path: root/kernel/cgroup.c (plain)
blob: 4585b7ddf267b4967d1d9943bce3b739c5147577

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 <linux/cgroup.h>
32	#include <linux/cred.h>
33	#include <linux/ctype.h>
34	#include <linux/errno.h>
35	#include <linux/init_task.h>
36	#include <linux/kernel.h>
37	#include <linux/list.h>
38	#include <linux/magic.h>
39	#include <linux/mm.h>
40	#include <linux/mutex.h>
41	#include <linux/mount.h>
42	#include <linux/pagemap.h>
43	#include <linux/proc_fs.h>
44	#include <linux/rcupdate.h>
45	#include <linux/sched.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/sort.h>
51	#include <linux/kmod.h>
52	#include <linux/delayacct.h>
53	#include <linux/cgroupstats.h>
54	#include <linux/hashtable.h>
55	#include <linux/pid_namespace.h>
56	#include <linux/idr.h>
57	#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58	#include <linux/kthread.h>
59	#include <linux/delay.h>
60	#include <linux/atomic.h>
61	#include <linux/cpuset.h>
62	#include <linux/proc_ns.h>
63	#include <linux/nsproxy.h>
64	#include <linux/file.h>
65	#include <linux/psi.h>
66	#include <net/sock.h>
67
68	#define CREATE_TRACE_POINTS
69	#include <trace/events/cgroup.h>
70
71	/*
72	* pidlists linger the following amount before being destroyed. The goal
73	* is avoiding frequent destruction in the middle of consecutive read calls
74	* Expiring in the middle is a performance problem not a correctness one.
75	* 1 sec should be enough.
76	*/
77	#define CGROUP_PIDLIST_DESTROY_DELAY HZ
78
79	#define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
80	MAX_CFTYPE_NAME + 2)
81
82	/*
83	* cgroup_mutex is the master lock. Any modification to cgroup or its
84	* hierarchy must be performed while holding it.
85	*
86	* css_set_lock protects task->cgroups pointer, the list of css_set
87	* objects, and the chain of tasks off each css_set.
88	*
89	* These locks are exported if CONFIG_PROVE_RCU so that accessors in
90	* cgroup.h can use them for lockdep annotations.
91	*/
92	#ifdef CONFIG_PROVE_RCU
93	DEFINE_MUTEX(cgroup_mutex);
94	DEFINE_SPINLOCK(css_set_lock);
95	EXPORT_SYMBOL_GPL(cgroup_mutex);
96	EXPORT_SYMBOL_GPL(css_set_lock);
97	#else
98	static DEFINE_MUTEX(cgroup_mutex);
99	static DEFINE_SPINLOCK(css_set_lock);
100	#endif
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	/*
115	* Protects cgroup_subsys->release_agent_path. Modifying it also requires
116	* cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
117	*/
118	static DEFINE_SPINLOCK(release_agent_path_lock);
119
120	struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
121
122	#define cgroup_assert_mutex_or_rcu_locked() \
123	RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
124	!lockdep_is_held(&cgroup_mutex), \
125	"cgroup_mutex or RCU read lock required");
126
127	/*
128	* cgroup destruction makes heavy use of work items and there can be a lot
129	* of concurrent destructions. Use a separate workqueue so that cgroup
130	* destruction work items don't end up filling up max_active of system_wq
131	* which may lead to deadlock.
132	*/
133	static struct workqueue_struct *cgroup_destroy_wq;
134
135	/*
136	* pidlist destructions need to be flushed on cgroup destruction. Use a
137	* separate workqueue as flush domain.
138	*/
139	static struct workqueue_struct *cgroup_pidlist_destroy_wq;
140
141	/* generate an array of cgroup subsystem pointers */
142	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
143	static struct cgroup_subsys *cgroup_subsys[] = {
144	#include <linux/cgroup_subsys.h>
145	};
146	#undef SUBSYS
147
148	/* array of cgroup subsystem names */
149	#define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
150	static const char *cgroup_subsys_name[] = {
151	#include <linux/cgroup_subsys.h>
152	};
153	#undef SUBSYS
154
155	/* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
156	#define SUBSYS(_x) \
157	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
158	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
159	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
160	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
161	#include <linux/cgroup_subsys.h>
162	#undef SUBSYS
163
164	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
165	static struct static_key_true *cgroup_subsys_enabled_key[] = {
166	#include <linux/cgroup_subsys.h>
167	};
168	#undef SUBSYS
169
170	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
171	static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
172	#include <linux/cgroup_subsys.h>
173	};
174	#undef SUBSYS
175
176	/*
177	* The default hierarchy, reserved for the subsystems that are otherwise
178	* unattached - it never has more than a single cgroup, and all tasks are
179	* part of that cgroup.
180	*/
181	struct cgroup_root cgrp_dfl_root;
182	EXPORT_SYMBOL_GPL(cgrp_dfl_root);
183
184	/*
185	* The default hierarchy always exists but is hidden until mounted for the
186	* first time. This is for backward compatibility.
187	*/
188	static bool cgrp_dfl_visible;
189
190	/* Controllers blocked by the commandline in v1 */
191	static u16 cgroup_no_v1_mask;
192
193	/* some controllers are not supported in the default hierarchy */
194	static u16 cgrp_dfl_inhibit_ss_mask;
195
196	/* some controllers are implicitly enabled on the default hierarchy */
197	static unsigned long cgrp_dfl_implicit_ss_mask;
198
199	/* The list of hierarchy roots */
200
201	static LIST_HEAD(cgroup_roots);
202	static int cgroup_root_count;
203
204	/* hierarchy ID allocation and mapping, protected by cgroup_mutex */
205	static DEFINE_IDR(cgroup_hierarchy_idr);
206
207	/*
208	* Assign a monotonically increasing serial number to csses. It guarantees
209	* cgroups with bigger numbers are newer than those with smaller numbers.
210	* Also, as csses are always appended to the parent's ->children list, it
211	* guarantees that sibling csses are always sorted in the ascending serial
212	* number order on the list. Protected by cgroup_mutex.
213	*/
214	static u64 css_serial_nr_next = 1;
215
216	/*
217	* These bitmask flags indicate whether tasks in the fork and exit paths have
218	* fork/exit handlers to call. This avoids us having to do extra work in the
219	* fork/exit path to check which subsystems have fork/exit callbacks.
220	*/
221	static u16 have_fork_callback __read_mostly;
222	static u16 have_exit_callback __read_mostly;
223	static u16 have_free_callback __read_mostly;
224
225	/* cgroup namespace for init task */
226	struct cgroup_namespace init_cgroup_ns = {
227	.count = { .counter = 2, },
228	.user_ns = &init_user_ns,
229	.ns.ops = &cgroupns_operations,
230	.ns.inum = PROC_CGROUP_INIT_INO,
231	.root_cset = &init_css_set,
232	};
233
234	/* Ditto for the can_fork callback. */
235	static u16 have_canfork_callback __read_mostly;
236
237	static struct file_system_type cgroup2_fs_type;
238	static struct cftype cgroup_dfl_base_files[];
239	static struct cftype cgroup_legacy_base_files[];
240
241	static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
242	static void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
243	static int cgroup_apply_control(struct cgroup *cgrp);
244	static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
245	static void css_task_iter_advance(struct css_task_iter *it);
246	static int cgroup_destroy_locked(struct cgroup *cgrp);
247	static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
248	struct cgroup_subsys *ss);
249	static void css_release(struct percpu_ref *ref);
250	static void kill_css(struct cgroup_subsys_state *css);
251	static int cgroup_addrm_files(struct cgroup_subsys_state *css,
252	struct cgroup *cgrp, struct cftype cfts[],
253	bool is_add);
254
255	/**
256	* cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
257	* @ssid: subsys ID of interest
258	*
259	* cgroup_subsys_enabled() can only be used with literal subsys names which
260	* is fine for individual subsystems but unsuitable for cgroup core. This
261	* is slower static_key_enabled() based test indexed by @ssid.
262	*/
263	static bool cgroup_ssid_enabled(int ssid)
264	{
265	if (CGROUP_SUBSYS_COUNT == 0)
266	return false;
267
268	return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
269	}
270
271	static bool cgroup_ssid_no_v1(int ssid)
272	{
273	return cgroup_no_v1_mask & (1 << 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 differnetly depending on the
283	* interface version.
284	*
285	* The set of behaviors which change on the default hierarchy are still
286	* being determined and the mount option is prefixed with __DEVEL__.
287	*
288	* List of changed behaviors:
289	*
290	* - Mount options "noprefix", "xattr", "clone_children", "release_agent"
291	* and "name" are disallowed.
292	*
293	* - When mounting an existing superblock, mount options should match.
294	*
295	* - Remount is disallowed.
296	*
297	* - rename(2) is disallowed.
298	*
299	* - "tasks" is removed. Everything should be at process granularity. Use
300	* "cgroup.procs" instead.
301	*
302	* - "cgroup.procs" is not sorted. pids will be unique unless they got
303	* recycled inbetween reads.
304	*
305	* - "release_agent" and "notify_on_release" are removed. Replacement
306	* notification mechanism will be implemented.
307	*
308	* - "cgroup.clone_children" is removed.
309	*
310	* - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
311	* and its descendants contain no task; otherwise, 1. The file also
312	* generates kernfs notification which can be monitored through poll and
313	* [di]notify when the value of the file changes.
314	*
315	* - cpuset: tasks will be kept in empty cpusets when hotplug happens and
316	* take masks of ancestors with non-empty cpus/mems, instead of being
317	* moved to an ancestor.
318	*
319	* - cpuset: a task can be moved into an empty cpuset, and again it takes
320	* masks of ancestors.
321	*
322	* - memcg: use_hierarchy is on by default and the cgroup file for the flag
323	* is not created.
324	*
325	* - blkcg: blk-throttle becomes properly hierarchical.
326	*
327	* - debug: disallowed on the default hierarchy.
328	*/
329	static bool cgroup_on_dfl(const struct cgroup *cgrp)
330	{
331	return cgrp->root == &cgrp_dfl_root;
332	}
333
334	/* IDR wrappers which synchronize using cgroup_idr_lock */
335	static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
336	gfp_t gfp_mask)
337	{
338	int ret;
339
340	idr_preload(gfp_mask);
341	spin_lock_bh(&cgroup_idr_lock);
342	ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
343	spin_unlock_bh(&cgroup_idr_lock);
344	idr_preload_end();
345	return ret;
346	}
347
348	static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
349	{
350	void *ret;
351
352	spin_lock_bh(&cgroup_idr_lock);
353	ret = idr_replace(idr, ptr, id);
354	spin_unlock_bh(&cgroup_idr_lock);
355	return ret;
356	}
357
358	static void cgroup_idr_remove(struct idr *idr, int id)
359	{
360	spin_lock_bh(&cgroup_idr_lock);
361	idr_remove(idr, id);
362	spin_unlock_bh(&cgroup_idr_lock);
363	}
364
365	/* subsystems visibly enabled on a cgroup */
366	static u16 cgroup_control(struct cgroup *cgrp)
367	{
368	struct cgroup *parent = cgroup_parent(cgrp);
369	u16 root_ss_mask = cgrp->root->subsys_mask;
370
371	if (parent)
372	return parent->subtree_control;
373
374	if (cgroup_on_dfl(cgrp))
375	root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
376	cgrp_dfl_implicit_ss_mask);
377	return root_ss_mask;
378	}
379
380	/* subsystems enabled on a cgroup */
381	static u16 cgroup_ss_mask(struct cgroup *cgrp)
382	{
383	struct cgroup *parent = cgroup_parent(cgrp);
384
385	if (parent)
386	return parent->subtree_ss_mask;
387
388	return cgrp->root->subsys_mask;
389	}
390
391	/**
392	* cgroup_css - obtain a cgroup's css for the specified subsystem
393	* @cgrp: the cgroup of interest
394	* @ss: the subsystem of interest (%NULL returns @cgrp->self)
395	*
396	* Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
397	* function must be called either under cgroup_mutex or rcu_read_lock() and
398	* the caller is responsible for pinning the returned css if it wants to
399	* keep accessing it outside the said locks. This function may return
400	* %NULL if @cgrp doesn't have @subsys_id enabled.
401	*/
402	static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
403	struct cgroup_subsys *ss)
404	{
405	if (ss)
406	return rcu_dereference_check(cgrp->subsys[ss->id],
407	lockdep_is_held(&cgroup_mutex));
408	else
409	return &cgrp->self;
410	}
411
412	/**
413	* cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
414	* @cgrp: the cgroup of interest
415	* @ss: the subsystem of interest (%NULL returns @cgrp->self)
416	*
417	* Similar to cgroup_css() but returns the effective css, which is defined
418	* as the matching css of the nearest ancestor including self which has @ss
419	* enabled. If @ss is associated with the hierarchy @cgrp is on, this
420	* function is guaranteed to return non-NULL css.
421	*/
422	static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
423	struct cgroup_subsys *ss)
424	{
425	lockdep_assert_held(&cgroup_mutex);
426
427	if (!ss)
428	return &cgrp->self;
429
430	/*
431	* This function is used while updating css associations and thus
432	* can't test the csses directly. Test ss_mask.
433	*/
434	while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
435	cgrp = cgroup_parent(cgrp);
436	if (!cgrp)
437	return NULL;
438	}
439
440	return cgroup_css(cgrp, ss);
441	}
442
443	/**
444	* cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
445	* @cgrp: the cgroup of interest
446	* @ss: the subsystem of interest
447	*
448	* Find and get the effective css of @cgrp for @ss. The effective css is
449	* defined as the matching css of the nearest ancestor including self which
450	* has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
451	* the root css is returned, so this function always returns a valid css.
452	* The returned css must be put using css_put().
453	*/
454	struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
455	struct cgroup_subsys *ss)
456	{
457	struct cgroup_subsys_state *css;
458
459	rcu_read_lock();
460
461	do {
462	css = cgroup_css(cgrp, ss);
463
464	if (css && css_tryget_online(css))
465	goto out_unlock;
466	cgrp = cgroup_parent(cgrp);
467	} while (cgrp);
468
469	css = init_css_set.subsys[ss->id];
470	css_get(css);
471	out_unlock:
472	rcu_read_unlock();
473	return css;
474	}
475
476	/* convenient tests for these bits */
477	static inline bool cgroup_is_dead(const struct cgroup *cgrp)
478	{
479	return !(cgrp->self.flags & CSS_ONLINE);
480	}
481
482	struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
483	{
484	struct cgroup *cgrp = of->kn->parent->priv;
485	struct cftype *cft = of_cft(of);
486
487	/*
488	* This is open and unprotected implementation of cgroup_css().
489	* seq_css() is only called from a kernfs file operation which has
490	* an active reference on the file. Because all the subsystem
491	* files are drained before a css is disassociated with a cgroup,
492	* the matching css from the cgroup's subsys table is guaranteed to
493	* be and stay valid until the enclosing operation is complete.
494	*/
495	if (cft->ss)
496	return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
497	else
498	return &cgrp->self;
499	}
500	EXPORT_SYMBOL_GPL(of_css);
501
502	static int notify_on_release(const struct cgroup *cgrp)
503	{
504	return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
505	}
506
507	/**
508	* for_each_css - iterate all css's of a cgroup
509	* @css: the iteration cursor
510	* @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
511	* @cgrp: the target cgroup to iterate css's of
512	*
513	* Should be called under cgroup_[tree_]mutex.
514	*/
515	#define for_each_css(css, ssid, cgrp) \
516	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
517	if (!((css) = rcu_dereference_check( \
518	(cgrp)->subsys[(ssid)], \
519	lockdep_is_held(&cgroup_mutex)))) { } \
520	else
521
522	/**
523	* for_each_e_css - iterate all effective css's of a cgroup
524	* @css: the iteration cursor
525	* @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
526	* @cgrp: the target cgroup to iterate css's of
527	*
528	* Should be called under cgroup_[tree_]mutex.
529	*/
530	#define for_each_e_css(css, ssid, cgrp) \
531	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
532	if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
533	; \
534	else
535
536	/**
537	* for_each_subsys - iterate all enabled cgroup subsystems
538	* @ss: the iteration cursor
539	* @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
540	*/
541	#define for_each_subsys(ss, ssid) \
542	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
543	(((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
544
545	/**
546	* do_each_subsys_mask - filter for_each_subsys with a bitmask
547	* @ss: the iteration cursor
548	* @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
549	* @ss_mask: the bitmask
550	*
551	* The block will only run for cases where the ssid-th bit (1 << ssid) of
552	* @ss_mask is set.
