &cgrp->self.sibling, head: &cgroup_parent(cgrp)->self.children);
5936	atomic_inc(v: &root->nr_cgrps);
5937	cgroup_get_live(cgrp: parent);
5938
5939	/*
5940	* On the default hierarchy, a child doesn't automatically inherit
5941	* subtree_control from the parent. Each is configured manually.
5942	*/
5943	if (!cgroup_on_dfl(cgrp))
5944	cgrp->subtree_control = cgroup_control(cgrp);
5945
5946	cgroup_propagate_control(cgrp);
5947
5948	return cgrp;
5949
5950	out_psi_free:
5951	psi_cgroup_free(cgrp);
5952	out_stat_exit:
5953	css_rstat_exit(css: &cgrp->self);
5954	out_kernfs_remove:
5955	kernfs_remove(kn: cgrp->kn);
5956	out_cancel_ref:
5957	percpu_ref_exit(ref: &cgrp->self.refcnt);
5958	out_free_cgrp:
5959	kfree(objp: cgrp);
5960	return ERR_PTR(error: ret);
5961	}
5962
5963	static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5964	{
5965	struct cgroup *cgroup;
5966	int ret = false;
5967	int level = 0;
5968
5969	lockdep_assert_held(&cgroup_mutex);
5970
5971	for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgrp: cgroup)) {
5972	if (cgroup->nr_descendants >= cgroup->max_descendants)
5973	goto fail;
5974
5975	if (level >= cgroup->max_depth)
5976	goto fail;
5977
5978	level++;
5979	}
5980
5981	ret = true;
5982	fail:
5983	return ret;
5984	}
5985
5986	int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5987	{
5988	struct cgroup *parent, *cgrp;
5989	int ret;
5990
5991	/* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5992	if (strchr(name, '\n'))
5993	return -EINVAL;
5994
5995	parent = cgroup_kn_lock_live(kn: parent_kn, drain_offline: false);
5996	if (!parent)
5997	return -ENODEV;
5998
5999	if (!cgroup_check_hierarchy_limits(parent)) {
6000	ret = -EAGAIN;
6001	goto out_unlock;
6002	}
6003
6004	cgrp = cgroup_create(parent, name, mode);
6005	if (IS_ERR(ptr: cgrp)) {
6006	ret = PTR_ERR(ptr: cgrp);
6007	goto out_unlock;
6008	}
6009
6010	/*
6011	* This extra ref will be put in css_free_rwork_fn() and guarantees
6012	* that @cgrp->kn is always accessible.
6013	*/
6014	kernfs_get(kn: cgrp->kn);
6015
6016	ret = css_populate_dir(css: &cgrp->self);
6017	if (ret)
6018	goto out_destroy;
6019
6020	ret = cgroup_apply_control_enable(cgrp);
6021	if (ret)
6022	goto out_destroy;
6023
6024	TRACE_CGROUP_PATH(mkdir, cgrp);
6025
6026	/* let's create and online css's */
6027	kernfs_activate(kn: cgrp->kn);
6028
6029	ret = 0;
6030	goto out_unlock;
6031
6032	out_destroy:
6033	cgroup_destroy_locked(cgrp);
6034	out_unlock:
6035	cgroup_kn_unlock(kn: parent_kn);
6036	return ret;
6037	}
6038
6039	/*
6040	* This is called when the refcnt of a css is confirmed to be killed.
6041	* css_tryget_online() is now guaranteed to fail. Tell the subsystem to
6042	* initiate destruction and put the css ref from kill_css().
6043	*/
6044	static void css_killed_work_fn(struct work_struct *work)
6045	{
6046	struct cgroup_subsys_state *css =
6047	container_of(work, struct cgroup_subsys_state, destroy_work);
6048
6049	cgroup_lock();
6050
6051	do {
6052	offline_css(css);
6053	css_put(css);
6054	/* @css can't go away while we're holding cgroup_mutex */
6055	css = css->parent;
6056	} while (css && atomic_dec_and_test(v: &css->online_cnt));
6057
6058	cgroup_unlock();
6059	}
6060
6061	/* css kill confirmation processing requires process context, bounce */
6062	static void css_killed_ref_fn(struct percpu_ref *ref)
6063	{
6064	struct cgroup_subsys_state *css =
6065	container_of(ref, struct cgroup_subsys_state, refcnt);
6066
6067	if (atomic_dec_and_test(v: &css->online_cnt)) {
6068	INIT_WORK(&css->destroy_work, css_killed_work_fn);
6069	queue_work(wq: cgroup_offline_wq, work: &css->destroy_work);
6070	}
6071	}
6072
6073	/**
6074	* kill_css - destroy a css
6075	* @css: css to destroy
6076	*
6077	* This function initiates destruction of @css by removing cgroup interface
6078	* files and putting its base reference. ->css_offline() will be invoked
6079	* asynchronously once css_tryget_online() is guaranteed to fail and when
6080	* the reference count reaches zero, @css will be released.
6081	*/
6082	static void kill_css(struct cgroup_subsys_state *css)
6083	{
6084	lockdep_assert_held(&cgroup_mutex);
6085
6086	if (css->flags & CSS_DYING)
6087	return;
6088
6089	/*
6090	* Call css_killed(), if defined, before setting the CSS_DYING flag
6091	*/
6092	if (css->ss->css_killed)
6093	css->ss->css_killed(css);
6094
6095	css->flags |= CSS_DYING;
6096
6097	/*
6098	* This must happen before css is disassociated with its cgroup.
6099	* See seq_css() for details.
6100	*/
6101	css_clear_dir(css);
6102
6103	/*
6104	* Killing would put the base ref, but we need to keep it alive
6105	* until after ->css_offline().
6106	*/
6107	css_get(css);
6108
6109	/*
6110	* cgroup core guarantees that, by the time ->css_offline() is
6111	* invoked, no new css reference will be given out via
6112	* css_tryget_online(). We can't simply call percpu_ref_kill() and
6113	* proceed to offlining css's because percpu_ref_kill() doesn't
6114	* guarantee that the ref is seen as killed on all CPUs on return.
6115	*
6116	* Use percpu_ref_kill_and_confirm() to get notifications as each
6117	* css is confirmed to be seen as killed on all CPUs.
6118	*/
6119	percpu_ref_kill_and_confirm(ref: &css->refcnt, confirm_kill: css_killed_ref_fn);
6120	}
6121
6122	/**
6123	* cgroup_destroy_locked - the first stage of cgroup destruction
6124	* @cgrp: cgroup to be destroyed
6125	*
6126	* css's make use of percpu refcnts whose killing latency shouldn't be
6127	* exposed to userland and are RCU protected. Also, cgroup core needs to
6128	* guarantee that css_tryget_online() won't succeed by the time
6129	* ->css_offline() is invoked. To satisfy all the requirements,
6130	* destruction is implemented in the following two steps.
