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