553	*/
554	#define do_each_subsys_mask(ss, ssid, ss_mask) do { \
555	unsigned long __ss_mask = (ss_mask); \
556	if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
557	(ssid) = 0; \
558	break; \
559	} \
560	for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
561	(ss) = cgroup_subsys[ssid]; \
562	{
563
564	#define while_each_subsys_mask() \
565	} \
566	} \
567	} while (false)
568
569	/* iterate across the hierarchies */
570	#define for_each_root(root) \
571	list_for_each_entry((root), &cgroup_roots, root_list)
572
573	/* iterate over child cgrps, lock should be held throughout iteration */
574	#define cgroup_for_each_live_child(child, cgrp) \
575	list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
576	if (({ lockdep_assert_held(&cgroup_mutex); \
577	cgroup_is_dead(child); })) \
578	; \
579	else
580
581	/* walk live descendants in preorder */
582	#define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
583	css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
584	if (({ lockdep_assert_held(&cgroup_mutex); \
585	(dsct) = (d_css)->cgroup; \
586	cgroup_is_dead(dsct); })) \
587	; \
588	else
589
590	/* walk live descendants in postorder */
591	#define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
592	css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
593	if (({ lockdep_assert_held(&cgroup_mutex); \
594	(dsct) = (d_css)->cgroup; \
595	cgroup_is_dead(dsct); })) \
596	; \
597	else
598
599	static void cgroup_release_agent(struct work_struct *work);
600	static void check_for_release(struct cgroup *cgrp);
601
602	/*
603	* A cgroup can be associated with multiple css_sets as different tasks may
604	* belong to different cgroups on different hierarchies. In the other
605	* direction, a css_set is naturally associated with multiple cgroups.
606	* This M:N relationship is represented by the following link structure
607	* which exists for each association and allows traversing the associations
608	* from both sides.
609	*/
610	struct cgrp_cset_link {
611	/* the cgroup and css_set this link associates */
612	struct cgroup *cgrp;
613	struct css_set *cset;
614
615	/* list of cgrp_cset_links anchored at cgrp->cset_links */
616	struct list_head cset_link;
617
618	/* list of cgrp_cset_links anchored at css_set->cgrp_links */
619	struct list_head cgrp_link;
620	};
621
622	/*
623	* The default css_set - used by init and its children prior to any
624	* hierarchies being mounted. It contains a pointer to the root state
625	* for each subsystem. Also used to anchor the list of css_sets. Not
626	* reference-counted, to improve performance when child cgroups
627	* haven't been created.
628	*/
629	struct css_set init_css_set = {
630	.refcount = ATOMIC_INIT(1),
631	.cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
632	.tasks = LIST_HEAD_INIT(init_css_set.tasks),
633	.mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
634	.mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
635	.mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
636	.task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
637	};
638
639	static int css_set_count = 1; /* 1 for init_css_set */
640
641	/**
642	* css_set_populated - does a css_set contain any tasks?
643	* @cset: target css_set
644	*/
645	static bool css_set_populated(struct css_set *cset)
646	{
647	lockdep_assert_held(&css_set_lock);
648
649	return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
650	}
651
652	/**
653	* cgroup_update_populated - updated populated count of a cgroup
654	* @cgrp: the target cgroup
655	* @populated: inc or dec populated count
656	*
657	* One of the css_sets associated with @cgrp is either getting its first
658	* task or losing the last. Update @cgrp->populated_cnt accordingly. The
659	* count is propagated towards root so that a given cgroup's populated_cnt
660	* is zero iff the cgroup and all its descendants don't contain any tasks.
661	*
662	* @cgrp's interface file "cgroup.populated" is zero if
663	* @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
664	* changes from or to zero, userland is notified that the content of the
665	* interface file has changed. This can be used to detect when @cgrp and
666	* its descendants become populated or empty.
667	*/
668	static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
669	{
670	lockdep_assert_held(&css_set_lock);
671
672	do {
673	bool trigger;
674
675	if (populated)
676	trigger = !cgrp->populated_cnt++;
677	else
678	trigger = !--cgrp->populated_cnt;
679
680	if (!trigger)
681	break;
682
683	check_for_release(cgrp);
684	cgroup_file_notify(&cgrp->events_file);
685
686	cgrp = cgroup_parent(cgrp);
687	} while (cgrp);
688	}
689
690	/**
691	* css_set_update_populated - update populated state of a css_set
692	* @cset: target css_set
693	* @populated: whether @cset is populated or depopulated
694	*
695	* @cset is either getting the first task or losing the last. Update the
696	* ->populated_cnt of all associated cgroups accordingly.
697	*/
698	static void css_set_update_populated(struct css_set *cset, bool populated)
699	{
700	struct cgrp_cset_link *link;
701
702	lockdep_assert_held(&css_set_lock);
703
704	list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
705	cgroup_update_populated(link->cgrp, populated);
706	}
707
708	/**
709	* css_set_move_task - move a task from one css_set to another
710	* @task: task being moved
711	* @from_cset: css_set @task currently belongs to (may be NULL)
712	* @to_cset: new css_set @task is being moved to (may be NULL)
713	* @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
714	*
715	* Move @task from @from_cset to @to_cset. If @task didn't belong to any
716	* css_set, @from_cset can be NULL. If @task is being disassociated
717	* instead of moved, @to_cset can be NULL.
718	*
719	* This function automatically handles populated_cnt updates and
720	* css_task_iter adjustments but the caller is responsible for managing
721	* @from_cset and @to_cset's reference counts.
722	*/
723	static void css_set_move_task(struct task_struct *task,
724	struct css_set *from_cset, struct css_set *to_cset,
725	bool use_mg_tasks)
726	{
727	lockdep_assert_held(&css_set_lock);
728
729	if (to_cset && !css_set_populated(to_cset))
730	css_set_update_populated(to_cset, true);
731
732	if (from_cset) {
733	struct css_task_iter *it, *pos;
734
735	WARN_ON_ONCE(list_empty(&task->cg_list));
736
737	/*
738	* @task is leaving, advance task iterators which are
739	* pointing to it so that they can resume at the next
740	* position. Advancing an iterator might remove it from
741	* the list, use safe walk. See css_task_iter_advance*()
742	* for details.
743	*/
744	list_for_each_entry_safe(it, pos, &from_cset->task_iters,
745	iters_node)
746	if (it->task_pos == &task->cg_list)
747	css_task_iter_advance(it);
748
749	list_del_init(&task->cg_list);
750	if (!css_set_populated(from_cset))
751	css_set_update_populated(from_cset, false);
752	} else {
753	WARN_ON_ONCE(!list_empty(&task->cg_list));
754	}
755
756	if (to_cset) {
757	/*
758	* We are synchronized through cgroup_threadgroup_rwsem
759	* against PF_EXITING setting such that we can't race
760	* against cgroup_exit() changing the css_set to
761	* init_css_set and dropping the old one.
762	*/
763	WARN_ON_ONCE(task->flags & PF_EXITING);
764
765	cgroup_move_task(task, to_cset);
766	list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
767	&to_cset->tasks);
768	}
769	}
770
771	/*
772	* hash table for cgroup groups. This improves the performance to find
773	* an existing css_set. This hash doesn't (currently) take into
774	* account cgroups in empty hierarchies.
775	*/
776	#define CSS_SET_HASH_BITS 7
777	static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
778
779	static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
780	{
781	unsigned long key = 0UL;
782	struct cgroup_subsys *ss;
783	int i;
784
785	for_each_subsys(ss, i)
786	key += (unsigned long)css[i];
787	key = (key >> 16) ^ key;
788
789	return key;
790	}
791
792	static void put_css_set_locked(struct css_set *cset)
793	{
794	struct cgrp_cset_link *link, *tmp_link;
795	struct cgroup_subsys *ss;
796	int ssid;
797
798	lockdep_assert_held(&css_set_lock);
799
800	if (!atomic_dec_and_test(&cset->refcount))
801	return;
802
803	/* This css_set is dead. unlink it and release cgroup and css refs */
804	for_each_subsys(ss, ssid) {
805	list_del(&cset->e_cset_node[ssid]);
806	css_put(cset->subsys[ssid]);
807	}
808	hash_del(&cset->hlist);
809	css_set_count--;
810
811	list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
812	list_del(&link->cset_link);
813	list_del(&link->cgrp_link);
814	if (cgroup_parent(link->cgrp))
815	cgroup_put(link->cgrp);
816	kfree(link);
817	}
818
819	kfree_rcu(cset, rcu_head);
820	}
821
822	static void put_css_set(struct css_set *cset)
823	{
824	unsigned long flags;
825
826	/*
827	* Ensure that the refcount doesn't hit zero while any readers
828	* can see it. Similar to atomic_dec_and_lock(), but for an
829	* rwlock
830	*/
831	if (atomic_add_unless(&cset->refcount, -1, 1))
832	return;
833
834	spin_lock_irqsave(&css_set_lock, flags);
835	put_css_set_locked(cset);
836	spin_unlock_irqrestore(&css_set_lock, flags);
837	}
838
839	/*
840	* refcounted get/put for css_set objects
841	*/
842	static inline void get_css_set(struct css_set *cset)
843	{
844	atomic_inc(&cset->refcount);
845	}
846
847	/**
848	* compare_css_sets - helper function for find_existing_css_set().
849	* @cset: candidate css_set being tested
850	* @old_cset: existing css_set for a task
851	* @new_cgrp: cgroup that's being entered by the task
852	* @template: desired set of css pointers in css_set (pre-calculated)
853	*
854	* Returns true if "cset" matches "old_cset" except for the hierarchy
855	* which "new_cgrp" belongs to, for which it should match "new_cgrp".
856	*/
857	static bool compare_css_sets(struct css_set *cset,
858	struct css_set *old_cset,
859	struct cgroup *new_cgrp,
860	struct cgroup_subsys_state *template[])
861	{
862	struct list_head *l1, *l2;
863
864	/*
865	* On the default hierarchy, there can be csets which are
866	* associated with the same set of cgroups but different csses.
867	* Let's first ensure that csses match.
868	*/
869	if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
870	return false;
871
872	/*
873	* Compare cgroup pointers in order to distinguish between
874	* different cgroups in hierarchies. As different cgroups may
875	* share the same effective css, this comparison is always
876	* necessary.
877	*/
878	l1 = &cset->cgrp_links;
879	l2 = &old_cset->cgrp_links;
880	while (1) {
881	struct cgrp_cset_link *link1, *link2;
882	struct cgroup *cgrp1, *cgrp2;
883
884	l1 = l1->next;
885	l2 = l2->next;
886	/* See if we reached the end - both lists are equal length. */
887	if (l1 == &cset->cgrp_links) {
888	BUG_ON(l2 != &old_cset->cgrp_links);
889	break;
890	} else {
891	BUG_ON(l2 == &old_cset->cgrp_links);
892	}
893	/* Locate the cgroups associated with these links. */
894	link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
895	link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
896	cgrp1 = link1->cgrp;
897	cgrp2 = link2->cgrp;
898	/* Hierarchies should be linked in the same order. */
899	BUG_ON(cgrp1->root != cgrp2->root);
900
901	/*
902	* If this hierarchy is the hierarchy of the cgroup
903	* that's changing, then we need to check that this
904	* css_set points to the new cgroup; if it's any other
905	* hierarchy, then this css_set should point to the
906	* same cgroup as the old css_set.
907	*/
908	if (cgrp1->root == new_cgrp->root) {
909	if (cgrp1 != new_cgrp)
910	return false;
911	} else {
912	if (cgrp1 != cgrp2)
913	return false;
914	}
915	}
916	return true;
917	}
918
919	/**
920	* find_existing_css_set - init css array and find the matching css_set
921	* @old_cset: the css_set that we're using before the cgroup transition
922	* @cgrp: the cgroup that we're moving into
923	* @template: out param for the new set of csses, should be clear on entry
924	*/
925	static struct css_set *find_existing_css_set(struct css_set *old_cset,
926	struct cgroup *cgrp,
927	struct cgroup_subsys_state *template[])
928	{
929	struct cgroup_root *root = cgrp->root;
930	struct cgroup_subsys *ss;
931	struct css_set *cset;
932	unsigned long key;
933	int i;
934
935	/*
936	* Build the set of subsystem state objects that we want to see in the
937	* new css_set. while subsystems can change globally, the entries here
938	* won't change, so no need for locking.
939	*/
940	for_each_subsys(ss, i) {
941	if (root->subsys_mask & (1UL << i)) {
942	/*
943	* @ss is in this hierarchy, so we want the
944	* effective css from @cgrp.
945	*/
946	template[i] = cgroup_e_css(cgrp, ss);
947	} else {
948	/*
949	* @ss is not in this hierarchy, so we don't want
950	* to change the css.
951	*/
952	template[i] = old_cset->subsys[i];
953	}
954	}
955
956	key = css_set_hash(template);
957	hash_for_each_possible(css_set_table, cset, hlist, key) {
958	if (!compare_css_sets(cset, old_cset, cgrp, template))
959	continue;
960
961	/* This css_set matches what we need */
962	return cset;
963	}
964
965	/* No existing cgroup group matched */
966	return NULL;
967	}
968
969	static void free_cgrp_cset_links(struct list_head *links_to_free)
970	{
971	struct cgrp_cset_link *link, *tmp_link;
972
973	list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
974	list_del(&link->cset_link);
975	kfree(link);
976	}
977	}
978
979	/**
980	* allocate_cgrp_cset_links - allocate cgrp_cset_links
981	* @count: the number of links to allocate
982	* @tmp_links: list_head the allocated links are put on
983	*
984	* Allocate @count cgrp_cset_link structures and chain them on @tmp_links
985	* through ->cset_link. Returns 0 on success or -errno.
986	*/
987	static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
988	{
989	struct cgrp_cset_link *link;
990	int i;
991
992	INIT_LIST_HEAD(tmp_links);
993
994	for (i = 0; i < count; i++) {
995	link = kzalloc(sizeof(*link), GFP_KERNEL);
996	if (!link) {
997	free_cgrp_cset_links(tmp_links);
998	return -ENOMEM;
999	}
1000	list_add(&link->cset_link, tmp_links);
1001	}
1002	return 0;
1003	}
1004
1005	/**
1006	* link_css_set - a helper function to link a css_set to a cgroup
1007	* @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1008	* @cset: the css_set to be linked
1009	* @cgrp: the destination cgroup
1010	*/
1011	static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1012	struct cgroup *cgrp)
1013	{
1014	struct cgrp_cset_link *link;
1015
1016	BUG_ON(list_empty(tmp_links));
1017
1018	if (cgroup_on_dfl(cgrp))
1019	cset->dfl_cgrp = cgrp;
1020
1021	link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1022	link->cset = cset;
1023	link->cgrp = cgrp;
1024
1025	/*
1026	* Always add links to the tail of the lists so that the lists are
1027	* in choronological order.
1028	*/
1029	list_move_tail(&link->cset_link, &cgrp->cset_links);
1030	list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1031
1032	if (cgroup_parent(cgrp))
1033	cgroup_get(cgrp);
1034	}
1035
1036	/**
1037	* find_css_set - return a new css_set with one cgroup updated
1038	* @old_cset: the baseline css_set
1039	* @cgrp: the cgroup to be updated
1040	*
1041	* Return a new css_set that's equivalent to @old_cset, but with @cgrp
1042	* substituted into the appropriate hierarchy.