6131	*
6132	* s1. Verify @cgrp can be destroyed and mark it dying. Remove all
6133	* userland visible parts and start killing the percpu refcnts of
6134	* css's. Set up so that the next stage will be kicked off once all
6135	* the percpu refcnts are confirmed to be killed.
6136	*
6137	* s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
6138	* rest of destruction. Once all cgroup references are gone, the
6139	* cgroup is RCU-freed.
6140	*
6141	* This function implements s1. After this step, @cgrp is gone as far as
6142	* the userland is concerned and a new cgroup with the same name may be
6143	* created. As cgroup doesn't care about the names internally, this
6144	* doesn't cause any problem.
6145	*/
6146	static int cgroup_destroy_locked(struct cgroup *cgrp)
6147	__releases(&cgroup_mutex) __acquires(&cgroup_mutex)
6148	{
6149	struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
6150	struct cgroup_subsys_state *css;
6151	struct cgrp_cset_link *link;
6152	int ssid, ret;
6153
6154	lockdep_assert_held(&cgroup_mutex);
6155
6156	/*
6157	* Only migration can raise populated from zero and we're already
6158	* holding cgroup_mutex.
6159	*/
6160	if (cgroup_is_populated(cgrp))
6161	return -EBUSY;
6162
6163	/*
6164	* Make sure there's no live children. We can't test emptiness of
6165	* ->self.children as dead children linger on it while being
6166	* drained; otherwise, "rmdir parent/child parent" may fail.
6167	*/
6168	if (css_has_online_children(css: &cgrp->self))
6169	return -EBUSY;
6170
6171	/*
6172	* Mark @cgrp and the associated csets dead. The former prevents
6173	* further task migration and child creation by disabling
6174	* cgroup_kn_lock_live(). The latter makes the csets ignored by
6175	* the migration path.
6176	*/
6177	cgrp->self.flags &= ~CSS_ONLINE;
6178
6179	spin_lock_irq(lock: &css_set_lock);
6180	list_for_each_entry(link, &cgrp->cset_links, cset_link)
6181	link->cset->dead = true;
6182	spin_unlock_irq(lock: &css_set_lock);
6183
6184	/* initiate massacre of all css's */
6185	for_each_css(css, ssid, cgrp)
6186	kill_css(css);
6187
6188	/* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
6189	css_clear_dir(css: &cgrp->self);
6190	kernfs_remove(kn: cgrp->kn);
6191
6192	if (cgroup_is_threaded(cgrp))
6193	parent->nr_threaded_children--;
6194
6195	spin_lock_irq(lock: &css_set_lock);
6196	for (tcgrp = parent; tcgrp; tcgrp = cgroup_parent(cgrp: tcgrp)) {
6197	tcgrp->nr_descendants--;
6198	tcgrp->nr_dying_descendants++;
6199	/*
6200	* If the dying cgroup is frozen, decrease frozen descendants
6201	* counters of ancestor cgroups.
6202	*/
6203	if (test_bit(CGRP_FROZEN, &cgrp->flags))
6204	tcgrp->freezer.nr_frozen_descendants--;
6205	}
6206	spin_unlock_irq(lock: &css_set_lock);
6207
6208	cgroup1_check_for_release(cgrp: parent);
6209
6210	ret = blocking_notifier_call_chain(nh: &cgroup_lifetime_notifier,
6211	val: CGROUP_LIFETIME_OFFLINE, v: cgrp);
6212	WARN_ON_ONCE(notifier_to_errno(ret));
6213
6214	/* put the base reference */
6215	percpu_ref_kill(ref: &cgrp->self.refcnt);
6216
6217	return 0;
6218	};
6219
6220	int cgroup_rmdir(struct kernfs_node *kn)
6221	{
6222	struct cgroup *cgrp;
6223	int ret = 0;
6224
6225	cgrp = cgroup_kn_lock_live(kn, drain_offline: false);
6226	if (!cgrp)
6227	return 0;
6228
6229	ret = cgroup_destroy_locked(cgrp);
6230	if (!ret)
6231	TRACE_CGROUP_PATH(rmdir, cgrp);
6232
6233	cgroup_kn_unlock(kn);
6234	return ret;
6235	}
6236
6237	static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
6238	.show_options = cgroup_show_options,
6239	.mkdir = cgroup_mkdir,
6240	.rmdir = cgroup_rmdir,
6241	.show_path = cgroup_show_path,
6242	};
6243
6244	static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
6245	{
6246	struct cgroup_subsys_state *css;
6247
6248	pr_debug("Initializing cgroup subsys %s\n", ss->name);
6249
6250	cgroup_lock();
6251
6252	idr_init(idr: &ss->css_idr);
6253	INIT_LIST_HEAD(list: &ss->cfts);
6254
6255	/* Create the root cgroup state for this subsystem */
6256	ss->root = &cgrp_dfl_root;
6257	css = ss->css_alloc(NULL);
6258	/* We don't handle early failures gracefully */
6259	BUG_ON(IS_ERR(css));
6260	init_and_link_css(css, ss, cgrp: &cgrp_dfl_root.cgrp);
6261
6262	/*
6263	* Root csses are never destroyed and we can't initialize
6264	* percpu_ref during early init. Disable refcnting.
6265	*/
6266	css->flags |= CSS_NO_REF;
6267
6268	if (early) {
6269	/* allocation can't be done safely during early init */
6270	css->id = 1;
6271	} else {
6272	css->id = cgroup_idr_alloc(idr: &ss->css_idr, ptr: css, start: 1, end: 2, GFP_KERNEL);
6273	BUG_ON(css->id < 0);
6274
6275	BUG_ON(ss_rstat_init(ss));
6276	BUG_ON(css_rstat_init(css));
6277	}
6278
6279	/* Update the init_css_set to contain a subsys
6280	* pointer to this state - since the subsystem is
6281	* newly registered, all tasks and hence the
6282	* init_css_set is in the subsystem's root cgroup. */
6283	init_css_set.subsys[ss->id] = css;
6284
6285	have_fork_callback |= (bool)ss->fork << ss->id;
6286	have_exit_callback |= (bool)ss->exit << ss->id;
6287	have_release_callback |= (bool)ss->release << ss->id;
6288	have_canfork_callback |= (bool)ss->can_fork << ss->id;
6289
6290	/* At system boot, before all subsystems have been
6291	* registered, no tasks have been forked, so we don't
6292	* need to invoke fork callbacks here. */
6293	BUG_ON(!list_empty(&init_task.tasks));
6294
6295	BUG_ON(online_css(css));
6296
6297	cgroup_unlock();
6298	}
6299
6300	/**
6301	* cgroup_init_early - cgroup initialization at system boot
6302	*
6303	* Initialize cgroups at system boot, and initialize any
6304	* subsystems that request early init.