1043	*/
1044	static struct css_set *find_css_set(struct css_set *old_cset,
1045	struct cgroup *cgrp)
1046	{
1047	struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1048	struct css_set *cset;
1049	struct list_head tmp_links;
1050	struct cgrp_cset_link *link;
1051	struct cgroup_subsys *ss;
1052	unsigned long key;
1053	int ssid;
1054
1055	lockdep_assert_held(&cgroup_mutex);
1056
1057	/* First see if we already have a cgroup group that matches
1058	* the desired set */
1059	spin_lock_irq(&css_set_lock);
1060	cset = find_existing_css_set(old_cset, cgrp, template);
1061	if (cset)
1062	get_css_set(cset);
1063	spin_unlock_irq(&css_set_lock);
1064
1065	if (cset)
1066	return cset;
1067
1068	cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1069	if (!cset)
1070	return NULL;
1071
1072	/* Allocate all the cgrp_cset_link objects that we'll need */
1073	if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1074	kfree(cset);
1075	return NULL;
1076	}
1077
1078	atomic_set(&cset->refcount, 1);
1079	INIT_LIST_HEAD(&cset->cgrp_links);
1080	INIT_LIST_HEAD(&cset->tasks);
1081	INIT_LIST_HEAD(&cset->mg_tasks);
1082	INIT_LIST_HEAD(&cset->mg_preload_node);
1083	INIT_LIST_HEAD(&cset->mg_node);
1084	INIT_LIST_HEAD(&cset->task_iters);
1085	INIT_HLIST_NODE(&cset->hlist);
1086
1087	/* Copy the set of subsystem state objects generated in
1088	* find_existing_css_set() */
1089	memcpy(cset->subsys, template, sizeof(cset->subsys));
1090
1091	spin_lock_irq(&css_set_lock);
1092	/* Add reference counts and links from the new css_set. */
1093	list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1094	struct cgroup *c = link->cgrp;
1095
1096	if (c->root == cgrp->root)
1097	c = cgrp;
1098	link_css_set(&tmp_links, cset, c);
1099	}
1100
1101	BUG_ON(!list_empty(&tmp_links));
1102
1103	css_set_count++;
1104
1105	/* Add @cset to the hash table */
1106	key = css_set_hash(cset->subsys);
1107	hash_add(css_set_table, &cset->hlist, key);
1108
1109	for_each_subsys(ss, ssid) {
1110	struct cgroup_subsys_state *css = cset->subsys[ssid];
1111
1112	list_add_tail(&cset->e_cset_node[ssid],
1113	&css->cgroup->e_csets[ssid]);
1114	css_get(css);
1115	}
1116
1117	spin_unlock_irq(&css_set_lock);
1118
1119	return cset;
1120	}
1121
1122	static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1123	{
1124	struct cgroup *root_cgrp = kf_root->kn->priv;
1125
1126	return root_cgrp->root;
1127	}
1128
1129	static int cgroup_init_root_id(struct cgroup_root *root)
1130	{
1131	int id;
1132
1133	lockdep_assert_held(&cgroup_mutex);
1134
1135	id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1136	if (id < 0)
1137	return id;
1138
1139	root->hierarchy_id = id;
1140	return 0;
1141	}
1142
1143	static void cgroup_exit_root_id(struct cgroup_root *root)
1144	{
1145	lockdep_assert_held(&cgroup_mutex);
1146
1147	idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1148	}
1149
1150	static void cgroup_free_root(struct cgroup_root *root)
1151	{
1152	if (root) {
1153	idr_destroy(&root->cgroup_idr);
1154	kfree(root);
1155	}
1156	}
1157
1158	static void cgroup_destroy_root(struct cgroup_root *root)
1159	{
1160	struct cgroup *cgrp = &root->cgrp;
1161	struct cgrp_cset_link *link, *tmp_link;
1162
1163	trace_cgroup_destroy_root(root);
1164
1165	cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1166
1167	BUG_ON(atomic_read(&root->nr_cgrps));
1168	BUG_ON(!list_empty(&cgrp->self.children));
1169
1170	/* Rebind all subsystems back to the default hierarchy */
1171	WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1172
1173	/*
1174	* Release all the links from cset_links to this hierarchy's
1175	* root cgroup
1176	*/
1177	spin_lock_irq(&css_set_lock);
1178
1179	list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1180	list_del(&link->cset_link);
1181	list_del(&link->cgrp_link);
1182	kfree(link);
1183	}
1184
1185	spin_unlock_irq(&css_set_lock);
1186
1187	if (!list_empty(&root->root_list)) {
1188	list_del(&root->root_list);
1189	cgroup_root_count--;
1190	}
1191
1192	cgroup_exit_root_id(root);
1193
1194	mutex_unlock(&cgroup_mutex);
1195
1196	kernfs_destroy_root(root->kf_root);
1197	cgroup_free_root(root);
1198	}
1199
1200	/*
1201	* look up cgroup associated with current task's cgroup namespace on the
1202	* specified hierarchy
1203	*/
1204	static struct cgroup *
1205	current_cgns_cgroup_from_root(struct cgroup_root *root)
1206	{
1207	struct cgroup *res = NULL;
1208	struct css_set *cset;
1209
1210	lockdep_assert_held(&css_set_lock);
1211
1212	rcu_read_lock();
1213
1214	cset = current->nsproxy->cgroup_ns->root_cset;
1215	if (cset == &init_css_set) {
1216	res = &root->cgrp;
1217	} else {
1218	struct cgrp_cset_link *link;
1219
1220	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1221	struct cgroup *c = link->cgrp;
1222
1223	if (c->root == root) {
1224	res = c;
1225	break;
1226	}
1227	}
1228	}
1229	rcu_read_unlock();
1230
1231	BUG_ON(!res);
1232	return res;
1233	}
1234
1235	/* look up cgroup associated with given css_set on the specified hierarchy */
1236	static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1237	struct cgroup_root *root)
1238	{
1239	struct cgroup *res = NULL;
1240
1241	lockdep_assert_held(&cgroup_mutex);
1242	lockdep_assert_held(&css_set_lock);
1243
1244	if (cset == &init_css_set) {
1245	res = &root->cgrp;
1246	} else {
1247	struct cgrp_cset_link *link;
1248
1249	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1250	struct cgroup *c = link->cgrp;
1251
1252	if (c->root == root) {
1253	res = c;
1254	break;
1255	}
1256	}
1257	}
1258
1259	BUG_ON(!res);
1260	return res;
1261	}
1262
1263	/*
1264	* Return the cgroup for "task" from the given hierarchy. Must be
1265	* called with cgroup_mutex and css_set_lock held.
1266	*/
1267	static struct cgroup *task_cgroup_from_root(struct task_struct *task,
1268	struct cgroup_root *root)
1269	{
1270	/*
1271	* No need to lock the task - since we hold cgroup_mutex the
1272	* task can't change groups, so the only thing that can happen
1273	* is that it exits and its css is set back to init_css_set.
1274	*/
1275	return cset_cgroup_from_root(task_css_set(task), root);
1276	}
1277
1278	/*
1279	* A task must hold cgroup_mutex to modify cgroups.
1280	*
1281	* Any task can increment and decrement the count field without lock.
1282	* So in general, code holding cgroup_mutex can't rely on the count
1283	* field not changing. However, if the count goes to zero, then only
1284	* cgroup_attach_task() can increment it again. Because a count of zero
1285	* means that no tasks are currently attached, therefore there is no
1286	* way a task attached to that cgroup can fork (the other way to
1287	* increment the count). So code holding cgroup_mutex can safely
1288	* assume that if the count is zero, it will stay zero. Similarly, if
1289	* a task holds cgroup_mutex on a cgroup with zero count, it
1290	* knows that the cgroup won't be removed, as cgroup_rmdir()
1291	* needs that mutex.
1292	*
1293	* A cgroup can only be deleted if both its 'count' of using tasks
1294	* is zero, and its list of 'children' cgroups is empty. Since all
1295	* tasks in the system use _some_ cgroup, and since there is always at
1296	* least one task in the system (init, pid == 1), therefore, root cgroup
1297	* always has either children cgroups and/or using tasks. So we don't
1298	* need a special hack to ensure that root cgroup cannot be deleted.
1299	*
1300	* P.S. One more locking exception. RCU is used to guard the
1301	* update of a tasks cgroup pointer by cgroup_attach_task()
1302	*/
1303
1304	static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1305	static const struct file_operations proc_cgroupstats_operations;
1306
1307	static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1308	char *buf)
1309	{
1310	struct cgroup_subsys *ss = cft->ss;
1311
1312	if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1313	!(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1314	snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1315	cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1316	cft->name);
1317	else
1318	strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1319	return buf;
1320	}
1321
1322	/**
1323	* cgroup_file_mode - deduce file mode of a control file
1324	* @cft: the control file in question
1325	*
1326	* S_IRUGO for read, S_IWUSR for write.
1327	*/
1328	static umode_t cgroup_file_mode(const struct cftype *cft)
1329	{
1330	umode_t mode = 0;
1331
1332	if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1333	mode |= S_IRUGO;
1334
1335	if (cft->write_u64 || cft->write_s64 || cft->write) {
1336	if (cft->flags & CFTYPE_WORLD_WRITABLE)
1337	mode |= S_IWUGO;
1338	else
1339	mode |= S_IWUSR;
1340	}
1341
1342	return mode;
1343	}
1344
1345	/**
1346	* cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1347	* @subtree_control: the new subtree_control mask to consider
1348	* @this_ss_mask: available subsystems
1349	*
1350	* On the default hierarchy, a subsystem may request other subsystems to be
1351	* enabled together through its ->depends_on mask. In such cases, more
1352	* subsystems than specified in "cgroup.subtree_control" may be enabled.
1353	*
1354	* This function calculates which subsystems need to be enabled if
1355	* @subtree_control is to be applied while restricted to @this_ss_mask.
1356	*/
1357	static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1358	{
1359	u16 cur_ss_mask = subtree_control;
1360	struct cgroup_subsys *ss;
1361	int ssid;
1362
1363	lockdep_assert_held(&cgroup_mutex);
1364
1365	cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1366
1367	while (true) {
1368	u16 new_ss_mask = cur_ss_mask;
1369
1370	do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1371	new_ss_mask |= ss->depends_on;
1372	} while_each_subsys_mask();
1373
1374	/*
1375	* Mask out subsystems which aren't available. This can
1376	* happen only if some depended-upon subsystems were bound
1377	* to non-default hierarchies.
1378	*/
1379	new_ss_mask &= this_ss_mask;
1380
1381	if (new_ss_mask == cur_ss_mask)
1382	break;
1383	cur_ss_mask = new_ss_mask;
1384	}
1385
1386	return cur_ss_mask;
1387	}
1388
1389	/**
1390	* cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1391	* @kn: the kernfs_node being serviced
1392	*
1393	* This helper undoes cgroup_kn_lock_live() and should be invoked before
1394	* the method finishes if locking succeeded. Note that once this function
1395	* returns the cgroup returned by cgroup_kn_lock_live() may become
1396	* inaccessible any time. If the caller intends to continue to access the
1397	* cgroup, it should pin it before invoking this function.
1398	*/
1399	static void cgroup_kn_unlock(struct kernfs_node *kn)
1400	{
1401	struct cgroup *cgrp;
1402
1403	if (kernfs_type(kn) == KERNFS_DIR)
1404	cgrp = kn->priv;
1405	else
1406	cgrp = kn->parent->priv;
1407
1408	mutex_unlock(&cgroup_mutex);
1409
1410	kernfs_unbreak_active_protection(kn);
1411	cgroup_put(cgrp);
1412	}
1413
1414	/**
1415	* cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1416	* @kn: the kernfs_node being serviced
1417	* @drain_offline: perform offline draining on the cgroup
1418	*
1419	* This helper is to be used by a cgroup kernfs method currently servicing
1420	* @kn. It breaks the active protection, performs cgroup locking and
1421	* verifies that the associated cgroup is alive. Returns the cgroup if
1422	* alive; otherwise, %NULL. A successful return should be undone by a
1423	* matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1424	* cgroup is drained of offlining csses before return.
1425	*
1426	* Any cgroup kernfs method implementation which requires locking the
1427	* associated cgroup should use this helper. It avoids nesting cgroup
1428	* locking under kernfs active protection and allows all kernfs operations
1429	* including self-removal.
1430	*/
1431	static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn,
1432	bool drain_offline)
1433	{
1434	struct cgroup *cgrp;
1435
1436	if (kernfs_type(kn) == KERNFS_DIR)
1437	cgrp = kn->priv;
1438	else
1439	cgrp = kn->parent->priv;
1440
1441	/*
1442	* We're gonna grab cgroup_mutex which nests outside kernfs
1443	* active_ref. cgroup liveliness check alone provides enough
1444	* protection against removal. Ensure @cgrp stays accessible and
1445	* break the active_ref protection.
1446	*/
1447	if (!cgroup_tryget(cgrp))
1448	return NULL;
1449	kernfs_break_active_protection(kn);
1450
1451	if (drain_offline)
1452	cgroup_lock_and_drain_offline(cgrp);
1453	else
1454	mutex_lock(&cgroup_mutex);
1455
1456	if (!cgroup_is_dead(cgrp))
1457	return cgrp;
1458
1459	cgroup_kn_unlock(kn);
1460	return NULL;
1461	}
1462
1463	static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1464	{
1465	char name[CGROUP_FILE_NAME_MAX];
1466
1467	lockdep_assert_held(&cgroup_mutex);
1468
1469	if (cft->file_offset) {
1470	struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1471	struct cgroup_file *cfile = (void *)css + cft->file_offset;
1472
1473	spin_lock_irq(&cgroup_file_kn_lock);
1474	cfile->kn = NULL;
1475	spin_unlock_irq(&cgroup_file_kn_lock);
1476	}
1477
1478	kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1479	}
1480
1481	/**
1482	* css_clear_dir - remove subsys files in a cgroup directory
1483	* @css: taget css
1484	*/
1485	static void css_clear_dir(struct cgroup_subsys_state *css)
1486	{
1487	struct cgroup *cgrp = css->cgroup;
1488	struct cftype *cfts;
1489
1490	if (!(css->flags & CSS_VISIBLE))
1491	return;
1492
1493	css->flags &= ~CSS_VISIBLE;
1494
1495	list_for_each_entry(cfts, &css->ss->cfts, node)
1496	cgroup_addrm_files(css, cgrp, cfts, false);
1497	}
1498
1499	/**
1500	* css_populate_dir - create subsys files in a cgroup directory
1501	* @css: target css
1502	*
1503	* On failure, no file is added.
1504	*/
1505	static int css_populate_dir(struct cgroup_subsys_state *css)
1506	{
1507	struct cgroup *cgrp = css->cgroup;
1508	struct cftype *cfts, *failed_cfts;
1509	int ret;
1510
1511	if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1512	return 0;
1513
1514	if (!css->ss) {
1515	if (cgroup_on_dfl(cgrp))
1516	cfts = cgroup_dfl_base_files;
1517	else
1518	cfts = cgroup_legacy_base_files;
1519
1520	return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1521	}
1522
1523	list_for_each_entry(cfts, &css->ss->cfts, node) {
1524	ret = cgroup_addrm_files(css, cgrp, cfts, true);
1525	if (ret < 0) {
1526	failed_cfts = cfts;
1527	goto err;
1528	}
1529	}
1530
1531	css->flags |= CSS_VISIBLE;
1532
1533	return 0;
1534	err:
1535	list_for_each_entry(cfts, &css->ss->cfts, node) {
1536	if (cfts == failed_cfts)
1537	break;
1538	cgroup_addrm_files(css, cgrp, cfts, false);
1539	}
1540	return ret;
1541	}
1542
1543	static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1544	{
1545	struct cgroup *dcgrp = &dst_root->cgrp;
1546	struct cgroup_subsys *ss;
1547	int ssid, i, ret;
1548
1549	lockdep_assert_held(&cgroup_mutex);
1550
1551	do_each_subsys_mask(ss, ssid, ss_mask) {
1552	/*
1553	* If @ss has non-root csses attached to it, can't move.
1554	* If @ss is an implicit controller, it is exempt from this
1555	* rule and can be stolen.
1556	*/
1557	if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1558	!ss->implicit_on_dfl)
1559	return -EBUSY;
1560
1561	/* can't move between two non-dummy roots either */
1562	if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1563	return -EBUSY;
1564	} while_each_subsys_mask();
1565
1566	do_each_subsys_mask(ss, ssid, ss_mask) {
1567	struct cgroup_root *src_root = ss->root;
1568	struct cgroup *scgrp = &src_root->cgrp;
1569	struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1570	struct css_set *cset;
1571
1572	WARN_ON(!css || cgroup_css(dcgrp, ss));
1573
1574	/* disable from the source */
1575	src_root->subsys_mask &= ~(1 << ssid);
1576	WARN_ON(cgroup_apply_control(scgrp));
1577	cgroup_finalize_control(scgrp, 0);
1578
1579	/* rebind */
1580	RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1581	rcu_assign_pointer(dcgrp->subsys[ssid], css);
1582	ss->root = dst_root;
1583	css->cgroup = dcgrp;
1584
1585	spin_lock_irq(&css_set_lock);
1586	hash_for_each(css_set_table, i, cset, hlist)
1587	list_move_tail(&cset->e_cset_node[ss->id],
1588	&dcgrp->e_csets[ss->id]);
1589	spin_unlock_irq(&css_set_lock);
1590
1591	/* default hierarchy doesn't enable controllers by default */
1592	dst_root->subsys_mask |= 1 << ssid;
1593	if (dst_root == &cgrp_dfl_root) {
1594	static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1595	} else {
1596	dcgrp->subtree_control |= 1 << ssid;
1597	static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1598	}
1599
1600	ret = cgroup_apply_control(dcgrp);
1601	if (ret)
1602	pr_warn("partial failure to rebind %s controller (err=%d)\n",
1603	ss->name, ret);
1604
1605	if (ss->bind)
1606	ss->bind(css);
1607	} while_each_subsys_mask();
1608
1609	kernfs_activate(dcgrp->kn);
1610	return 0;
1611	}
1612
1613	static int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1614	struct kernfs_root *kf_root)
1615	{
1616	int len = 0;
1617	char *buf = NULL;
1618	struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1619	struct cgroup *ns_cgroup;
1620
1621	buf = kmalloc(PATH_MAX, GFP_KERNEL);
1622	if (!buf)
1623	return -ENOMEM;
1624
1625	spin_lock_irq(&css_set_lock);
1626	ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1627	len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1628	spin_unlock_irq(&css_set_lock);
1629
1630	if (len >= PATH_MAX)
1631	len = -ERANGE;
1632	else if (len > 0) {
1633	seq_escape(sf, buf, " \t\n\\");
1634	len = 0;
1635	}
1636	kfree(buf);
1637	return len;
1638	}
1639
1640	static int cgroup_show_options(struct seq_file *seq,
1641	struct kernfs_root *kf_root)
1642	{
1643	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1644	struct cgroup_subsys *ss;
1645	int ssid;
1646
1647	if (root != &cgrp_dfl_root)
1648	for_each_subsys(ss, ssid)
1649	if (root->subsys_mask & (1 << ssid))
1650	seq_show_option(seq, ss->legacy_name, NULL);
1651	if (root->flags & CGRP_ROOT_NOPREFIX)
1652	seq_puts(seq, ",noprefix");
1653	if (root->flags & CGRP_ROOT_XATTR)
1654	seq_puts(seq, ",xattr");
1655
1656	spin_lock(&release_agent_path_lock);
1657	if (strlen(root->release_agent_path))
1658	seq_show_option(seq, "release_agent",
1659	root->release_agent_path);
1660	spin_unlock(&release_agent_path_lock);
1661
1662	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
1663	seq_puts(seq, ",clone_children");
1664	if (strlen(root->name))
1665	seq_show_option(seq, "name", root->name);
1666	return 0;
1667	}
1668
1669	struct cgroup_sb_opts {
1670	u16 subsys_mask;
1671	unsigned int flags;
1672	char *release_agent;
1673	bool cpuset_clone_children;
1674	char *name;
1675	/* User explicitly requested empty subsystem */
1676	bool none;
1677	};
1678
1679	static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1680	{
1681	char *token, *o = data;
1682	bool all_ss = false, one_ss = false;
1683	u16 mask = U16_MAX;
1684	struct cgroup_subsys *ss;
1685	int nr_opts = 0;
1686	int i;
1687
1688	#ifdef CONFIG_CPUSETS
1689	mask = ~((u16)1 << cpuset_cgrp_id);
1690	#endif
1691
1692	memset(opts, 0, sizeof(*opts));
1693
1694	while ((token = strsep(&o, ",")) != NULL) {
1695	nr_opts++;
1696
1697	if (!*token)
1698	return -EINVAL;
1699	if (!strcmp(token, "none")) {
1700	/* Explicitly have no subsystems */
1701	opts->none = true;
1702	continue;
1703	}
1704	if (!strcmp(token, "all")) {
1705	/* Mutually exclusive option 'all' + subsystem name */
1706	if (one_ss)
1707	return -EINVAL;
1708	all_ss = true;
1709	continue;
1710	}
1711	if (!strcmp(token, "noprefix")) {
1712	opts->flags |= CGRP_ROOT_NOPREFIX;
1713	continue;
1714	}
1715	if (!strcmp(token, "clone_children")) {
1716	opts->cpuset_clone_children = true;
1717	continue;
1718	}
1719	if (!strcmp(token, "xattr")) {
1720	opts->flags |= CGRP_ROOT_XATTR;
1721	continue;
1722	}
1723	if (!strncmp(token, "release_agent=", 14)) {
1724	/* Specifying two release agents is forbidden */
1725	if (opts->release_agent)
1726	return -EINVAL;
1727	opts->release_agent =
1728	kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1729	if (!opts->release_agent)
1730	return -ENOMEM;
1731	continue;
1732	}
1733	if (!strncmp(token, "name=", 5)) {
1734	const char *name = token + 5;
1735	/* Can't specify an empty name */
1736	if (!strlen(name))
1737	return -EINVAL;
1738	/* Must match [\w.-]+ */
1739	for (i = 0; i < strlen(name); i++) {
1740	char c = name[i];
1741	if (isalnum(c))
1742	continue;
1743	if ((c == '.') || (c == '-') || (c == '_'))
1744	continue;
1745	return -EINVAL;
1746	}
1747	/* Specifying two names is forbidden */
1748	if (opts->name)
1749	return -EINVAL;
1750	opts->name = kstrndup(name,
1751	MAX_CGROUP_ROOT_NAMELEN - 1,
1752	GFP_KERNEL);
1753	if (!opts->name)
1754	return -ENOMEM;
1755
1756	continue;
1757	}
1758
1759	for_each_subsys(ss, i) {
1760	if (strcmp(token, ss->legacy_name))
1761	continue;
1762	if (!cgroup_ssid_enabled(i))
1763	continue;
1764	if (cgroup_ssid_no_v1(i))
1765	continue;
1766
1767	/* Mutually exclusive option 'all' + subsystem name */
1768	if (all_ss)
1769	return -EINVAL;
1770	opts->subsys_mask |= (1 << i);
1771	one_ss = true;
1772
1773	break;
1774	}
1775	if (i == CGROUP_SUBSYS_COUNT)
1776	return -ENOENT;
1777	}
1778
1779	/*
1780	* If the 'all' option was specified select all the subsystems,
1781	* otherwise if 'none', 'name=' and a subsystem name options were
1782	* not specified, let's default to 'all'
1783	*/
1784	if (all_ss || (!one_ss && !opts->none && !opts->name))
1785	for_each_subsys(ss, i)
1786	if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i))
1787	opts->subsys_mask |= (1 << i);
1788
1789	/*
1790	* We either have to specify by name or by subsystems. (So all
1791	* empty hierarchies must have a name).
1792	*/
1793	if (!opts->subsys_mask && !opts->name)
1794	return -EINVAL;
1795
1796	/*
1797	* Option noprefix was introduced just for backward compatibility
1798	* with the old cpuset, so we allow noprefix only if mounting just
1799	* the cpuset subsystem.
1800	*/
1801	if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1802	return -EINVAL;
1803
1804	/* Can't specify "none" and some subsystems */
1805	if (opts->subsys_mask && opts->none)
1806	return -EINVAL;
1807
1808	return 0;
1809	}
1810
1811	static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1812	{
1813	int ret = 0;
1814	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1815	struct cgroup_sb_opts opts;
1816	u16 added_mask, removed_mask;
1817
1818	if (root == &cgrp_dfl_root) {
1819	pr_err("remount is not allowed\n");
1820	return -EINVAL;
1821	}
1822
1823	cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1824
1825	/* See what subsystems are wanted */
1826	ret = parse_cgroupfs_options(data, &opts);
1827	if (ret)
1828	goto out_unlock;
1829
1830	if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
1831	pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1832	task_tgid_nr(current), current->comm);
1833
1834	added_mask = opts.subsys_mask & ~root->subsys_mask;
1835	removed_mask = root->subsys_mask & ~opts.subsys_mask;
1836
1837	/* Don't allow flags or name to change at remount */
1838	if ((opts.flags ^ root->flags) ||
1839	(opts.name && strcmp(opts.name, root->name))) {
1840	pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1841	opts.flags, opts.name ?: "", root->flags, root->name);
1842	ret = -EINVAL;
1843	goto out_unlock;
1844	}
1845
1846	/* remounting is not allowed for populated hierarchies */
1847	if (!list_empty(&root->cgrp.self.children)) {
1848	ret = -EBUSY;
1849	goto out_unlock;
1850	}
1851
1852	ret = rebind_subsystems(root, added_mask);
1853	if (ret)
1854	goto out_unlock;
1855
1856	WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1857
1858	if (opts.release_agent) {
1859	spin_lock(&release_agent_path_lock);
1860	strcpy(root->release_agent_path, opts.release_agent);
1861	spin_unlock(&release_agent_path_lock);
1862	}
1863
1864	trace_cgroup_remount(root);
1865
1866	out_unlock:
1867	kfree(opts.release_agent);
1868	kfree(opts.name);
1869	mutex_unlock(&cgroup_mutex);
1870	return ret;
1871	}
1872
1873	/*
1874	* To reduce the fork() overhead for systems that are not actually using
1875	* their cgroups capability, we don't maintain the lists running through
1876	* each css_set to its tasks until we see the list actually used - in other
1877	* words after the first mount.
1878	*/
1879	static bool use_task_css_set_links __read_mostly;
1880
1881	static void cgroup_enable_task_cg_lists(void)
1882	{
1883	struct task_struct *p, *g;
1884
1885	spin_lock_irq(&css_set_lock);
1886
1887	if (use_task_css_set_links)
1888	goto out_unlock;
1889
1890	use_task_css_set_links = true;
1891
1892	/*
1893	* We need tasklist_lock because RCU is not safe against
1894	* while_each_thread(). Besides, a forking task that has passed
1895	* cgroup_post_fork() without seeing use_task_css_set_links = 1
1896	* is not guaranteed to have its child immediately visible in the
1897	* tasklist if we walk through it with RCU.
1898	*/
1899	read_lock(&tasklist_lock);
1900	do_each_thread(g, p) {
1901	WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1902	task_css_set(p) != &init_css_set);
1903
1904	/*
1905	* We should check if the process is exiting, otherwise
1906	* it will race with cgroup_exit() in that the list
1907	* entry won't be deleted though the process has exited.
1908	* Do it while holding siglock so that we don't end up
1909	* racing against cgroup_exit().
1910	*
1911	* Interrupts were already disabled while acquiring
1912	* the css_set_lock, so we do not need to disable it
1913	* again when acquiring the sighand->siglock here.
1914	*/
1915	spin_lock(&p->sighand->siglock);
1916	if (!(p->flags & PF_EXITING)) {
1917	struct css_set *cset = task_css_set(p);
1918
1919	if (!css_set_populated(cset))
1920	css_set_update_populated(cset, true);
1921	list_add_tail(&p->cg_list, &cset->tasks);
1922	get_css_set(cset);
1923	}
1924	spin_unlock(&p->sighand->siglock);
1925	} while_each_thread(g, p);
1926	read_unlock(&tasklist_lock);
1927	out_unlock:
1928	spin_unlock_irq(&css_set_lock);
1929	}
1930
1931	static void init_cgroup_housekeeping(struct cgroup *cgrp)
1932	{
1933	struct cgroup_subsys *ss;
1934	int ssid;
1935
1936	INIT_LIST_HEAD(&cgrp->self.sibling);
1937	INIT_LIST_HEAD(&cgrp->self.children);
1938	INIT_LIST_HEAD(&cgrp->cset_links);
1939	INIT_LIST_HEAD(&cgrp->pidlists);
1940	mutex_init(&cgrp->pidlist_mutex);
1941	cgrp->self.cgroup = cgrp;
1942	cgrp->self.flags |= CSS_ONLINE;
1943
1944	for_each_subsys(ss, ssid)
1945	INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1946
1947	init_waitqueue_head(&cgrp->offline_waitq);
1948	INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent);
1949	}
1950
1951	static void init_cgroup_root(struct cgroup_root *root,
1952	struct cgroup_sb_opts *opts)
1953	{
1954	struct cgroup *cgrp = &root->cgrp;
1955
1956	INIT_LIST_HEAD(&root->root_list);
1957	atomic_set(&root->nr_cgrps, 1);
1958	cgrp->root = root;
1959	init_cgroup_housekeeping(cgrp);
1960	idr_init(&root->cgroup_idr);
1961
1962	root->flags = opts->flags;
1963	if (opts->release_agent)
1964	strcpy(root->release_agent_path, opts->release_agent);
1965	if (opts->name)
1966	strcpy(root->name, opts->name);
1967	if (opts->cpuset_clone_children)
1968	set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1969	}
1970
1971	static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1972	{
1973	LIST_HEAD(tmp_links);
1974	struct cgroup *root_cgrp = &root->cgrp;
1975	struct css_set *cset;
1976	int i, ret;
1977
1978	lockdep_assert_held(&cgroup_mutex);
1979
1980	ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1981	if (ret < 0)
1982	goto out;
1983	root_cgrp->id = ret;
1984	root_cgrp->ancestor_ids[0] = ret;
1985
1986	ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1987	GFP_KERNEL);
1988	if (ret)
1989	goto out;
1990
1991	/*
1992	* We're accessing css_set_count without locking css_set_lock here,
1993	* but that's OK - it can only be increased by someone holding
1994	* cgroup_lock, and that's us. Later rebinding may disable
1995	* controllers on the default hierarchy and thus create new csets,
1996	* which can't be more than the existing ones. Allocate 2x.
1997	*/
1998	ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1999	if (ret)
2000	goto cancel_ref;
2001
2002	ret = cgroup_init_root_id(root);
2003	if (ret)
2004	goto cancel_ref;
2005
2006	root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
2007	KERNFS_ROOT_CREATE_DEACTIVATED,
2008	root_cgrp);
2009	if (IS_ERR(root->kf_root)) {
2010	ret = PTR_ERR(root->kf_root);
2011	goto exit_root_id;
2012	}
2013	root_cgrp->kn = root->kf_root->kn;
2014
2015	ret = css_populate_dir(&root_cgrp->self);
2016	if (ret)
2017	goto destroy_root;
2018
2019	ret = rebind_subsystems(root, ss_mask);
2020	if (ret)
2021	goto destroy_root;
2022
2023	trace_cgroup_setup_root(root);
2024
2025	/*
2026	* There must be no failure case after here, since rebinding takes
2027	* care of subsystems' refcounts, which are explicitly dropped in
2028	* the failure exit path.
2029	*/
2030	list_add(&root->root_list, &cgroup_roots);
2031	cgroup_root_count++;
2032
2033	/*
2034	* Link the root cgroup in this hierarchy into all the css_set
2035	* objects.
2036	*/
2037	spin_lock_irq(&css_set_lock);
2038	hash_for_each(css_set_table, i, cset, hlist) {
2039	link_css_set(&tmp_links, cset, root_cgrp);
2040	if (css_set_populated(cset))
2041	cgroup_update_populated(root_cgrp, true);
2042	}
2043	spin_unlock_irq(&css_set_lock);
2044
2045	BUG_ON(!list_empty(&root_cgrp->self.children));
2046	BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2047
2048	kernfs_activate(root_cgrp->kn);
2049	ret = 0;
2050	goto out;
2051
2052	destroy_root:
2053	kernfs_destroy_root(root->kf_root);
2054	root->kf_root = NULL;
2055	exit_root_id:
2056	cgroup_exit_root_id(root);
2057	cancel_ref:
2058	percpu_ref_exit(&root_cgrp->self.refcnt);
2059	out:
2060	free_cgrp_cset_links(&tmp_links);
2061	return ret;
2062	}
2063
2064	static struct dentry *cgroup_mount(struct file_system_type *fs_type,
2065	int flags, const char *unused_dev_name,
2066	void *data)
2067	{
2068	bool is_v2 = fs_type == &cgroup2_fs_type;
2069	struct super_block *pinned_sb = NULL;
2070	struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
2071	struct cgroup_subsys *ss;
2072	struct cgroup_root *root;
2073	struct cgroup_sb_opts opts;
2074	struct dentry *dentry;
2075	int ret;
2076	int i;
2077	bool new_sb;
2078
2079	get_cgroup_ns(ns);
2080
2081	/* Check if the caller has permission to mount. */
2082	if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
2083	put_cgroup_ns(ns);
2084	return ERR_PTR(-EPERM);
2085	}
2086
2087	/*
2088	* The first time anyone tries to mount a cgroup, enable the list
2089	* linking each css_set to its tasks and fix up all existing tasks.
2090	*/
2091	if (!use_task_css_set_links)
2092	cgroup_enable_task_cg_lists();
2093
2094	if (is_v2) {
2095	if (data) {
2096	pr_err("cgroup2: unknown option \"%s\"\n", (char *)data);
2097	put_cgroup_ns(ns);
2098	return ERR_PTR(-EINVAL);
2099	}
2100	cgrp_dfl_visible = true;
2101	root = &cgrp_dfl_root;
2102	cgroup_get(&root->cgrp);
2103	goto out_mount;
2104	}
2105
2106	cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
2107
2108	/* First find the desired set of subsystems */
2109	ret = parse_cgroupfs_options(data, &opts);
2110	if (ret)
2111	goto out_unlock;
2112
2113	/*
2114	* Destruction of cgroup root is asynchronous, so subsystems may
2115	* still be dying after the previous unmount. Let's drain the
2116	* dying subsystems. We just need to ensure that the ones
2117	* unmounted previously finish dying and don't care about new ones
2118	* starting. Testing ref liveliness is good enough.
2119	*/
2120	for_each_subsys(ss, i) {
2121	if (!(opts.subsys_mask & (1 << i)) ||
2122	ss->root == &cgrp_dfl_root)
2123	continue;
2124
2125	if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) {
2126	mutex_unlock(&cgroup_mutex);
2127	msleep(10);
2128	ret = restart_syscall();
2129	goto out_free;
2130	}
2131	cgroup_put(&ss->root->cgrp);
2132	}
2133
2134	for_each_root(root) {
2135	bool name_match = false;
2136
2137	if (root == &cgrp_dfl_root)
2138	continue;
2139
2140	/*
2141	* If we asked for a name then it must match. Also, if
2142	* name matches but sybsys_mask doesn't, we should fail.
2143	* Remember whether name matched.
2144	*/
2145	if (opts.name) {
2146	if (strcmp(opts.name, root->name))
2147	continue;
2148	name_match = true;
2149	}
2150
2151	/*
2152	* If we asked for subsystems (or explicitly for no
2153	* subsystems) then they must match.
2154	*/
2155	if ((opts.subsys_mask || opts.none) &&
2156	(opts.subsys_mask != root->subsys_mask)) {
2157	if (!name_match)
2158	continue;
2159	ret = -EBUSY;
2160	goto out_unlock;
2161	}
2162
2163	if (root->flags ^ opts.flags)
2164	pr_warn("new mount options do not match the existing superblock, will be ignored\n");
2165
2166	/*
2167	* We want to reuse @root whose lifetime is governed by its
2168	* ->cgrp. Let's check whether @root is alive and keep it
2169	* that way. As cgroup_kill_sb() can happen anytime, we
2170	* want to block it by pinning the sb so that @root doesn't
2171	* get killed before mount is complete.
2172	*
2173	* With the sb pinned, tryget_live can reliably indicate
2174	* whether @root can be reused. If it's being killed,
2175	* drain it. We can use wait_queue for the wait but this
2176	* path is super cold. Let's just sleep a bit and retry.
2177	*/
2178	pinned_sb = kernfs_pin_sb(root->kf_root, NULL);
2179	if (IS_ERR(pinned_sb) ||
2180	!percpu_ref_tryget_live(&root->cgrp.self.refcnt)) {
2181	mutex_unlock(&cgroup_mutex);
2182	if (!IS_ERR_OR_NULL(pinned_sb))
2183	deactivate_super(pinned_sb);
2184	msleep(10);
2185	ret = restart_syscall();
2186	goto out_free;
2187	}
2188
2189	ret = 0;
2190	goto out_unlock;
2191	}
2192
2193	/*
2194	* No such thing, create a new one. name= matching without subsys
2195	* specification is allowed for already existing hierarchies but we
2196	* can't create new one without subsys specification.