6305	*/
6306	int __init cgroup_init_early(void)
6307	{
6308	static struct cgroup_fs_context __initdata ctx;
6309	struct cgroup_subsys *ss;
6310	int i;
6311
6312	ctx.root = &cgrp_dfl_root;
6313	init_cgroup_root(ctx: &ctx);
6314	cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
6315
6316	RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
6317
6318	for_each_subsys(ss, i) {
6319	WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
6320	"invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
6321	i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
6322	ss->id, ss->name);
6323	WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
6324	"cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
6325	WARN(ss->early_init && ss->css_rstat_flush,
6326	"cgroup rstat cannot be used with early init subsystem\n");
6327
6328	ss->id = i;
6329	ss->name = cgroup_subsys_name[i];
6330	if (!ss->legacy_name)
6331	ss->legacy_name = cgroup_subsys_name[i];
6332
6333	if (ss->early_init)
6334	cgroup_init_subsys(ss, early: true);
6335	}
6336	return 0;
6337	}
6338
6339	/**
6340	* cgroup_init - cgroup initialization
6341	*
6342	* Register cgroup filesystem and /proc file, and initialize
6343	* any subsystems that didn't request early init.
6344	*/
6345	int __init cgroup_init(void)
6346	{
6347	struct cgroup_subsys *ss;
6348	int ssid;
6349
6350	BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
6351	BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
6352	BUG_ON(cgroup_init_cftypes(NULL, cgroup_psi_files));
6353	BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
6354
6355	BUG_ON(ss_rstat_init(NULL));
6356
6357	get_user_ns(ns: init_cgroup_ns.user_ns);
6358	cgroup_rt_init();
6359
6360	cgroup_lock();
6361
6362	/*
6363	* Add init_css_set to the hash table so that dfl_root can link to
6364	* it during init.
6365	*/
6366	hash_add(css_set_table, &init_css_set.hlist,
6367	css_set_hash(init_css_set.subsys));
6368
6369	cgroup_bpf_lifetime_notifier_init();
6370
6371	BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
6372
6373	cgroup_unlock();
6374
6375	for_each_subsys(ss, ssid) {
6376	if (ss->early_init) {
6377	struct cgroup_subsys_state *css =
6378	init_css_set.subsys[ss->id];
6379
6380	css->id = cgroup_idr_alloc(idr: &ss->css_idr, ptr: css, start: 1, end: 2,
6381	GFP_KERNEL);
6382	BUG_ON(css->id < 0);
6383	} else {
6384	cgroup_init_subsys(ss, early: false);
6385	}
6386
6387	list_add_tail(new: &init_css_set.e_cset_node[ssid],
6388	head: &cgrp_dfl_root.cgrp.e_csets[ssid]);
6389
6390	/*
6391	* Setting dfl_root subsys_mask needs to consider the
6392	* disabled flag and cftype registration needs kmalloc,
6393	* both of which aren't available during early_init.
6394	*/
6395	if (!cgroup_ssid_enabled(ssid))
6396	continue;
6397
6398	if (cgroup1_ssid_disabled(ssid))
6399	pr_info("Disabling %s control group subsystem in v1 mounts\n",
6400	ss->legacy_name);
6401
6402	cgrp_dfl_root.subsys_mask |= 1 << ss->id;
6403
6404	/* implicit controllers must be threaded too */
6405	WARN_ON(ss->implicit_on_dfl && !ss->threaded);
6406
6407	if (ss->implicit_on_dfl)
6408	cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
6409	else if (!ss->dfl_cftypes)
6410	cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
6411
6412	if (ss->threaded)
6413	cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
6414
6415	if (ss->dfl_cftypes == ss->legacy_cftypes) {
6416	WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
6417	} else {
6418	WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
6419	WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
6420	}
6421
6422	if (ss->bind)
6423	ss->bind(init_css_set.subsys[ssid]);
6424
6425	cgroup_lock();
6426	css_populate_dir(css: init_css_set.subsys[ssid]);
6427	cgroup_unlock();
6428	}
6429
6430	/* init_css_set.subsys[] has been updated, re-hash */
6431	hash_del(node: &init_css_set.hlist);
6432	hash_add(css_set_table, &init_css_set.hlist,
6433	css_set_hash(init_css_set.subsys));
6434
6435	WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
6436	WARN_ON(register_filesystem(&cgroup_fs_type));
6437	WARN_ON(register_filesystem(&cgroup2_fs_type));
6438	WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
6439	#ifdef CONFIG_CPUSETS_V1
6440	WARN_ON(register_filesystem(&cpuset_fs_type));
6441	#endif
6442
6443	ns_tree_add(&init_cgroup_ns);
6444	return 0;
6445	}
6446
6447	static int __init cgroup_wq_init(void)
6448	{
6449	/*
6450	* There isn't much point in executing destruction path in
6451	* parallel. Good chunk is serialized with cgroup_mutex anyway.
6452	* Use 1 for @max_active.
6453	*
6454	* We would prefer to do this in cgroup_init() above, but that
6455	* is called before init_workqueues(): so leave this until after.
6456	*/
6457	cgroup_offline_wq = alloc_workqueue("cgroup_offline", WQ_PERCPU, 1);
6458	BUG_ON(!cgroup_offline_wq);
6459
6460	cgroup_release_wq = alloc_workqueue("cgroup_release", WQ_PERCPU, 1);
6461	BUG_ON(!cgroup_release_wq);
6462
6463	cgroup_free_wq = alloc_workqueue("cgroup_free", WQ_PERCPU, 1);
6464	BUG_ON(!cgroup_free_wq);
6465	return 0;
6466	}
6467	core_initcall(cgroup_wq_init);
6468
6469	void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
6470	{
6471	struct kernfs_node *kn;
6472
6473	kn = kernfs_find_and_get_node_by_id(root: cgrp_dfl_root.kf_root, id);
6474	if (!kn)
6475	return;
6476	kernfs_path(kn, buf, buflen);
6477	kernfs_put(kn);
6478	}
6479
6480	/*
6481	* __cgroup_get_from_id : get the cgroup associated with cgroup id
6482	* @id: cgroup id
6483	* On success return the cgrp or ERR_PTR on failure
6484	* There are no cgroup NS restrictions.
6485	*/
6486	struct cgroup *__cgroup_get_from_id(u64 id)
6487	{
6488	struct kernfs_node *kn;
6489	struct cgroup *cgrp;
6490
6491	kn = kernfs_find_and_get_node_by_id(root: cgrp_dfl_root.kf_root, id);
6492	if (!kn)
6493	return ERR_PTR(error: -ENOENT);
6494
6495	if (kernfs_type(kn) != KERNFS_DIR) {
6496	kernfs_put(kn);
6497	return ERR_PTR(error: -ENOENT);
6498	}
6499
6500	rcu_read_lock();
6501
6502	cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6503	if (cgrp && !cgroup_tryget(cgrp))
6504	cgrp = NULL;
6505
6506	rcu_read_unlock();
6507	kernfs_put(kn);
6508
6509	if (!cgrp)
6510	return ERR_PTR(error: -ENOENT);
6511	return cgrp;
6512	}
6513
6514	/*
6515	* cgroup_get_from_id : get the cgroup associated with cgroup id
6516	* @id: cgroup id
6517	* On success return the cgrp or ERR_PTR on failure
6518	* Only cgroups within current task's cgroup NS are valid.
6519	*/
6520	struct cgroup *cgroup_get_from_id(u64 id)
6521	{
6522	struct cgroup *cgrp, *root_cgrp;
6523
6524	cgrp = __cgroup_get_from_id(id);
6525	if (IS_ERR(ptr: cgrp))
6526	return cgrp;
6527
6528	root_cgrp = current_cgns_cgroup_dfl();
6529	if (!cgroup_is_descendant(cgrp, ancestor: root_cgrp)) {
6530	cgroup_put(cgrp);
6531	return ERR_PTR(error: -ENOENT);
6532	}
6533
6534	return cgrp;
6535	}
6536	EXPORT_SYMBOL_GPL(cgroup_get_from_id);
6537
6538	/*
6539	* proc_cgroup_show()
6540	* - Print task's cgroup paths into seq_file, one line for each hierarchy
6541	* - Used for /proc/<pid>/cgroup.
6542	*/
6543	int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
6544	struct pid *pid, struct task_struct *tsk)
6545	{
6546	char *buf;
6547	int retval;
6548	struct cgroup_root *root;
6549
6550	retval = -ENOMEM;
6551	buf = kmalloc(PATH_MAX, GFP_KERNEL);
6552	if (!buf)
6553	goto out;
6554
6555	rcu_read_lock();
6556	spin_lock_irq(lock: &css_set_lock);
6557
6558	for_each_root(root) {
6559	struct cgroup_subsys *ss;
6560	struct cgroup *cgrp;
6561	int ssid, count = 0;
6562
6563	if (root == &cgrp_dfl_root && !READ_ONCE(cgrp_dfl_visible))
6564	continue;
6565
6566	cgrp = task_cgroup_from_root(task: tsk, root);
6567	/* The root has already been unmounted. */
6568	if (!cgrp)
6569	continue;
6570
6571	seq_printf(m, fmt: "%d:", root->hierarchy_id);
6572	if (root != &cgrp_dfl_root)
6573	for_each_subsys(ss, ssid)
6574	if (root->subsys_mask & (1 << ssid))
6575	seq_printf(m, fmt: "%s%s", count++ ? "," : "",
6576	ss->legacy_name);
6577	if (strlen(root->name))
6578	seq_printf(m, fmt: "%sname=%s", count ? "," : "",
6579	root->name);
6580	seq_putc(m, c: ':');
6581	/*
6582	* On traditional hierarchies, all zombie tasks show up as
6583	* belonging to the root cgroup. On the default hierarchy,
6584	* while a zombie doesn't show up in "cgroup.procs" and
6585	* thus can't be migrated, its /proc/PID/cgroup keeps
6586	* reporting the cgroup it belonged to before exiting. If
6587	* the cgroup is removed before the zombie is reaped,
6588	* " (deleted)" is appended to the cgroup path.
6589	*/
6590	if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
6591	retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
6592	current->nsproxy->cgroup_ns);
6593	if (retval == -E2BIG)
6594	retval = -ENAMETOOLONG;
6595	if (retval < 0)
6596	goto out_unlock;
6597
6598	seq_puts(m, s: buf);
6599	} else {
6600	seq_puts(m, s: "/");
6601	}
6602
6603	if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
6604	seq_puts(m, s: " (deleted)\n");
6605	else
6606	seq_putc(m, c: '\n');
6607	}
6608
6609	retval = 0;
6610	out_unlock:
6611	spin_unlock_irq(lock: &css_set_lock);
6612	rcu_read_unlock();
6613	kfree(objp: buf);
6614	out:
6615	return retval;
6616	}
6617
6618	/**
6619	* cgroup_fork - initialize cgroup related fields during copy_process()
6620	* @child: pointer to task_struct of forking parent process.
6621	*
6622	* A task is associated with the init_css_set until cgroup_post_fork()
6623	* attaches it to the target css_set.
6624	*/
6625	void cgroup_fork(struct task_struct *child)
6626	{
6627	RCU_INIT_POINTER(child->cgroups, &init_css_set);
6628	INIT_LIST_HEAD(list: &child->cg_list);
6629	}
6630
6631	/**
6632	* cgroup_v1v2_get_from_file - get a cgroup pointer from a file pointer
6633	* @f: file corresponding to cgroup_dir
6634	*
6635	* Find the cgroup from a file pointer associated with a cgroup directory.
6636	* Returns a pointer to the cgroup on success. ERR_PTR is returned if the
6637	* cgroup cannot be found.
6638	*/
6639	static struct cgroup *cgroup_v1v2_get_from_file(struct file *f)
6640	{
6641	struct cgroup_subsys_state *css;
6642
6643	css = css_tryget_online_from_dir(dentry: f->f_path.dentry, NULL);
6644	if (IS_ERR(ptr: css))
6645	return ERR_CAST(ptr: css);
6646
6647	return css->cgroup;
6648	}
6649
6650	/**
6651	* cgroup_get_from_file - same as cgroup_v1v2_get_from_file, but only supports
6652	* cgroup2.
6653	* @f: file corresponding to cgroup2_dir
6654	*/
6655	static struct cgroup *cgroup_get_from_file(struct file *f)
6656	{
6657	struct cgroup *cgrp = cgroup_v1v2_get_from_file(f);
6658
6659	if (IS_ERR(ptr: cgrp))
6660	return ERR_CAST(ptr: cgrp);
6661
6662	if (!cgroup_on_dfl(cgrp)) {
6663	cgroup_put(cgrp);
6664	return ERR_PTR(error: -EBADF);
6665	}
6666
6667	return cgrp;
6668	}
6669
6670	/**
6671	* cgroup_css_set_fork - find or create a css_set for a child process
6672	* @kargs: the arguments passed to create the child process
6673	*
6674	* This functions finds or creates a new css_set which the child
6675	* process will be attached to in cgroup_post_fork(). By default,
6676	* the child process will be given the same css_set as its parent.
6677	*
6678	* If CLONE_INTO_CGROUP is specified this function will try to find an
6679	* existing css_set which includes the requested cgroup and if not create
6680	* a new css_set that the child will be attached to later. If this function
6681	* succeeds it will hold cgroup_threadgroup_rwsem on return. If
6682	* CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
6683	* before grabbing cgroup_threadgroup_rwsem and will hold a reference
6684	* to the target cgroup.