2197	*/
2198	if (!opts.subsys_mask && !opts.none) {
2199	ret = -EINVAL;
2200	goto out_unlock;
2201	}
2202
2203	/* Hierarchies may only be created in the initial cgroup namespace. */
2204	if (ns != &init_cgroup_ns) {
2205	ret = -EPERM;
2206	goto out_unlock;
2207	}
2208
2209	root = kzalloc(sizeof(*root), GFP_KERNEL);
2210	if (!root) {
2211	ret = -ENOMEM;
2212	goto out_unlock;
2213	}
2214
2215	init_cgroup_root(root, &opts);
2216
2217	ret = cgroup_setup_root(root, opts.subsys_mask);
2218	if (ret)
2219	cgroup_free_root(root);
2220
2221	out_unlock:
2222	mutex_unlock(&cgroup_mutex);
2223	out_free:
2224	kfree(opts.release_agent);
2225	kfree(opts.name);
2226
2227	if (ret) {
2228	put_cgroup_ns(ns);
2229	return ERR_PTR(ret);
2230	}
2231	out_mount:
2232	dentry = kernfs_mount(fs_type, flags, root->kf_root,
2233	is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC,
2234	&new_sb);
2235
2236	/*
2237	* In non-init cgroup namespace, instead of root cgroup's
2238	* dentry, we return the dentry corresponding to the
2239	* cgroupns->root_cgrp.
2240	*/
2241	if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
2242	struct dentry *nsdentry;
2243	struct cgroup *cgrp;
2244
2245	mutex_lock(&cgroup_mutex);
2246	spin_lock_irq(&css_set_lock);
2247
2248	cgrp = cset_cgroup_from_root(ns->root_cset, root);
2249
2250	spin_unlock_irq(&css_set_lock);
2251	mutex_unlock(&cgroup_mutex);
2252
2253	nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
2254	dput(dentry);
2255	dentry = nsdentry;
2256	}
2257
2258	if (IS_ERR(dentry) || !new_sb)
2259	cgroup_put(&root->cgrp);
2260
2261	/*
2262	* If @pinned_sb, we're reusing an existing root and holding an
2263	* extra ref on its sb. Mount is complete. Put the extra ref.
2264	*/
2265	if (pinned_sb) {
2266	WARN_ON(new_sb);
2267	deactivate_super(pinned_sb);
2268	}
2269
2270	put_cgroup_ns(ns);
2271	return dentry;
2272	}
2273
2274	static void cgroup_kill_sb(struct super_block *sb)
2275	{
2276	struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2277	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2278
2279	/*
2280	* If @root doesn't have any mounts or children, start killing it.
2281	* This prevents new mounts by disabling percpu_ref_tryget_live().
2282	* cgroup_mount() may wait for @root's release.
2283	*
2284	* And don't kill the default root.
2285	*/
2286	if (!list_empty(&root->cgrp.self.children) ||
2287	root == &cgrp_dfl_root)
2288	cgroup_put(&root->cgrp);
2289	else
2290	percpu_ref_kill(&root->cgrp.self.refcnt);
2291
2292	kernfs_kill_sb(sb);
2293	}
2294
2295	static struct file_system_type cgroup_fs_type = {
2296	.name = "cgroup",
2297	.mount = cgroup_mount,
2298	.kill_sb = cgroup_kill_sb,
2299	.fs_flags = FS_USERNS_MOUNT,
2300	};
2301
2302	static struct file_system_type cgroup2_fs_type = {
2303	.name = "cgroup2",
2304	.mount = cgroup_mount,
2305	.kill_sb = cgroup_kill_sb,
2306	.fs_flags = FS_USERNS_MOUNT,
2307	};
2308
2309	static int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2310	struct cgroup_namespace *ns)
2311	{
2312	struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2313
2314	return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2315	}
2316
2317	int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2318	struct cgroup_namespace *ns)
2319	{
2320	int ret;
2321
2322	mutex_lock(&cgroup_mutex);
2323	spin_lock_irq(&css_set_lock);
2324
2325	ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2326
2327	spin_unlock_irq(&css_set_lock);
2328	mutex_unlock(&cgroup_mutex);
2329
2330	return ret;
2331	}
2332	EXPORT_SYMBOL_GPL(cgroup_path_ns);
2333
2334	/**
2335	* task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2336	* @task: target task
2337	* @buf: the buffer to write the path into
2338	* @buflen: the length of the buffer
2339	*
2340	* Determine @task's cgroup on the first (the one with the lowest non-zero
2341	* hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2342	* function grabs cgroup_mutex and shouldn't be used inside locks used by
2343	* cgroup controller callbacks.
2344	*
2345	* Return value is the same as kernfs_path().
2346	*/
2347	int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2348	{
2349	struct cgroup_root *root;
2350	struct cgroup *cgrp;
2351	int hierarchy_id = 1;
2352	int ret;
2353
2354	mutex_lock(&cgroup_mutex);
2355	spin_lock_irq(&css_set_lock);
2356
2357	root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2358
2359	if (root) {
2360	cgrp = task_cgroup_from_root(task, root);
2361	ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2362	} else {
2363	/* if no hierarchy exists, everyone is in "/" */
2364	ret = strlcpy(buf, "/", buflen);
2365	}
2366
2367	spin_unlock_irq(&css_set_lock);
2368	mutex_unlock(&cgroup_mutex);
2369	return ret;
2370	}
2371	EXPORT_SYMBOL_GPL(task_cgroup_path);
2372
2373	/* used to track tasks and other necessary states during migration */
2374	struct cgroup_taskset {
2375	/* the src and dst cset list running through cset->mg_node */
2376	struct list_head src_csets;
2377	struct list_head dst_csets;
2378
2379	/* the subsys currently being processed */
2380	int ssid;
2381
2382	/*
2383	* Fields for cgroup_taskset_*() iteration.
2384	*
2385	* Before migration is committed, the target migration tasks are on
2386	* ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2387	* the csets on ->dst_csets. ->csets point to either ->src_csets
2388	* or ->dst_csets depending on whether migration is committed.
2389	*
2390	* ->cur_csets and ->cur_task point to the current task position
2391	* during iteration.
2392	*/
2393	struct list_head *csets;
2394	struct css_set *cur_cset;
2395	struct task_struct *cur_task;
2396	};
2397
2398	#define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2399	.src_csets = LIST_HEAD_INIT(tset.src_csets), \
2400	.dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2401	.csets = &tset.src_csets, \
2402	}
2403
2404	/**
2405	* cgroup_taskset_add - try to add a migration target task to a taskset
2406	* @task: target task
2407	* @tset: target taskset
2408	*
2409	* Add @task, which is a migration target, to @tset. This function becomes
2410	* noop if @task doesn't need to be migrated. @task's css_set should have
2411	* been added as a migration source and @task->cg_list will be moved from
2412	* the css_set's tasks list to mg_tasks one.
2413	*/
2414	static void cgroup_taskset_add(struct task_struct *task,
2415	struct cgroup_taskset *tset)
2416	{
2417	struct css_set *cset;
2418
2419	lockdep_assert_held(&css_set_lock);
2420
2421	/* @task either already exited or can't exit until the end */
2422	if (task->flags & PF_EXITING)
2423	return;
2424
2425	/* leave @task alone if post_fork() hasn't linked it yet */
2426	if (list_empty(&task->cg_list))
2427	return;
2428
2429	cset = task_css_set(task);
2430	if (!cset->mg_src_cgrp)
2431	return;
2432
2433	list_move_tail(&task->cg_list, &cset->mg_tasks);
2434	if (list_empty(&cset->mg_node))
2435	list_add_tail(&cset->mg_node, &tset->src_csets);
2436	if (list_empty(&cset->mg_dst_cset->mg_node))
2437	list_move_tail(&cset->mg_dst_cset->mg_node,
2438	&tset->dst_csets);
2439	}
2440
2441	/**
2442	* cgroup_taskset_first - reset taskset and return the first task
2443	* @tset: taskset of interest
2444	* @dst_cssp: output variable for the destination css
2445	*
2446	* @tset iteration is initialized and the first task is returned.
2447	*/
2448	struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2449	struct cgroup_subsys_state **dst_cssp)
2450	{
2451	tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2452	tset->cur_task = NULL;
2453
2454	return cgroup_taskset_next(tset, dst_cssp);
2455	}
2456
2457	/**
2458	* cgroup_taskset_next - iterate to the next task in taskset
2459	* @tset: taskset of interest
2460	* @dst_cssp: output variable for the destination css
2461	*
2462	* Return the next task in @tset. Iteration must have been initialized
2463	* with cgroup_taskset_first().
2464	*/
2465	struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2466	struct cgroup_subsys_state **dst_cssp)
2467	{
2468	struct css_set *cset = tset->cur_cset;
2469	struct task_struct *task = tset->cur_task;
2470
2471	while (&cset->mg_node != tset->csets) {
2472	if (!task)
2473	task = list_first_entry(&cset->mg_tasks,
2474	struct task_struct, cg_list);
2475	else
2476	task = list_next_entry(task, cg_list);
2477
2478	if (&task->cg_list != &cset->mg_tasks) {
2479	tset->cur_cset = cset;
2480	tset->cur_task = task;
2481
2482	/*
2483	* This function may be called both before and
2484	* after cgroup_taskset_migrate(). The two cases
2485	* can be distinguished by looking at whether @cset
2486	* has its ->mg_dst_cset set.
2487	*/
2488	if (cset->mg_dst_cset)
2489	*dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2490	else
2491	*dst_cssp = cset->subsys[tset->ssid];
2492
2493	return task;
2494	}
2495
2496	cset = list_next_entry(cset, mg_node);
2497	task = NULL;
2498	}
2499
2500	return NULL;
2501	}
2502
2503	/**
2504	* cgroup_taskset_migrate - migrate a taskset
2505	* @tset: taget taskset
2506	* @root: cgroup root the migration is taking place on
2507	*
2508	* Migrate tasks in @tset as setup by migration preparation functions.
2509	* This function fails iff one of the ->can_attach callbacks fails and
2510	* guarantees that either all or none of the tasks in @tset are migrated.
2511	* @tset is consumed regardless of success.
2512	*/
2513	static int cgroup_taskset_migrate(struct cgroup_taskset *tset,
2514	struct cgroup_root *root)
2515	{
2516	struct cgroup_subsys *ss;
2517	struct task_struct *task, *tmp_task;
2518	struct css_set *cset, *tmp_cset;
2519	int ssid, failed_ssid, ret;
2520
2521	/* methods shouldn't be called if no task is actually migrating */
2522	if (list_empty(&tset->src_csets))
2523	return 0;
2524
2525	/* check that we can legitimately attach to the cgroup */
2526	do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2527	if (ss->can_attach) {
2528	tset->ssid = ssid;
2529	ret = ss->can_attach(tset);
2530	if (ret) {
2531	failed_ssid = ssid;
2532	goto out_cancel_attach;
2533	}
2534	}
2535	} while_each_subsys_mask();
2536
2537	/*
2538	* Now that we're guaranteed success, proceed to move all tasks to
2539	* the new cgroup. There are no failure cases after here, so this
2540	* is the commit point.
2541	*/
2542	spin_lock_irq(&css_set_lock);
2543	list_for_each_entry(cset, &tset->src_csets, mg_node) {
2544	list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2545	struct css_set *from_cset = task_css_set(task);
2546	struct css_set *to_cset = cset->mg_dst_cset;
2547
2548	get_css_set(to_cset);
2549	css_set_move_task(task, from_cset, to_cset, true);
2550	put_css_set_locked(from_cset);
2551	}
2552	}
2553	spin_unlock_irq(&css_set_lock);
2554
2555	/*
2556	* Migration is committed, all target tasks are now on dst_csets.
2557	* Nothing is sensitive to fork() after this point. Notify
2558	* controllers that migration is complete.
2559	*/
2560	tset->csets = &tset->dst_csets;
2561
2562	do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2563	if (ss->attach) {
2564	tset->ssid = ssid;
2565	ss->attach(tset);
2566	}
2567	} while_each_subsys_mask();
2568
2569	ret = 0;
2570	goto out_release_tset;
2571
2572	out_cancel_attach:
2573	do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2574	if (ssid == failed_ssid)
2575	break;
2576	if (ss->cancel_attach) {
2577	tset->ssid = ssid;
2578	ss->cancel_attach(tset);
2579	}
2580	} while_each_subsys_mask();
2581	out_release_tset:
2582	spin_lock_irq(&css_set_lock);
2583	list_splice_init(&tset->dst_csets, &tset->src_csets);
2584	list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2585	list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2586	list_del_init(&cset->mg_node);
2587	}
2588	spin_unlock_irq(&css_set_lock);
2589	return ret;
2590	}
2591
2592	/**
2593	* cgroup_may_migrate_to - verify whether a cgroup can be migration destination
2594	* @dst_cgrp: destination cgroup to test
2595	*
2596	* On the default hierarchy, except for the root, subtree_control must be
2597	* zero for migration destination cgroups with tasks so that child cgroups
2598	* don't compete against tasks.
2599	*/
2600	static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp)
2601	{
2602	return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) ||
2603	!dst_cgrp->subtree_control;
2604	}
2605
2606	/**
2607	* cgroup_migrate_finish - cleanup after attach
2608	* @preloaded_csets: list of preloaded css_sets
2609	*
2610	* Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2611	* those functions for details.
2612	*/
2613	static void cgroup_migrate_finish(struct list_head *preloaded_csets)
2614	{
2615	struct css_set *cset, *tmp_cset;
2616
2617	lockdep_assert_held(&cgroup_mutex);
2618
2619	spin_lock_irq(&css_set_lock);
2620	list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
2621	cset->mg_src_cgrp = NULL;
2622	cset->mg_dst_cgrp = NULL;
2623	cset->mg_dst_cset = NULL;
2624	list_del_init(&cset->mg_preload_node);
2625	put_css_set_locked(cset);
2626	}
2627	spin_unlock_irq(&css_set_lock);
2628	}
2629
2630	/**
2631	* cgroup_migrate_add_src - add a migration source css_set
2632	* @src_cset: the source css_set to add
2633	* @dst_cgrp: the destination cgroup
2634	* @preloaded_csets: list of preloaded css_sets
2635	*
2636	* Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2637	* @src_cset and add it to @preloaded_csets, which should later be cleaned
2638	* up by cgroup_migrate_finish().
2639	*
2640	* This function may be called without holding cgroup_threadgroup_rwsem
2641	* even if the target is a process. Threads may be created and destroyed
2642	* but as long as cgroup_mutex is not dropped, no new css_set can be put
2643	* into play and the preloaded css_sets are guaranteed to cover all
2644	* migrations.
2645	*/
2646	static void cgroup_migrate_add_src(struct css_set *src_cset,
2647	struct cgroup *dst_cgrp,
2648	struct list_head *preloaded_csets)
2649	{
2650	struct cgroup *src_cgrp;
2651
2652	lockdep_assert_held(&cgroup_mutex);
2653	lockdep_assert_held(&css_set_lock);
2654
2655	/*
2656	* If ->dead, @src_set is associated with one or more dead cgroups
2657	* and doesn't contain any migratable tasks. Ignore it early so
2658	* that the rest of migration path doesn't get confused by it.
2659	*/
2660	if (src_cset->dead)
2661	return;
2662
2663	src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2664
2665	if (!list_empty(&src_cset->mg_preload_node))
2666	return;
2667
2668	WARN_ON(src_cset->mg_src_cgrp);
2669	WARN_ON(src_cset->mg_dst_cgrp);
2670	WARN_ON(!list_empty(&src_cset->mg_tasks));
2671	WARN_ON(!list_empty(&src_cset->mg_node));
2672
2673	src_cset->mg_src_cgrp = src_cgrp;
2674	src_cset->mg_dst_cgrp = dst_cgrp;
2675	get_css_set(src_cset);
2676	list_add(&src_cset->mg_preload_node, preloaded_csets);
2677	}
2678
2679	/**
2680	* cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2681	* @preloaded_csets: list of preloaded source css_sets
2682	*
2683	* Tasks are about to be moved and all the source css_sets have been
2684	* preloaded to @preloaded_csets. This function looks up and pins all
2685	* destination css_sets, links each to its source, and append them to
2686	* @preloaded_csets.
2687	*
2688	* This function must be called after cgroup_migrate_add_src() has been
2689	* called on each migration source css_set. After migration is performed
2690	* using cgroup_migrate(), cgroup_migrate_finish() must be called on
2691	* @preloaded_csets.
2692	*/
2693	static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets)
2694	{
2695	LIST_HEAD(csets);
2696	struct css_set *src_cset, *tmp_cset;
2697
2698	lockdep_assert_held(&cgroup_mutex);
2699
2700	/* look up the dst cset for each src cset and link it to src */
2701	list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) {
2702	struct css_set *dst_cset;
2703
2704	dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2705	if (!dst_cset)
2706	goto err;
2707
2708	WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2709
2710	/*
2711	* If src cset equals dst, it's noop. Drop the src.