6685	*/
6686	static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
6687	__acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
6688	{
6689	int ret;
6690	struct cgroup *dst_cgrp = NULL;
6691	struct css_set *cset;
6692	struct super_block *sb;
6693
6694	if (kargs->flags & CLONE_INTO_CGROUP)
6695	cgroup_lock();
6696
6697	cgroup_threadgroup_change_begin(current);
6698
6699	spin_lock_irq(lock: &css_set_lock);
6700	cset = task_css_set(current);
6701	get_css_set(cset);
6702	if (kargs->cgrp)
6703	kargs->kill_seq = kargs->cgrp->kill_seq;
6704	else
6705	kargs->kill_seq = cset->dfl_cgrp->kill_seq;
6706	spin_unlock_irq(lock: &css_set_lock);
6707
6708	if (!(kargs->flags & CLONE_INTO_CGROUP)) {
6709	kargs->cset = cset;
6710	return 0;
6711	}
6712
6713	CLASS(fd_raw, f)(fd: kargs->cgroup);
6714	if (fd_empty(f)) {
6715	ret = -EBADF;
6716	goto err;
6717	}
6718	sb = fd_file(f)->f_path.dentry->d_sb;
6719
6720	dst_cgrp = cgroup_get_from_file(fd_file(f));
6721	if (IS_ERR(ptr: dst_cgrp)) {
6722	ret = PTR_ERR(ptr: dst_cgrp);
6723	dst_cgrp = NULL;
6724	goto err;
6725	}
6726
6727	if (cgroup_is_dead(cgrp: dst_cgrp)) {
6728	ret = -ENODEV;
6729	goto err;
6730	}
6731
6732	/*
6733	* Verify that we the target cgroup is writable for us. This is
6734	* usually done by the vfs layer but since we're not going through
6735	* the vfs layer here we need to do it "manually".
6736	*/
6737	ret = cgroup_may_write(cgrp: dst_cgrp, sb);
6738	if (ret)
6739	goto err;
6740
6741	/*
6742	* Spawning a task directly into a cgroup works by passing a file
6743	* descriptor to the target cgroup directory. This can even be an O_PATH
6744	* file descriptor. But it can never be a cgroup.procs file descriptor.
6745	* This was done on purpose so spawning into a cgroup could be
6746	* conceptualized as an atomic
6747	*
6748	* fd = openat(dfd_cgroup, "cgroup.procs", ...);
6749	* write(fd, <child-pid>, ...);
6750	*
6751	* sequence, i.e. it's a shorthand for the caller opening and writing
6752	* cgroup.procs of the cgroup indicated by @dfd_cgroup. This allows us
6753	* to always use the caller's credentials.
6754	*/
6755	ret = cgroup_attach_permissions(src_cgrp: cset->dfl_cgrp, dst_cgrp, sb,
6756	threadgroup: !(kargs->flags & CLONE_THREAD),
6757	current->nsproxy->cgroup_ns);
6758	if (ret)
6759	goto err;
6760
6761	kargs->cset = find_css_set(old_cset: cset, cgrp: dst_cgrp);
6762	if (!kargs->cset) {
6763	ret = -ENOMEM;
6764	goto err;
6765	}
6766
6767	put_css_set(cset);
6768	kargs->cgrp = dst_cgrp;
6769	return ret;
6770
6771	err:
6772	cgroup_threadgroup_change_end(current);
6773	cgroup_unlock();
6774	if (dst_cgrp)
6775	cgroup_put(cgrp: dst_cgrp);
6776	put_css_set(cset);
6777	if (kargs->cset)
6778	put_css_set(cset: kargs->cset);
6779	return ret;
6780	}
6781
6782	/**
6783	* cgroup_css_set_put_fork - drop references we took during fork
6784	* @kargs: the arguments passed to create the child process
6785	*
6786	* Drop references to the prepared css_set and target cgroup if
6787	* CLONE_INTO_CGROUP was requested.
6788	*/
6789	static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
6790	__releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6791	{
6792	struct cgroup *cgrp = kargs->cgrp;
6793	struct css_set *cset = kargs->cset;
6794
6795	cgroup_threadgroup_change_end(current);
6796
6797	if (cset) {
6798	put_css_set(cset);
6799	kargs->cset = NULL;
6800	}
6801
6802	if (kargs->flags & CLONE_INTO_CGROUP) {
6803	cgroup_unlock();
6804	if (cgrp) {
6805	cgroup_put(cgrp);
6806	kargs->cgrp = NULL;
6807	}
6808	}
6809	}
6810
6811	/**
6812	* cgroup_can_fork - called on a new task before the process is exposed
6813	* @child: the child process
6814	* @kargs: the arguments passed to create the child process
6815	*
6816	* This prepares a new css_set for the child process which the child will
6817	* be attached to in cgroup_post_fork().
6818	* This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6819	* callback returns an error, the fork aborts with that error code. This
6820	* allows for a cgroup subsystem to conditionally allow or deny new forks.
6821	*/
6822	int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6823	{
6824	struct cgroup_subsys *ss;
6825	int i, j, ret;
6826
6827	ret = cgroup_css_set_fork(kargs);
6828	if (ret)
6829	return ret;
6830
6831	do_each_subsys_mask(ss, i, have_canfork_callback) {
6832	ret = ss->can_fork(child, kargs->cset);
6833	if (ret)
6834	goto out_revert;
6835	} while_each_subsys_mask();
6836
6837	return 0;
6838
6839	out_revert:
6840	for_each_subsys(ss, j) {
6841	if (j >= i)
6842	break;
6843	if (ss->cancel_fork)
6844	ss->cancel_fork(child, kargs->cset);
6845	}
6846
6847	cgroup_css_set_put_fork(kargs);
6848
6849	return ret;
6850	}
6851
6852	/**
6853	* cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6854	* @child: the child process
6855	* @kargs: the arguments passed to create the child process
6856	*
6857	* This calls the cancel_fork() callbacks if a fork failed *after*
6858	* cgroup_can_fork() succeeded and cleans up references we took to
6859	* prepare a new css_set for the child process in cgroup_can_fork().
6860	*/
6861	void cgroup_cancel_fork(struct task_struct *child,
6862	struct kernel_clone_args *kargs)
6863	{
6864	struct cgroup_subsys *ss;
6865	int i;
6866
6867	for_each_subsys(ss, i)
6868	if (ss->cancel_fork)
6869	ss->cancel_fork(child, kargs->cset);
6870
6871	cgroup_css_set_put_fork(kargs);
6872	}
6873
6874	/**
6875	* cgroup_post_fork - finalize cgroup setup for the child process
6876	* @child: the child process
6877	* @kargs: the arguments passed to create the child process
6878	*
6879	* Attach the child process to its css_set calling the subsystem fork()
6880	* callbacks.