2712	* cgroup_migrate() will skip the cset too. Note that we
2713	* can't handle src == dst as some nodes are used by both.
2714	*/
2715	if (src_cset == dst_cset) {
2716	src_cset->mg_src_cgrp = NULL;
2717	src_cset->mg_dst_cgrp = NULL;
2718	list_del_init(&src_cset->mg_preload_node);
2719	put_css_set(src_cset);
2720	put_css_set(dst_cset);
2721	continue;
2722	}
2723
2724	src_cset->mg_dst_cset = dst_cset;
2725
2726	if (list_empty(&dst_cset->mg_preload_node))
2727	list_add(&dst_cset->mg_preload_node, &csets);
2728	else
2729	put_css_set(dst_cset);
2730	}
2731
2732	list_splice_tail(&csets, preloaded_csets);
2733	return 0;
2734	err:
2735	cgroup_migrate_finish(&csets);
2736	return -ENOMEM;
2737	}
2738
2739	/**
2740	* cgroup_migrate - migrate a process or task to a cgroup
2741	* @leader: the leader of the process or the task to migrate
2742	* @threadgroup: whether @leader points to the whole process or a single task
2743	* @root: cgroup root migration is taking place on
2744	*
2745	* Migrate a process or task denoted by @leader. If migrating a process,
2746	* the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2747	* responsible for invoking cgroup_migrate_add_src() and
2748	* cgroup_migrate_prepare_dst() on the targets before invoking this
2749	* function and following up with cgroup_migrate_finish().
2750	*
2751	* As long as a controller's ->can_attach() doesn't fail, this function is
2752	* guaranteed to succeed. This means that, excluding ->can_attach()
2753	* failure, when migrating multiple targets, the success or failure can be
2754	* decided for all targets by invoking group_migrate_prepare_dst() before
2755	* actually starting migrating.
2756	*/
2757	static int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2758	struct cgroup_root *root)
2759	{
2760	struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2761	struct task_struct *task;
2762
2763	/*
2764	* Prevent freeing of tasks while we take a snapshot. Tasks that are
2765	* already PF_EXITING could be freed from underneath us unless we
2766	* take an rcu_read_lock.
2767	*/
2768	spin_lock_irq(&css_set_lock);
2769	rcu_read_lock();
2770	task = leader;
2771	do {
2772	cgroup_taskset_add(task, &tset);
2773	if (!threadgroup)
2774	break;
2775	} while_each_thread(leader, task);
2776	rcu_read_unlock();
2777	spin_unlock_irq(&css_set_lock);
2778
2779	return cgroup_taskset_migrate(&tset, root);
2780	}
2781
2782	/**
2783	* cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2784	* @dst_cgrp: the cgroup to attach to
2785	* @leader: the task or the leader of the threadgroup to be attached
2786	* @threadgroup: attach the whole threadgroup?
2787	*
2788	* Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2789	*/
2790	static int cgroup_attach_task(struct cgroup *dst_cgrp,
2791	struct task_struct *leader, bool threadgroup)
2792	{
2793	LIST_HEAD(preloaded_csets);
2794	struct task_struct *task;
2795	int ret;
2796
2797	if (!cgroup_may_migrate_to(dst_cgrp))
2798	return -EBUSY;
2799
2800	/* look up all src csets */
2801	spin_lock_irq(&css_set_lock);
2802	rcu_read_lock();
2803	task = leader;
2804	do {
2805	cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2806	&preloaded_csets);
2807	if (!threadgroup)
2808	break;
2809	} while_each_thread(leader, task);
2810	rcu_read_unlock();
2811	spin_unlock_irq(&css_set_lock);
2812
2813	/* prepare dst csets and commit */
2814	ret = cgroup_migrate_prepare_dst(&preloaded_csets);
2815	if (!ret)
2816	ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root);
2817
2818	cgroup_migrate_finish(&preloaded_csets);
2819
2820	if (!ret)
2821	trace_cgroup_attach_task(dst_cgrp, leader, threadgroup);
2822
2823	return ret;
2824	}
2825
2826	static int cgroup_procs_write_permission(struct task_struct *task,
2827	struct cgroup *dst_cgrp,
2828	struct kernfs_open_file *of)
2829	{
2830	const struct cred *cred = current_cred();
2831	const struct cred *tcred = get_task_cred(task);
2832	int ret = 0;
2833
2834	/*
2835	* even if we're attaching all tasks in the thread group, we only
2836	* need to check permissions on one of them.
2837	*/
2838	if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2839	!uid_eq(cred->euid, tcred->uid) &&
2840	!uid_eq(cred->euid, tcred->suid) &&
2841	!ns_capable(tcred->user_ns, CAP_SYS_NICE))
2842	ret = -EACCES;
2843
2844	if (!ret && cgroup_on_dfl(dst_cgrp)) {
2845	struct super_block *sb = of->file->f_path.dentry->d_sb;
2846	struct cgroup *cgrp;
2847	struct inode *inode;
2848
2849	spin_lock_irq(&css_set_lock);
2850	cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
2851	spin_unlock_irq(&css_set_lock);
2852
2853	while (!cgroup_is_descendant(dst_cgrp, cgrp))
2854	cgrp = cgroup_parent(cgrp);
2855
2856	ret = -ENOMEM;
2857	inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
2858	if (inode) {
2859	ret = inode_permission(inode, MAY_WRITE);
2860	iput(inode);
2861	}
2862	}
2863
2864	put_cred(tcred);
2865	return ret;
2866	}
2867
2868	/*
2869	* Find the task_struct of the task to attach by vpid and pass it along to the
2870	* function to attach either it or all tasks in its threadgroup. Will lock
2871	* cgroup_mutex and threadgroup.
2872	*/
2873	static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
2874	size_t nbytes, loff_t off, bool threadgroup)
2875	{
2876	struct task_struct *tsk;
2877	struct cgroup_subsys *ss;
2878	struct cgroup *cgrp;
2879	pid_t pid;
2880	int ssid, ret;
2881
2882	if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2883	return -EINVAL;
2884
2885	cgrp = cgroup_kn_lock_live(of->kn, false);
2886	if (!cgrp)
2887	return -ENODEV;
2888
2889	percpu_down_write(&cgroup_threadgroup_rwsem);
2890	rcu_read_lock();
2891	if (pid) {
2892	tsk = find_task_by_vpid(pid);
2893	if (!tsk) {
2894	ret = -ESRCH;
2895	goto out_unlock_rcu;
2896	}
2897	} else {
2898	tsk = current;
2899	}
2900
2901	if (threadgroup)
2902	tsk = tsk->group_leader;
2903
2904	/*
2905	* kthreads may acquire PF_NO_SETAFFINITY during initialization.
2906	* If userland migrates such a kthread to a non-root cgroup, it can
2907	* become trapped in a cpuset, or RT kthread may be born in a
2908	* cgroup with no rt_runtime allocated. Just say no.
2909	*/
2910	if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2911	ret = -EINVAL;
2912	goto out_unlock_rcu;
2913	}
2914
2915	get_task_struct(tsk);
2916	rcu_read_unlock();
2917
2918	ret = cgroup_procs_write_permission(tsk, cgrp, of);
2919	if (!ret)
2920	ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2921
2922	put_task_struct(tsk);
2923	goto out_unlock_threadgroup;
2924
2925	out_unlock_rcu:
2926	rcu_read_unlock();
2927	out_unlock_threadgroup:
2928	percpu_up_write(&cgroup_threadgroup_rwsem);
2929	for_each_subsys(ss, ssid)
2930	if (ss->post_attach)
2931	ss->post_attach();
2932	cgroup_kn_unlock(of->kn);
2933	return ret ?: nbytes;
2934	}
2935
2936	/**
2937	* cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2938	* @from: attach to all cgroups of a given task
2939	* @tsk: the task to be attached
2940	*/
2941	int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2942	{
2943	struct cgroup_root *root;
2944	int retval = 0;
2945
2946	mutex_lock(&cgroup_mutex);
2947	percpu_down_write(&cgroup_threadgroup_rwsem);
2948	for_each_root(root) {
2949	struct cgroup *from_cgrp;
2950
2951	if (root == &cgrp_dfl_root)
2952	continue;
2953
2954	spin_lock_irq(&css_set_lock);
2955	from_cgrp = task_cgroup_from_root(from, root);
2956	spin_unlock_irq(&css_set_lock);
2957
2958	retval = cgroup_attach_task(from_cgrp, tsk, false);
2959	if (retval)
2960	break;
2961	}
2962	percpu_up_write(&cgroup_threadgroup_rwsem);
2963	mutex_unlock(&cgroup_mutex);
2964
2965	return retval;
2966	}
2967	EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2968
2969	static ssize_t cgroup_tasks_write(struct kernfs_open_file *of,
2970	char *buf, size_t nbytes, loff_t off)
2971	{
2972	return __cgroup_procs_write(of, buf, nbytes, off, false);
2973	}
2974
2975	static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
2976	char *buf, size_t nbytes, loff_t off)
2977	{
2978	return __cgroup_procs_write(of, buf, nbytes, off, true);
2979	}
2980
2981	static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
2982	char *buf, size_t nbytes, loff_t off)
2983	{
2984	struct cgroup *cgrp;
2985
2986	BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2987
2988	cgrp = cgroup_kn_lock_live(of->kn, false);
2989	if (!cgrp)
2990	return -ENODEV;
2991	spin_lock(&release_agent_path_lock);
2992	strlcpy(cgrp->root->release_agent_path, strstrip(buf),
2993	sizeof(cgrp->root->release_agent_path));
2994	spin_unlock(&release_agent_path_lock);
2995	cgroup_kn_unlock(of->kn);
2996	return nbytes;
2997	}
2998
2999	static int cgroup_release_agent_show(struct seq_file *seq, void *v)
3000	{
3001	struct cgroup *cgrp = seq_css(seq)->cgroup;
3002
3003	spin_lock(&release_agent_path_lock);
3004	seq_puts(seq, cgrp->root->release_agent_path);
3005	spin_unlock(&release_agent_path_lock);
3006	seq_putc(seq, '\n');
3007	return 0;
3008	}
3009
3010	static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
3011	{
3012	seq_puts(seq, "0\n");
3013	return 0;
3014	}
3015
3016	static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
3017	{
3018	struct cgroup_subsys *ss;
3019	bool printed = false;
3020	int ssid;
3021
3022	do_each_subsys_mask(ss, ssid, ss_mask) {
3023	if (printed)
3024	seq_putc(seq, ' ');
3025	seq_printf(seq, "%s", ss->name);
3026	printed = true;
3027	} while_each_subsys_mask();
3028	if (printed)
3029	seq_putc(seq, '\n');
3030	}
3031
3032	/* show controllers which are enabled from the parent */
3033	static int cgroup_controllers_show(struct seq_file *seq, void *v)
3034	{
3035	struct cgroup *cgrp = seq_css(seq)->cgroup;
3036
3037	cgroup_print_ss_mask(seq, cgroup_control(cgrp));
3038	return 0;
3039	}
3040
3041	/* show controllers which are enabled for a given cgroup's children */
3042	static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
3043	{
3044	struct cgroup *cgrp = seq_css(seq)->cgroup;
3045
3046	cgroup_print_ss_mask(seq, cgrp->subtree_control);
3047	return 0;
3048	}
3049
3050	/**
3051	* cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
3052	* @cgrp: root of the subtree to update csses for
3053	*
3054	* @cgrp's control masks have changed and its subtree's css associations
3055	* need to be updated accordingly. This function looks up all css_sets
3056	* which are attached to the subtree, creates the matching updated css_sets
3057	* and migrates the tasks to the new ones.
3058	*/
3059	static int cgroup_update_dfl_csses(struct cgroup *cgrp)
3060	{
3061	LIST_HEAD(preloaded_csets);
3062	struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
3063	struct cgroup_subsys_state *d_css;
3064	struct cgroup *dsct;
3065	struct css_set *src_cset;
3066	int ret;
3067
3068	lockdep_assert_held(&cgroup_mutex);
3069
3070	percpu_down_write(&cgroup_threadgroup_rwsem);
3071
3072	/* look up all csses currently attached to @cgrp's subtree */
3073	spin_lock_irq(&css_set_lock);
3074	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3075	struct cgrp_cset_link *link;
3076
3077	list_for_each_entry(link, &dsct->cset_links, cset_link)
3078	cgroup_migrate_add_src(link->cset, dsct,
3079	&preloaded_csets);
3080	}
3081	spin_unlock_irq(&css_set_lock);
3082
3083	/* NULL dst indicates self on default hierarchy */
3084	ret = cgroup_migrate_prepare_dst(&preloaded_csets);
3085	if (ret)
3086	goto out_finish;
3087
3088	spin_lock_irq(&css_set_lock);
3089	list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) {
3090	struct task_struct *task, *ntask;
3091
3092	/* src_csets precede dst_csets, break on the first dst_cset */
3093	if (!src_cset->mg_src_cgrp)
3094	break;
3095
3096	/* all tasks in src_csets need to be migrated */
3097	list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
3098	cgroup_taskset_add(task, &tset);
3099	}
3100	spin_unlock_irq(&css_set_lock);
3101
3102	ret = cgroup_taskset_migrate(&tset, cgrp->root);
3103	out_finish:
3104	cgroup_migrate_finish(&preloaded_csets);
3105	percpu_up_write(&cgroup_threadgroup_rwsem);
3106	return ret;
3107	}
3108
3109	/**
3110	* cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3111	* @cgrp: root of the target subtree
3112	*
3113	* Because css offlining is asynchronous, userland may try to re-enable a
3114	* controller while the previous css is still around. This function grabs
3115	* cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3116	*/
3117	static void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
3118	__acquires(&cgroup_mutex)
3119	{
3120	struct cgroup *dsct;
3121	struct cgroup_subsys_state *d_css;
3122	struct cgroup_subsys *ss;
3123	int ssid;
3124
3125	restart:
3126	mutex_lock(&cgroup_mutex);
3127
3128	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3129	for_each_subsys(ss, ssid) {
3130	struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3131	DEFINE_WAIT(wait);
3132
3133	if (!css || !percpu_ref_is_dying(&css->refcnt))
3134	continue;
3135
3136	cgroup_get(dsct);
3137	prepare_to_wait(&dsct->offline_waitq, &wait,
3138	TASK_UNINTERRUPTIBLE);
3139
3140	mutex_unlock(&cgroup_mutex);
3141	schedule();
3142	finish_wait(&dsct->offline_waitq, &wait);
3143
3144	cgroup_put(dsct);
3145	goto restart;
3146	}
3147	}
3148	}
3149
3150	/**
3151	* cgroup_save_control - save control masks of a subtree
3152	* @cgrp: root of the target subtree
3153	*
3154	* Save ->subtree_control and ->subtree_ss_mask to the respective old_
3155	* prefixed fields for @cgrp's subtree including @cgrp itself.
3156	*/
3157	static void cgroup_save_control(struct cgroup *cgrp)
3158	{
3159	struct cgroup *dsct;
3160	struct cgroup_subsys_state *d_css;
3161
3162	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3163	dsct->old_subtree_control = dsct->subtree_control;
3164	dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3165	}
3166	}
3167
3168	/**
3169	* cgroup_propagate_control - refresh control masks of a subtree
3170	* @cgrp: root of the target subtree
3171	*
3172	* For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3173	* ->subtree_control and propagate controller availability through the
3174	* subtree so that descendants don't have unavailable controllers enabled.
3175	*/
3176	static void cgroup_propagate_control(struct cgroup *cgrp)
3177	{
3178	struct cgroup *dsct;
3179	struct cgroup_subsys_state *d_css;
3180
3181	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3182	dsct->subtree_control &= cgroup_control(dsct);
3183	dsct->subtree_ss_mask =
3184	cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3185	cgroup_ss_mask(dsct));
3186	}
3187	}
3188
3189	/**
3190	* cgroup_restore_control - restore control masks of a subtree
3191	* @cgrp: root of the target subtree
3192	*
3193	* Restore ->subtree_control and ->subtree_ss_mask from the respective old_
3194	* prefixed fields for @cgrp's subtree including @cgrp itself.
3195	*/
3196	static void cgroup_restore_control(struct cgroup *cgrp)
3197	{
3198	struct cgroup *dsct;
3199	struct cgroup_subsys_state *d_css;
3200
3201	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3202	dsct->subtree_control = dsct->old_subtree_control;
3203	dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3204	}
3205	}
3206
3207	static bool css_visible(struct cgroup_subsys_state *css)
3208	{
3209	struct cgroup_subsys *ss = css->ss;
3210	struct cgroup *cgrp = css->cgroup;
3211
3212	if (cgroup_control(cgrp) & (1 << ss->id))
3213	return true;
3214	if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3215	return false;
3216	return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3217	}
3218
3219	/**
3220	* cgroup_apply_control_enable - enable or show csses according to control
3221	* @cgrp: root of the target subtree
3222	*
3223	* Walk @cgrp's subtree and create new csses or make the existing ones
3224	* visible. A css is created invisible if it's being implicitly enabled
3225	* through dependency. An invisible css is made visible when the userland
3226	* explicitly enables it.
3227	*
3228	* Returns 0 on success, -errno on failure. On failure, csses which have
3229	* been processed already aren't cleaned up. The caller is responsible for
3230	* cleaning up with cgroup_apply_control_disble().
3231	*/
3232	static int cgroup_apply_control_enable(struct cgroup *cgrp)
3233	{
3234	struct cgroup *dsct;
3235	struct cgroup_subsys_state *d_css;
3236	struct cgroup_subsys *ss;
3237	int ssid, ret;
3238
3239	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3240	for_each_subsys(ss, ssid) {
3241	struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3242
3243	WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3244
3245	if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3246	continue;
3247
3248	if (!css) {
3249	css = css_create(dsct, ss);
3250	if (IS_ERR(css))
3251	return PTR_ERR(css);
3252	}
3253
3254	if (css_visible(css)) {
3255	ret = css_populate_dir(css);
3256	if (ret)
3257	return ret;
3258	}
3259	}
3260	}
3261
3262	return 0;
3263	}
3264
3265	/**
3266	* cgroup_apply_control_disable - kill or hide csses according to control
3267	* @cgrp: root of the target subtree
3268	*
3269	* Walk @cgrp's subtree and kill and hide csses so that they match
3270	* cgroup_ss_mask() and cgroup_visible_mask().
3271	*
3272	* A css is hidden when the userland requests it to be disabled while other
3273	* subsystems are still depending on it. The css must not actively control
3274	* resources and be in the vanilla state if it's made visible again later.
3275	* Controllers which may be depended upon should provide ->css_reset() for
3276	* this purpose.
3277	*/
3278	static void cgroup_apply_control_disable(struct cgroup *cgrp)
3279	{
3280	struct cgroup *dsct;
3281	struct cgroup_subsys_state *d_css;
3282	struct cgroup_subsys *ss;
3283	int ssid;
3284
3285	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3286	for_each_subsys(ss, ssid) {
3287	struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3288
3289	WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3290
3291	if (!css)
3292	continue;
3293
3294	if (css->parent &&
3295	!(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3296	kill_css(css);
3297	} else if (!css_visible(css)) {
3298	css_clear_dir(css);
3299	if (ss->css_reset)
3300	ss->css_reset(css);
3301	}
3302	}
3303	}
3304	}
3305
3306	/**
3307	* cgroup_apply_control - apply control mask updates to the subtree
3308	* @cgrp: root of the target subtree
3309	*
3310	* subsystems can be enabled and disabled in a subtree using the following
3311	* steps.
3312	*
3313	* 1. Call cgroup_save_control() to stash the current state.
3314	* 2. Update ->subtree_control masks in the subtree as desired.
3315	* 3. Call cgroup_apply_control() to apply the changes.
3316	* 4. Optionally perform other related operations.
3317	* 5. Call cgroup_finalize_control() to finish up.
3318	*
3319	* This function implements step 3 and propagates the mask changes
3320	* throughout @cgrp's subtree, updates csses accordingly and perform
3321	* process migrations.
3322	*/
3323	static int cgroup_apply_control(struct cgroup *cgrp)
3324	{
3325	int ret;
3326
3327	cgroup_propagate_control(cgrp);
3328
3329	ret = cgroup_apply_control_enable(cgrp);
3330	if (ret)
3331	return ret;
3332
3333	/*
3334	* At this point, cgroup_e_css() results reflect the new csses
3335	* making the following cgroup_update_dfl_csses() properly update
3336	* css associations of all tasks in the subtree.
3337	*/
3338	ret = cgroup_update_dfl_csses(cgrp);
3339	if (ret)
3340	return ret;
3341
3342	return 0;
3343	}
3344
3345	/**
3346	* cgroup_finalize_control - finalize control mask update
3347	* @cgrp: root of the target subtree
3348	* @ret: the result of the update
3349	*
3350	* Finalize control mask update. See cgroup_apply_control() for more info.
3351	*/
3352	static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3353	{
3354	if (ret) {
3355	cgroup_restore_control(cgrp);
3356	cgroup_propagate_control(cgrp);
3357	}
3358
3359	cgroup_apply_control_disable(cgrp);
3360	}
3361
3362	/* change the enabled child controllers for a cgroup in the default hierarchy */
3363	static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3364	char *buf, size_t nbytes,
3365	loff_t off)
3366	{
3367	u16 enable = 0, disable = 0;
3368	struct cgroup *cgrp, *child;
3369	struct cgroup_subsys *ss;
3370	char *tok;
3371	int ssid, ret;
3372
3373	/*
3374	* Parse input - space separated list of subsystem names prefixed
3375	* with either + or -.
3376	*/
3377	buf = strstrip(buf);
3378	while ((tok = strsep(&buf, " "))) {
3379	if (tok[0] == '\0')
3380	continue;
3381	do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3382	if (!cgroup_ssid_enabled(ssid) ||
3383	strcmp(tok + 1, ss->name))
3384	continue;
3385
3386	if (*tok == '+') {
3387	enable |= 1 << ssid;
3388	disable &= ~(1 << ssid);
3389	} else if (*tok == '-') {
3390	disable |= 1 << ssid;
3391	enable &= ~(1 << ssid);
3392	} else {
3393	return -EINVAL;
3394	}
3395	break;
3396	} while_each_subsys_mask();
3397	if (ssid == CGROUP_SUBSYS_COUNT)
3398	return -EINVAL;
3399	}
3400
3401	cgrp = cgroup_kn_lock_live(of->kn, true);
3402	if (!cgrp)
3403	return -ENODEV;
3404
3405	for_each_subsys(ss, ssid) {
3406	if (enable & (1 << ssid)) {
3407	if (cgrp->subtree_control & (1 << ssid)) {
3408	enable &= ~(1 << ssid);
3409	continue;
3410	}
3411
3412	if (!(cgroup_control(cgrp) & (1 << ssid))) {
3413	ret = -ENOENT;
3414	goto out_unlock;
3415	}
3416	} else if (disable & (1 << ssid)) {
3417	if (!(cgrp->subtree_control & (1 << ssid))) {
3418	disable &= ~(1 << ssid);
3419	continue;
3420	}
3421
3422	/* a child has it enabled? */
3423	cgroup_for_each_live_child(child, cgrp) {
3424	if (child->subtree_control & (1 << ssid)) {
3425	ret = -EBUSY;
3426	goto out_unlock;
3427	}
3428	}
3429	}
3430	}
3431
3432	if (!enable && !disable) {
3433	ret = 0;
3434	goto out_unlock;
3435	}
3436
3437	/*
3438	* Except for the root, subtree_control must be zero for a cgroup
3439	* with tasks so that child cgroups don't compete against tasks.
3440	*/
3441	if (enable && cgroup_parent(cgrp)) {
3442	struct cgrp_cset_link *link;
3443
3444	/*
3445	* Because namespaces pin csets too, @cgrp->cset_links
3446	* might not be empty even when @cgrp is empty. Walk and
3447	* verify each cset.
3448	*/
3449	spin_lock_irq(&css_set_lock);
3450
3451	ret = 0;
3452	list_for_each_entry(link, &cgrp->cset_links, cset_link) {
3453	if (css_set_populated(link->cset)) {
3454	ret = -EBUSY;
3455	break;
3456	}
3457	}
3458
3459	spin_unlock_irq(&css_set_lock);
3460
3461	if (ret)
3462	goto out_unlock;
3463	}
3464
3465	/* save and update control masks and prepare csses */
3466	cgroup_save_control(cgrp);
3467
3468	cgrp->subtree_control |= enable;
3469	cgrp->subtree_control &= ~disable;
3470
3471	ret = cgroup_apply_control(cgrp);
3472	cgroup_finalize_control(cgrp, ret);
3473	if (ret)
3474	goto out_unlock;
3475
3476	kernfs_activate(cgrp->kn);
3477	out_unlock:
3478	cgroup_kn_unlock(of->kn);
3479	return ret ?: nbytes;
3480	}
3481
3482	static int cgroup_events_show(struct seq_file *seq, void *v)
3483	{
3484	seq_printf(seq, "populated %d\n",
3485	cgroup_is_populated(seq_css(seq)->cgroup));
3486	return 0;
3487	}
3488
3489	#ifdef CONFIG_PSI
3490	static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3491	{
3492	return psi_show(seq, &seq_css(seq)->cgroup->psi, PSI_IO);
3493	}
3494	static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3495	{
3496	return psi_show(seq, &seq_css(seq)->cgroup->psi, PSI_MEM);
3497	}
3498	static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3499	{
3500	return psi_show(seq, &seq_css(seq)->cgroup->psi, PSI_CPU);
3501	}
3502
3503	static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3504	size_t nbytes, enum psi_res res)
3505	{
3506	struct psi_trigger *new;
3507	struct cgroup *cgrp;
3508
3509	cgrp = cgroup_kn_lock_live(of->kn, false);
3510	if (!cgrp)
3511	return -ENODEV;
3512
3513	cgroup_get(cgrp);
3514	cgroup_kn_unlock(of->kn);
3515
3516	new = psi_trigger_create(&cgrp->psi, buf, nbytes, res);
3517	if (IS_ERR(new)) {
3518	cgroup_put(cgrp);
3519	return PTR_ERR(new);
3520	}
3521
3522	psi_trigger_replace(&of->priv, new);
3523
3524	cgroup_put(cgrp);
3525
3526	return nbytes;
3527	}
3528
3529	static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3530	char *buf, size_t nbytes,
3531	loff_t off)
3532	{
3533	return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3534	}
3535
3536	static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3537	char *buf, size_t nbytes,
3538	loff_t off)
3539	{
3540	return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3541	}
3542
3543	static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3544	char *buf, size_t nbytes,
3545	loff_t off)
3546	{
3547	return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3548	}
3549
3550	static unsigned int cgroup_pressure_poll(struct kernfs_open_file *of,
3551	poll_table *pt)
3552	{
3553	return psi_trigger_poll(&of->priv, of->file, pt);
3554	}
3555
3556	static void cgroup_pressure_release(struct kernfs_open_file *of)
3557	{
3558	psi_trigger_replace(&of->priv, NULL);
3559	}
3560	#endif /* CONFIG_PSI */
3561
3562	static int cgroup_file_open(struct kernfs_open_file *of)
3563	{
3564	struct cftype *cft = of->kn->priv;
3565
3566	if (cft->open)
3567	return cft->open(of);
3568	return 0;
3569	}
3570
3571	static void cgroup_file_release(struct kernfs_open_file *of)
3572	{
3573	struct cftype *cft = of->kn->priv;
3574
3575	if (cft->release)
3576	cft->release(of);
3577	}
3578
3579	static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3580	size_t nbytes, loff_t off)
3581	{
3582	struct cgroup *cgrp = of->kn->parent->priv;
3583	struct cftype *cft = of->kn->priv;
3584	struct cgroup_subsys_state *css;
3585	int ret;
3586
3587	if (cft->write)
3588	return cft->write(of, buf, nbytes, off);
3589
3590	/*
3591	* kernfs guarantees that a file isn't deleted with operations in
3592	* flight, which means that the matching css is and stays alive and
3593	* doesn't need to be pinned. The RCU locking is not necessary
3594	* either. It's just for the convenience of using cgroup_css().
3595	*/
3596	rcu_read_lock();
3597	css = cgroup_css(cgrp, cft->ss);
3598	rcu_read_unlock();
3599
3600	if (cft->write_u64) {
3601	unsigned long long v;
3602	ret = kstrtoull(buf, 0, &v);
3603	if (!ret)
3604	ret = cft->write_u64(css, cft, v);
3605	} else if (cft->write_s64) {
3606	long long v;
3607	ret = kstrtoll(buf, 0, &v);
3608	if (!ret)
3609	ret = cft->write_s64(css, cft, v);
3610	} else {
3611	ret = -EINVAL;
3612	}
3613
3614	return ret ?: nbytes;
3615	}
3616
3617	static unsigned int cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3618	{
3619	struct cftype *cft = of->kn->priv;
3620
3621	if (cft->poll)
3622	return cft->poll(of, pt);
3623
3624	return kernfs_generic_poll(of, pt);
3625	}
3626
3627	static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3628	{
3629	return seq_cft(seq)->seq_start(seq, ppos);
3630	}
3631
3632	static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3633	{
3634	return seq_cft(seq)->seq_next(seq, v, ppos);
3635	}
3636
3637	static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3638	{
3639	if (seq_cft(seq)->seq_stop)
3640	seq_cft(seq)->seq_stop(seq, v);
3641	}
3642
3643	static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3644	{
3645	struct cftype *cft = seq_cft(m);
3646	struct cgroup_subsys_state *css = seq_css(m);
3647
3648	if (cft->seq_show)
3649	return cft->seq_show(m, arg);
3650
3651	if (cft->read_u64)
3652	seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3653	else if (cft->read_s64)
3654	seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3655	else
3656	return -EINVAL;
3657	return 0;
3658	}
3659
3660	static struct kernfs_ops cgroup_kf_single_ops = {
3661	.atomic_write_len = PAGE_SIZE,
3662	.open = cgroup_file_open,
3663	.release = cgroup_file_release,
3664	.write = cgroup_file_write,
3665	.poll = cgroup_file_poll,
3666	.seq_show = cgroup_seqfile_show,
3667	};
3668
3669	static struct kernfs_ops cgroup_kf_ops = {
3670	.atomic_write_len = PAGE_SIZE,
3671	.open = cgroup_file_open,
3672	.release = cgroup_file_release,
3673	.write = cgroup_file_write,
3674	.poll = cgroup_file_poll,
3675	.seq_start = cgroup_seqfile_start,
3676	.seq_next = cgroup_seqfile_next,
3677	.seq_stop = cgroup_seqfile_stop,
3678	.seq_show = cgroup_seqfile_show,
3679	};
3680
3681	/*
3682	* cgroup_rename - Only allow simple rename of directories in place.
3683	*/
3684	static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
3685	const char *new_name_str)
3686	{
3687	struct cgroup *cgrp = kn->priv;
3688	int ret;
3689
3690	if (kernfs_type(kn) != KERNFS_DIR)
3691	return -ENOTDIR;
3692	if (kn->parent != new_parent)
3693	return -EIO;
3694
3695	/*
3696	* This isn't a proper migration and its usefulness is very
3697	* limited. Disallow on the default hierarchy.
3698	*/
3699	if (cgroup_on_dfl(cgrp))
3700	return -EPERM;
3701
3702	/*
3703	* We're gonna grab cgroup_mutex which nests outside kernfs
3704	* active_ref. kernfs_rename() doesn't require active_ref
3705	* protection. Break them before grabbing cgroup_mutex.
3706	*/
3707	kernfs_break_active_protection(new_parent);
3708	kernfs_break_active_protection(kn);
3709
3710	mutex_lock(&cgroup_mutex);
3711
3712	ret = kernfs_rename(kn, new_parent, new_name_str);
3713	if (!ret)
3714	trace_cgroup_rename(cgrp);
3715
3716	mutex_unlock(&cgroup_mutex);
3717
3718	kernfs_unbreak_active_protection(kn);
3719	kernfs_unbreak_active_protection(new_parent);
3720	return ret;
3721	}
3722
3723	/* set uid and gid of cgroup dirs and files to that of the creator */
3724	static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3725	{
3726	struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3727	.ia_uid = current_fsuid(),
3728	.ia_gid = current_fsgid(), };
3729
3730	if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3731	gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3732	return 0;
3733
3734	return kernfs_setattr(kn, &iattr);
3735	}
3736
3737	static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3738	struct cftype *cft)
3739	{
3740	char name[CGROUP_FILE_NAME_MAX];
3741	struct kernfs_node *kn;
3742	struct lock_class_key *key = NULL;
3743	int ret;
3744
3745	#ifdef CONFIG_DEBUG_LOCK_ALLOC
3746	key = &cft->lockdep_key;
3747	#endif
3748	kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3749	cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3750	NULL, key);
3751	if (IS_ERR(kn))
3752	return PTR_ERR(kn);
3753
3754	ret = cgroup_kn_set_ugid(kn);
3755	if (ret) {
3756	kernfs_remove(kn);
3757	return ret;
3758	}
3759
3760	if (cft->file_offset) {
3761	struct cgroup_file *cfile = (void *)css + cft->file_offset;
3762
3763	spin_lock_irq(&cgroup_file_kn_lock);
3764	cfile->kn = kn;
3765	spin_unlock_irq(&cgroup_file_kn_lock);
3766	}
3767
3768	return 0;
3769	}
3770
3771	/**
3772	* cgroup_addrm_files - add or remove files to a cgroup directory
3773	* @css: the target css
3774	* @cgrp: the target cgroup (usually css->cgroup)
3775	* @cfts: array of cftypes to be added
3776	* @is_add: whether to add or remove
3777	*
3778	* Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3779	* For removals, this function never fails.