6881	*/
6882	void cgroup_post_fork(struct task_struct *child,
6883	struct kernel_clone_args *kargs)
6884	__releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6885	{
6886	unsigned int cgrp_kill_seq = 0;
6887	unsigned long cgrp_flags = 0;
6888	bool kill = false;
6889	struct cgroup_subsys *ss;
6890	struct css_set *cset;
6891	int i;
6892
6893	cset = kargs->cset;
6894	kargs->cset = NULL;
6895
6896	spin_lock_irq(lock: &css_set_lock);
6897
6898	/* init tasks are special, only link regular threads */
6899	if (likely(child->pid)) {
6900	if (kargs->cgrp) {
6901	cgrp_flags = kargs->cgrp->flags;
6902	cgrp_kill_seq = kargs->cgrp->kill_seq;
6903	} else {
6904	cgrp_flags = cset->dfl_cgrp->flags;
6905	cgrp_kill_seq = cset->dfl_cgrp->kill_seq;
6906	}
6907
6908	WARN_ON_ONCE(!list_empty(&child->cg_list));
6909	cset->nr_tasks++;
6910	css_set_move_task(task: child, NULL, to_cset: cset, use_mg_tasks: false);
6911	} else {
6912	put_css_set(cset);
6913	cset = NULL;
6914	}
6915
6916	if (!(child->flags & PF_KTHREAD)) {
6917	if (unlikely(test_bit(CGRP_FREEZE, &cgrp_flags))) {
6918	/*
6919	* If the cgroup has to be frozen, the new task has
6920	* too. Let's set the JOBCTL_TRAP_FREEZE jobctl bit to
6921	* get the task into the frozen state.
6922	*/
6923	spin_lock(lock: &child->sighand->siglock);
6924	WARN_ON_ONCE(child->frozen);
6925	child->jobctl |= JOBCTL_TRAP_FREEZE;
6926	spin_unlock(lock: &child->sighand->siglock);
6927
6928	/*
6929	* Calling cgroup_update_frozen() isn't required here,
6930	* because it will be called anyway a bit later from
6931	* do_freezer_trap(). So we avoid cgroup's transient
6932	* switch from the frozen state and back.
6933	*/
6934	}
6935
6936	/*
6937	* If the cgroup is to be killed notice it now and take the
6938	* child down right after we finished preparing it for
6939	* userspace.
6940	*/
6941	kill = kargs->kill_seq != cgrp_kill_seq;
6942	}
6943
6944	spin_unlock_irq(lock: &css_set_lock);
6945
6946	/*
6947	* Call ss->fork(). This must happen after @child is linked on
6948	* css_set; otherwise, @child might change state between ->fork()
6949	* and addition to css_set.
6950	*/
6951	do_each_subsys_mask(ss, i, have_fork_callback) {
6952	ss->fork(child);
6953	} while_each_subsys_mask();
6954
6955	/* Make the new cset the root_cset of the new cgroup namespace. */
6956	if (kargs->flags & CLONE_NEWCGROUP) {
6957	struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6958
6959	get_css_set(cset);
6960	child->nsproxy->cgroup_ns->root_cset = cset;
6961	put_css_set(cset: rcset);
6962	}
6963
6964	/* Cgroup has to be killed so take down child immediately. */
6965	if (unlikely(kill))
6966	do_send_sig_info(SIGKILL, SEND_SIG_NOINFO, p: child, type: PIDTYPE_TGID);
6967
6968	cgroup_css_set_put_fork(kargs);
6969	}
6970
6971	/**
6972	* cgroup_task_exit - detach cgroup from exiting task
6973	* @tsk: pointer to task_struct of exiting process
6974	*
6975	* Description: Detach cgroup from @tsk.
6976	*
6977	*/
6978	void cgroup_task_exit(struct task_struct *tsk)
6979	{
6980	struct cgroup_subsys *ss;
6981	int i;
6982
6983	/* see cgroup_post_fork() for details */
6984	do_each_subsys_mask(ss, i, have_exit_callback) {
6985	ss->exit(tsk);
6986	} while_each_subsys_mask();
6987	}
6988
6989	static void do_cgroup_task_dead(struct task_struct *tsk)
6990	{
6991	struct css_set *cset;
6992	unsigned long flags;
6993
6994	spin_lock_irqsave(&css_set_lock, flags);
6995
6996	WARN_ON_ONCE(list_empty(&tsk->cg_list));
6997	cset = task_css_set(task: tsk);
6998	css_set_move_task(task: tsk, from_cset: cset, NULL, use_mg_tasks: false);
6999	cset->nr_tasks--;
7000	/* matches the signal->live check in css_task_iter_advance() */
7001	if (thread_group_leader(p: tsk) && atomic_read(v: &tsk->signal->live))
7002	list_add_tail(new: &tsk->cg_list, head: &cset->dying_tasks);
7003
7004	if (dl_task(p: tsk))
7005	dec_dl_tasks_cs(task: tsk);
7006
7007	WARN_ON_ONCE(cgroup_task_frozen(tsk));
7008	if (unlikely(!(tsk->flags & PF_KTHREAD) &&
7009	test_bit(CGRP_FREEZE, &task_dfl_cgroup(tsk)->flags)))
7010	cgroup_update_frozen(cgrp: task_dfl_cgroup(task: tsk));
7011
7012	spin_unlock_irqrestore(lock: &css_set_lock, flags);
7013	}
7014
7015	#ifdef CONFIG_PREEMPT_RT
7016	/*
7017	* cgroup_task_dead() is called from finish_task_switch() which doesn't allow
7018	* scheduling even in RT. As the task_dead path requires grabbing css_set_lock,
7019	* this lead to sleeping in the invalid context warning bug. css_set_lock is too
7020	* big to become a raw_spinlock. The task_dead path doesn't need to run
7021	* synchronously but can't be delayed indefinitely either as the dead task pins
7022	* the cgroup and task_struct can be pinned indefinitely. Bounce through lazy
7023	* irq_work to allow batching while ensuring timely completion.
7024	*/
7025	static DEFINE_PER_CPU(struct llist_head, cgrp_dead_tasks);
7026	static DEFINE_PER_CPU(struct irq_work, cgrp_dead_tasks_iwork);
7027
7028	static void cgrp_dead_tasks_iwork_fn(struct irq_work *iwork)
7029	{
7030	struct llist_node *lnode;
7031	struct task_struct *task, *next;
7032
7033	lnode = llist_del_all(this_cpu_ptr(&cgrp_dead_tasks));
7034	llist_for_each_entry_safe(task, next, lnode, cg_dead_lnode) {
7035	do_cgroup_task_dead(task);
7036	put_task_struct(task);
7037	}
7038	}
7039
7040	static void __init cgroup_rt_init(void)
7041	{
7042	int cpu;
7043
7044	for_each_possible_cpu(cpu) {
7045	init_llist_head(per_cpu_ptr(&cgrp_dead_tasks, cpu));
7046	per_cpu(cgrp_dead_tasks_iwork, cpu) =
7047	IRQ_WORK_INIT_LAZY(cgrp_dead_tasks_iwork_fn);
7048	}
7049	}
7050
7051	void cgroup_task_dead(struct task_struct *task)
7052	{
7053	get_task_struct(task);
7054	llist_add(&task->cg_dead_lnode, this_cpu_ptr(&cgrp_dead_tasks));
7055	irq_work_queue(this_cpu_ptr(&cgrp_dead_tasks_iwork));
7056	}
7057	#else /* CONFIG_PREEMPT_RT */
7058	static void __init cgroup_rt_init(void) {}
7059
7060	void cgroup_task_dead(struct task_struct *task)
7061	{
7062	do_cgroup_task_dead(tsk: task);
7063	}
7064	#endif /* CONFIG_PREEMPT_RT */
7065
7066	void cgroup_task_release(struct task_struct *task)
7067	{
7068	struct cgroup_subsys *ss;
7069	int ssid;
7070
7071	do_each_subsys_mask(ss, ssid, have_release_callback) {
7072	ss->release(task);
7073	} while_each_subsys_mask();
7074	}
7075
7076	void cgroup_task_free(struct task_struct *task)
7077	{
7078	struct css_set *cset = task_css_set(task);
7079
7080	if (!list_empty(head: &task->cg_list)) {
7081	spin_lock_irq(lock: &css_set_lock);
7082	css_set_skip_task_iters(cset: task_css_set(task), task);
7083	list_del_init(entry: &task->cg_list);
7084	spin_unlock_irq(lock: &css_set_lock);
7085	}
7086
7087	put_css_set(cset);
7088	}
7089
7090	static int __init cgroup_disable(char *str)
7091	{
7092	struct cgroup_subsys *ss;
7093	char *token;
7094	int i;
7095
7096	while ((token = strsep(&str, ",")) != NULL) {
7097	if (!*token)
7098	continue;
7099
7100	for_each_subsys(ss, i) {
7101	if (strcmp(token, ss->name) &&
7102	strcmp(token, ss->legacy_name))
7103	continue;
7104
7105	static_branch_disable(cgroup_subsys_enabled_key[i]);
7106	pr_info("Disabling %s control group subsystem\n",
7107	ss->name);
7108	}
7109
7110	for (i = 0; i < OPT_FEATURE_COUNT; i++) {
7111	if (strcmp(token, cgroup_opt_feature_names[i]))
7112	continue;
7113	cgroup_feature_disable_mask |= 1 << i;
7114	pr_info("Disabling %s control group feature\n",
7115	cgroup_opt_feature_names[i]);
7116	break;
7117	}
7118	}
7119	return 1;
7120	}
7121	__setup("cgroup_disable=", cgroup_disable);
7122
7123	void __init __weak enable_debug_cgroup(void) { }
7124
7125	static int __init enable_cgroup_debug(char *str)
7126	{
7127	cgroup_debug = true;
7128	enable_debug_cgroup();
7129	return 1;
7130	}
7131	__setup("cgroup_debug", enable_cgroup_debug);
7132
7133	static int __init cgroup_favordynmods_setup(char *str)
7134	{
7135	return (kstrtobool(s: str, res: &have_favordynmods) == 0);
7136	}
7137	__setup("cgroup_favordynmods=", cgroup_favordynmods_setup);
7138
7139	/**
7140	* css_tryget_online_from_dir - get corresponding css from a cgroup dentry
7141	* @dentry: directory dentry of interest
7142	* @ss: subsystem of interest
7143	*
7144	* If @dentry is a directory for a cgroup which has @ss enabled on it, try
7145	* to get the corresponding css and return it. If such css doesn't exist
7146	* or can't be pinned, an ERR_PTR value is returned.
7147	*/
7148	struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
7149	struct cgroup_subsys *ss)
7150	{
7151	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
7152	struct file_system_type *s_type = dentry->d_sb->s_type;
7153	struct cgroup_subsys_state *css = NULL;
7154	struct cgroup *cgrp;
7155
7156	/* is @dentry a cgroup dir? */
7157	if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
7158	!kn || kernfs_type(kn) != KERNFS_DIR)
7159	return ERR_PTR(error: -EBADF);
7160
7161	rcu_read_lock();
7162
7163	/*
7164	* This path doesn't originate from kernfs and @kn could already
7165	* have been or be removed at any point. @kn->priv is RCU
7166	* protected for this access. See css_release_work_fn() for details.
7167	*/
7168	cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
7169	if (cgrp)
7170	css = cgroup_css(cgrp, ss);
7171
7172	if (!css || !css_tryget_online(css))
7173	css = ERR_PTR(error: -ENOENT);
7174
7175	rcu_read_unlock();
7176	return css;
7177	}
7178
7179	/**
7180	* css_from_id - lookup css by id
7181	* @id: the cgroup id
7182	* @ss: cgroup subsys to be looked into
7183	*
7184	* Returns the css if there's valid one with @id, otherwise returns NULL.
7185	* Should be called under rcu_read_lock().
7186	*/
7187	struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
7188	{
7189	WARN_ON_ONCE(!rcu_read_lock_held());
7190	return idr_find(&ss->css_idr, id);
7191	}
7192
7193	/**
7194	* cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
7195	* @path: path on the default hierarchy
7196	*
7197	* Find the cgroup at @path on the default hierarchy, increment its
7198	* reference count and return it. Returns pointer to the found cgroup on
7199	* success, ERR_PTR(-ENOENT) if @path doesn't exist or if the cgroup has already
7200	* been released and ERR_PTR(-ENOTDIR) if @path points to a non-directory.
7201	*/
7202	struct cgroup *cgroup_get_from_path(const char *path)
7203	{
7204	struct kernfs_node *kn;
7205	struct cgroup *cgrp = ERR_PTR(error: -ENOENT);
7206	struct cgroup *root_cgrp;
7207
7208	root_cgrp = current_cgns_cgroup_dfl();
7209	kn = kernfs_walk_and_get(kn: root_cgrp->kn, path);
7210	if (!kn)
7211	goto out;
7212
7213	if (kernfs_type(kn) != KERNFS_DIR) {
7214	cgrp = ERR_PTR(error: -ENOTDIR);
7215	goto out_kernfs;
7216	}
7217
7218	rcu_read_lock();
7219
7220	cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
7221	if (!cgrp || !cgroup_tryget(cgrp))
7222	cgrp = ERR_PTR(error: -ENOENT);
7223
7224	rcu_read_unlock();
7225
7226	out_kernfs:
7227	kernfs_put(kn);
7228	out:
7229	return cgrp;
7230	}
7231	EXPORT_SYMBOL_GPL(cgroup_get_from_path);
7232
7233	/**
7234	* cgroup_v1v2_get_from_fd - get a cgroup pointer from a fd
7235	* @fd: fd obtained by open(cgroup_dir)
7236	*
7237	* Find the cgroup from a fd which should be obtained
7238	* by opening a cgroup directory. Returns a pointer to the
7239	* cgroup on success. ERR_PTR is returned if the cgroup
7240	* cannot be found.
7241	*/
7242	struct cgroup *cgroup_v1v2_get_from_fd(int fd)
7243	{
7244	CLASS(fd_raw, f)(fd);
7245	if (fd_empty(f))
7246	return ERR_PTR(error: -EBADF);
7247
7248	return cgroup_v1v2_get_from_file(fd_file(f));
7249	}
7250
7251	/**
7252	* cgroup_get_from_fd - same as cgroup_v1v2_get_from_fd, but only supports
7253	* cgroup2.
7254	* @fd: fd obtained by open(cgroup2_dir)
7255	*/
7256	struct cgroup *cgroup_get_from_fd(int fd)
7257	{
7258	struct cgroup *cgrp = cgroup_v1v2_get_from_fd(fd);
7259
7260	if (IS_ERR(ptr: cgrp))
7261	return ERR_CAST(ptr: cgrp);
7262
7263	if (!cgroup_on_dfl(cgrp)) {
7264	cgroup_put(cgrp);
7265	return ERR_PTR(error: -EBADF);
7266	}
7267	return cgrp;
7268	}
7269	EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
7270
7271	static u64 power_of_ten(int power)
7272	{
7273	u64 v = 1;
7274	while (power--)
7275	v *= 10;
7276	return v;
7277	}
7278
7279	/**
7280	* cgroup_parse_float - parse a floating number
7281	* @input: input string
7282	* @dec_shift: number of decimal digits to shift
7283	* @v: output
7284	*
7285	* Parse a decimal floating point number in @input and store the result in
7286	* @v with decimal point right shifted @dec_shift times. For example, if
7287	* @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
7288	* Returns 0 on success, -errno otherwise.
7289	*
7290	* There's nothing cgroup specific about this function except that it's
7291	* currently the only user.
7292	*/
7293	int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
7294	{
7295	s64 whole, frac = 0;
7296	int fstart = 0, fend = 0, flen;
7297
7298	if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
7299	return -EINVAL;
7300	if (frac < 0)
7301	return -EINVAL;
7302
7303	flen = fend > fstart ? fend - fstart : 0;
7304	if (flen < dec_shift)
7305	frac *= power_of_ten(power: dec_shift - flen);
7306	else
7307	frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
7308
7309	*v = whole * power_of_ten(power: dec_shift) + frac;
7310	return 0;
7311	}
7312
7313	/*
7314	* sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
7315	* definition in cgroup-defs.h.
7316	*/
7317	#ifdef CONFIG_SOCK_CGROUP_DATA
7318
7319	void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
7320	{
7321	struct cgroup *cgroup;
7322
7323	rcu_read_lock();
7324	/* Don't associate the sock with unrelated interrupted task's cgroup. */
7325	if (in_interrupt()) {
7326	cgroup = &cgrp_dfl_root.cgrp;
7327	cgroup_get(cgrp: cgroup);
7328	goto out;
7329	}
7330
7331	while (true) {
7332	struct css_set *cset;
7333
7334	cset = task_css_set(current);
7335	if (likely(cgroup_tryget(cset->dfl_cgrp))) {
7336	cgroup = cset->dfl_cgrp;
7337	break;
7338	}
7339	cpu_relax();
7340	}
7341	out:
7342	skcd->cgroup = cgroup;
7343	cgroup_bpf_get(cgrp: cgroup);
7344	rcu_read_unlock();
7345	}
7346
7347	void cgroup_sk_clone(struct sock_cgroup_data *skcd)
7348	{
7349	struct cgroup *cgrp = sock_cgroup_ptr(skcd);
7350
7351	/*
7352	* We might be cloning a socket which is left in an empty
7353	* cgroup and the cgroup might have already been rmdir'd.
7354	* Don't use cgroup_get_live().
7355	*/
7356	cgroup_get(cgrp);
7357	cgroup_bpf_get(cgrp);
7358	}
7359
7360	void cgroup_sk_free(struct sock_cgroup_data *skcd)
7361	{
7362	struct cgroup *cgrp = sock_cgroup_ptr(skcd);
7363
7364	cgroup_bpf_put(cgrp);
7365	cgroup_put(cgrp);
7366	}
7367
7368	#endif /* CONFIG_SOCK_CGROUP_DATA */
7369
7370	#ifdef CONFIG_SYSFS
7371	static ssize_t show_delegatable_files(struct cftype *files, char *buf,
7372	ssize_t size, const char *prefix)
7373	{
7374	struct cftype *cft;
7375	ssize_t ret = 0;
7376
7377	for (cft = files; cft && cft->name[0] != '\0'; cft++) {
7378	if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
7379	continue;
7380
7381	if (prefix)
7382	ret += snprintf(buf: buf + ret, size: size - ret, fmt: "%s.", prefix);
7383
7384	ret += snprintf(buf: buf + ret, size: size - ret, fmt: "%s\n", cft->name);
7385
7386	if (WARN_ON(ret >= size))
7387	break;
7388	}
7389
7390	return ret;
7391	}
7392
7393	static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
7394	char *buf)
7395	{
7396	struct cgroup_subsys *ss;
7397	int ssid;
7398	ssize_t ret = 0;
7399
7400	ret = show_delegatable_files(files: cgroup_base_files, buf: buf + ret,
7401	PAGE_SIZE - ret, NULL);
7402	if (cgroup_psi_enabled())
7403	ret += show_delegatable_files(files: cgroup_psi_files, buf: buf + ret,
7404	PAGE_SIZE - ret, NULL);
7405
7406	for_each_subsys(ss, ssid)
7407	ret += show_delegatable_files(files: ss->dfl_cftypes, buf: buf + ret,
7408	PAGE_SIZE - ret,
7409	prefix: cgroup_subsys_name[ssid]);
7410
7411	return ret;
7412	}
7413	static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
7414
7415	static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
7416	char *buf)
7417	{
7418	return snprintf(buf, PAGE_SIZE,
7419	fmt: "nsdelegate\n"
7420	"favordynmods\n"
7421	"memory_localevents\n"
7422	"memory_recursiveprot\n"
7423	"memory_hugetlb_accounting\n"
7424	"pids_localevents\n");
7425	}
7426	static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
7427
7428	static struct attribute *cgroup_sysfs_attrs[] = {
7429	&cgroup_delegate_attr.attr,
7430	&cgroup_features_attr.attr,
7431	NULL,
7432	};
7433
7434	static const struct attribute_group cgroup_sysfs_attr_group = {
7435	.attrs = cgroup_sysfs_attrs,
7436	.name = "cgroup",
7437	};
7438
7439	static int __init cgroup_sysfs_init(void)
7440	{
7441	return sysfs_create_group(kobj: kernel_kobj, grp: &cgroup_sysfs_attr_group);
7442	}
7443	subsys_initcall(cgroup_sysfs_init);
7444
7445	#endif /* CONFIG_SYSFS */
7446