3780	*/
3781	static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3782	struct cgroup *cgrp, struct cftype cfts[],
3783	bool is_add)
3784	{
3785	struct cftype *cft, *cft_end = NULL;
3786	int ret = 0;
3787
3788	lockdep_assert_held(&cgroup_mutex);
3789
3790	restart:
3791	for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3792	/* does cft->flags tell us to skip this file on @cgrp? */
3793	if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3794	continue;
3795	if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3796	continue;
3797	if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3798	continue;
3799	if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3800	continue;
3801
3802	if (is_add) {
3803	ret = cgroup_add_file(css, cgrp, cft);
3804	if (ret) {
3805	pr_warn("%s: failed to add %s, err=%d\n",
3806	__func__, cft->name, ret);
3807	cft_end = cft;
3808	is_add = false;
3809	goto restart;
3810	}
3811	} else {
3812	cgroup_rm_file(cgrp, cft);
3813	}
3814	}
3815	return ret;
3816	}
3817
3818	static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3819	{
3820	LIST_HEAD(pending);
3821	struct cgroup_subsys *ss = cfts[0].ss;
3822	struct cgroup *root = &ss->root->cgrp;
3823	struct cgroup_subsys_state *css;
3824	int ret = 0;
3825
3826	lockdep_assert_held(&cgroup_mutex);
3827
3828	/* add/rm files for all cgroups created before */
3829	css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3830	struct cgroup *cgrp = css->cgroup;
3831
3832	if (!(css->flags & CSS_VISIBLE))
3833	continue;
3834
3835	ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3836	if (ret)
3837	break;
3838	}
3839
3840	if (is_add && !ret)
3841	kernfs_activate(root->kn);
3842	return ret;
3843	}
3844
3845	static void cgroup_exit_cftypes(struct cftype *cfts)
3846	{
3847	struct cftype *cft;
3848
3849	for (cft = cfts; cft->name[0] != '\0'; cft++) {
3850	/* free copy for custom atomic_write_len, see init_cftypes() */
3851	if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3852	kfree(cft->kf_ops);
3853	cft->kf_ops = NULL;
3854	cft->ss = NULL;
3855
3856	/* revert flags set by cgroup core while adding @cfts */
3857	cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3858	}
3859	}
3860
3861	static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3862	{
3863	struct cftype *cft;
3864
3865	for (cft = cfts; cft->name[0] != '\0'; cft++) {
3866	struct kernfs_ops *kf_ops;
3867
3868	WARN_ON(cft->ss || cft->kf_ops);
3869
3870	if (cft->seq_start)
3871	kf_ops = &cgroup_kf_ops;
3872	else
3873	kf_ops = &cgroup_kf_single_ops;
3874
3875	/*
3876	* Ugh... if @cft wants a custom max_write_len, we need to
3877	* make a copy of kf_ops to set its atomic_write_len.
3878	*/
3879	if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3880	kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3881	if (!kf_ops) {
3882	cgroup_exit_cftypes(cfts);
3883	return -ENOMEM;
3884	}
3885	kf_ops->atomic_write_len = cft->max_write_len;
3886	}
3887
3888	cft->kf_ops = kf_ops;
3889	cft->ss = ss;
3890	}
3891
3892	return 0;
3893	}
3894
3895	static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3896	{
3897	lockdep_assert_held(&cgroup_mutex);
3898
3899	if (!cfts || !cfts[0].ss)
3900	return -ENOENT;
3901
3902	list_del(&cfts->node);
3903	cgroup_apply_cftypes(cfts, false);
3904	cgroup_exit_cftypes(cfts);
3905	return 0;
3906	}
3907
3908	/**
3909	* cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3910	* @cfts: zero-length name terminated array of cftypes
3911	*
3912	* Unregister @cfts. Files described by @cfts are removed from all
3913	* existing cgroups and all future cgroups won't have them either. This
3914	* function can be called anytime whether @cfts' subsys is attached or not.
3915	*
3916	* Returns 0 on successful unregistration, -ENOENT if @cfts is not
3917	* registered.
3918	*/
3919	int cgroup_rm_cftypes(struct cftype *cfts)
3920	{
3921	int ret;
3922
3923	mutex_lock(&cgroup_mutex);
3924	ret = cgroup_rm_cftypes_locked(cfts);
3925	mutex_unlock(&cgroup_mutex);
3926	return ret;
3927	}
3928
3929	/**
3930	* cgroup_add_cftypes - add an array of cftypes to a subsystem
3931	* @ss: target cgroup subsystem
3932	* @cfts: zero-length name terminated array of cftypes
3933	*
3934	* Register @cfts to @ss. Files described by @cfts are created for all
3935	* existing cgroups to which @ss is attached and all future cgroups will
3936	* have them too. This function can be called anytime whether @ss is
3937	* attached or not.
3938	*
3939	* Returns 0 on successful registration, -errno on failure. Note that this
3940	* function currently returns 0 as long as @cfts registration is successful
3941	* even if some file creation attempts on existing cgroups fail.
3942	*/
3943	static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3944	{
3945	int ret;
3946
3947	if (!cgroup_ssid_enabled(ss->id))
3948	return 0;
3949
3950	if (!cfts || cfts[0].name[0] == '\0')
3951	return 0;
3952
3953	ret = cgroup_init_cftypes(ss, cfts);
3954	if (ret)
3955	return ret;
3956
3957	mutex_lock(&cgroup_mutex);
3958
3959	list_add_tail(&cfts->node, &ss->cfts);
3960	ret = cgroup_apply_cftypes(cfts, true);
3961	if (ret)
3962	cgroup_rm_cftypes_locked(cfts);
3963
3964	mutex_unlock(&cgroup_mutex);
3965	return ret;
3966	}
3967
3968	/**
3969	* cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3970	* @ss: target cgroup subsystem
3971	* @cfts: zero-length name terminated array of cftypes
3972	*
3973	* Similar to cgroup_add_cftypes() but the added files are only used for
3974	* the default hierarchy.
3975	*/
3976	int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3977	{
3978	struct cftype *cft;
3979
3980	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3981	cft->flags |= __CFTYPE_ONLY_ON_DFL;
3982	return cgroup_add_cftypes(ss, cfts);
3983	}
3984
3985	/**
3986	* cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3987	* @ss: target cgroup subsystem
3988	* @cfts: zero-length name terminated array of cftypes
3989	*
3990	* Similar to cgroup_add_cftypes() but the added files are only used for
3991	* the legacy hierarchies.
3992	*/
3993	int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3994	{
3995	struct cftype *cft;
3996
3997	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3998	cft->flags |= __CFTYPE_NOT_ON_DFL;
3999	return cgroup_add_cftypes(ss, cfts);
4000	}
4001
4002	/**
4003	* cgroup_file_notify - generate a file modified event for a cgroup_file
4004	* @cfile: target cgroup_file
4005	*
4006	* @cfile must have been obtained by setting cftype->file_offset.
4007	*/
4008	void cgroup_file_notify(struct cgroup_file *cfile)
4009	{
4010	unsigned long flags;
4011
4012	spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4013	if (cfile->kn)
4014	kernfs_notify(cfile->kn);
4015	spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4016	}
4017
4018	/**
4019	* cgroup_task_count - count the number of tasks in a cgroup.
4020	* @cgrp: the cgroup in question
4021	*
4022	* Return the number of tasks in the cgroup. The returned number can be
4023	* higher than the actual number of tasks due to css_set references from
4024	* namespace roots and temporary usages.
4025	*/
4026	static int cgroup_task_count(const struct cgroup *cgrp)
4027	{
4028	int count = 0;
4029	struct cgrp_cset_link *link;
4030
4031	spin_lock_irq(&css_set_lock);
4032	list_for_each_entry(link, &cgrp->cset_links, cset_link)
4033	count += atomic_read(&link->cset->refcount);
4034	spin_unlock_irq(&css_set_lock);
4035	return count;
4036	}
4037
4038	/**
4039	* css_next_child - find the next child of a given css
4040	* @pos: the current position (%NULL to initiate traversal)
4041	* @parent: css whose children to walk
4042	*
4043	* This function returns the next child of @parent and should be called
4044	* under either cgroup_mutex or RCU read lock. The only requirement is
4045	* that @parent and @pos are accessible. The next sibling is guaranteed to
4046	* be returned regardless of their states.
4047	*
4048	* If a subsystem synchronizes ->css_online() and the start of iteration, a
4049	* css which finished ->css_online() is guaranteed to be visible in the
4050	* future iterations and will stay visible until the last reference is put.
4051	* A css which hasn't finished ->css_online() or already finished
4052	* ->css_offline() may show up during traversal. It's each subsystem's
4053	* responsibility to synchronize against on/offlining.
4054	*/
4055	struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4056	struct cgroup_subsys_state *parent)
4057	{
4058	struct cgroup_subsys_state *next;
4059
4060	cgroup_assert_mutex_or_rcu_locked();
4061
4062	/*
4063	* @pos could already have been unlinked from the sibling list.
4064	* Once a cgroup is removed, its ->sibling.next is no longer
4065	* updated when its next sibling changes. CSS_RELEASED is set when
4066	* @pos is taken off list, at which time its next pointer is valid,
4067	* and, as releases are serialized, the one pointed to by the next
4068	* pointer is guaranteed to not have started release yet. This
4069	* implies that if we observe !CSS_RELEASED on @pos in this RCU
4070	* critical section, the one pointed to by its next pointer is
4071	* guaranteed to not have finished its RCU grace period even if we
4072	* have dropped rcu_read_lock() inbetween iterations.
4073	*
4074	* If @pos has CSS_RELEASED set, its next pointer can't be
4075	* dereferenced; however, as each css is given a monotonically
4076	* increasing unique serial number and always appended to the
4077	* sibling list, the next one can be found by walking the parent's
4078	* children until the first css with higher serial number than
4079	* @pos's. While this path can be slower, it happens iff iteration
4080	* races against release and the race window is very small.
4081	*/
4082	if (!pos) {
4083	next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4084	} else if (likely(!(pos->flags & CSS_RELEASED))) {
4085	next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4086	} else {
4087	list_for_each_entry_rcu(next, &parent->children, sibling)
4088	if (next->serial_nr > pos->serial_nr)
4089	break;
4090	}
4091
4092	/*
4093	* @next, if not pointing to the head, can be dereferenced and is
4094	* the next sibling.
4095	*/
4096	if (&next->sibling != &parent->children)
4097	return next;
4098	return NULL;
4099	}
4100
4101	/**
4102	* css_next_descendant_pre - find the next descendant for pre-order walk
4103	* @pos: the current position (%NULL to initiate traversal)
4104	* @root: css whose descendants to walk
4105	*
4106	* To be used by css_for_each_descendant_pre(). Find the next descendant
4107	* to visit for pre-order traversal of @root's descendants. @root is
4108	* included in the iteration and the first node to be visited.
4109	*
4110	* While this function requires cgroup_mutex or RCU read locking, it
4111	* doesn't require the whole traversal to be contained in a single critical
4112	* section. This function will return the correct next descendant as long
4113	* as both @pos and @root are accessible and @pos is a descendant of @root.
4114	*
4115	* If a subsystem synchronizes ->css_online() and the start of iteration, a
4116	* css which finished ->css_online() is guaranteed to be visible in the
4117	* future iterations and will stay visible until the last reference is put.
4118	* A css which hasn't finished ->css_online() or already finished
4119	* ->css_offline() may show up during traversal. It's each subsystem's
4120	* responsibility to synchronize against on/offlining.
4121	*/
4122	struct cgroup_subsys_state *
4123	css_next_descendant_pre(struct cgroup_subsys_state *pos,
4124	struct cgroup_subsys_state *root)
4125	{
4126	struct cgroup_subsys_state *next;
4127
4128	cgroup_assert_mutex_or_rcu_locked();
4129
4130	/* if first iteration, visit @root */
4131	if (!pos)
4132	return root;
4133
4134	/* visit the first child if exists */
4135	next = css_next_child(NULL, pos);
4136	if (next)
4137	return next;
4138
4139	/* no child, visit my or the closest ancestor's next sibling */
4140	while (pos != root) {
4141	next = css_next_child(pos, pos->parent);
4142	if (next)
4143	return next;
4144	pos = pos->parent;
4145	}
4146
4147	return NULL;
4148	}
4149
4150	/**
4151	* css_rightmost_descendant - return the rightmost descendant of a css
4152	* @pos: css of interest
4153	*
4154	* Return the rightmost descendant of @pos. If there's no descendant, @pos
4155	* is returned. This can be used during pre-order traversal to skip
4156	* subtree of @pos.
4157	*
4158	* While this function requires cgroup_mutex or RCU read locking, it
4159	* doesn't require the whole traversal to be contained in a single critical
4160	* section. This function will return the correct rightmost descendant as
4161	* long as @pos is accessible.
4162	*/
4163	struct cgroup_subsys_state *
4164	css_rightmost_descendant(struct cgroup_subsys_state *pos)
4165	{
4166	struct cgroup_subsys_state *last, *tmp;
4167
4168	cgroup_assert_mutex_or_rcu_locked();
4169
4170	do {
4171	last = pos;
4172	/* ->prev isn't RCU safe, walk ->next till the end */
4173	pos = NULL;
4174	css_for_each_child(tmp, last)
4175	pos = tmp;
4176	} while (pos);
4177
4178	return last;
4179	}
4180
4181	static struct cgroup_subsys_state *
4182	css_leftmost_descendant(struct cgroup_subsys_state *pos)
4183	{
4184	struct cgroup_subsys_state *last;
4185
4186	do {
4187	last = pos;
4188	pos = css_next_child(NULL, pos);
4189	} while (pos);
4190
4191	return last;
4192	}
4193
4194	/**
4195	* css_next_descendant_post - find the next descendant for post-order walk
4196	* @pos: the current position (%NULL to initiate traversal)
4197	* @root: css whose descendants to walk
4198	*
4199	* To be used by css_for_each_descendant_post(). Find the next descendant
4200	* to visit for post-order traversal of @root's descendants. @root is
4201	* included in the iteration and the last node to be visited.
4202	*
4203	* While this function requires cgroup_mutex or RCU read locking, it
4204	* doesn't require the whole traversal to be contained in a single critical
4205	* section. This function will return the correct next descendant as long
4206	* as both @pos and @cgroup are accessible and @pos is a descendant of
4207	* @cgroup.
4208	*
4209	* If a subsystem synchronizes ->css_online() and the start of iteration, a
4210	* css which finished ->css_online() is guaranteed to be visible in the
4211	* future iterations and will stay visible until the last reference is put.
4212	* A css which hasn't finished ->css_online() or already finished
4213	* ->css_offline() may show up during traversal. It's each subsystem's
4214	* responsibility to synchronize against on/offlining.
4215	*/
4216	struct cgroup_subsys_state *
4217	css_next_descendant_post(struct cgroup_subsys_state *pos,
4218	struct cgroup_subsys_state *root)
4219	{
4220	struct cgroup_subsys_state *next;
4221
4222	cgroup_assert_mutex_or_rcu_locked();
4223
4224	/* if first iteration, visit leftmost descendant which may be @root */
4225	if (!pos)
4226	return css_leftmost_descendant(root);
4227
4228	/* if we visited @root, we're done */
4229	if (pos == root)
4230	return NULL;
4231
4232	/* if there's an unvisited sibling, visit its leftmost descendant */
4233	next = css_next_child(pos, pos->parent);
4234	if (next)
4235	return css_leftmost_descendant(next);
4236
4237	/* no sibling left, visit parent */
4238	return pos->parent;
4239	}
4240
4241	/**
4242	* css_has_online_children - does a css have online children
4243	* @css: the target css
4244	*
4245	* Returns %true if @css has any online children; otherwise, %false. This
4246	* function can be called from any context but the caller is responsible
4247	* for synchronizing against on/offlining as necessary.
4248	*/
4249	bool css_has_online_children(struct cgroup_subsys_state *css)
4250	{
4251	struct cgroup_subsys_state *child;
4252	bool ret = false;
4253
4254	rcu_read_lock();
4255	css_for_each_child(child, css) {
4256	if (child->flags & CSS_ONLINE) {
4257	ret = true;
4258	break;
4259	}
4260	}
4261	rcu_read_unlock();
4262	return ret;
4263	}
4264
4265	/**
4266	* css_task_iter_advance_css_set - advance a task itererator to the next css_set
4267	* @it: the iterator to advance
4268	*
4269	* Advance @it to the next css_set to walk.
4270	*/
4271	static void css_task_iter_advance_css_set(struct css_task_iter *it)
4272	{
4273	struct list_head *l = it->cset_pos;
4274	struct cgrp_cset_link *link;
4275	struct css_set *cset;
4276
4277	lockdep_assert_held(&css_set_lock);
4278
4279	/* Advance to the next non-empty css_set */
4280	do {
4281	l = l->next;
4282	if (l == it->cset_head) {
4283	it->cset_pos = NULL;
4284	it->task_pos = NULL;
4285	return;
4286	}
4287
4288	if (it->ss) {
4289	cset = container_of(l, struct css_set,
4290	e_cset_node[it->ss->id]);
4291	} else {
4292	link = list_entry(l, struct cgrp_cset_link, cset_link);
4293	cset = link->cset;
4294	}
4295	} while (!css_set_populated(cset));
4296
4297	it->cset_pos = l;
4298
4299	if (!list_empty(&cset->tasks))
4300	it->task_pos = cset->tasks.next;
4301	else
4302	it->task_pos = cset->mg_tasks.next;
4303
4304	it->tasks_head = &cset->tasks;
4305	it->mg_tasks_head = &cset->mg_tasks;
4306
4307	/*
4308	* We don't keep css_sets locked across iteration steps and thus
4309	* need to take steps to ensure that iteration can be resumed after
4310	* the lock is re-acquired. Iteration is performed at two levels -
4311	* css_sets and tasks in them.
4312	*
4313	* Once created, a css_set never leaves its cgroup lists, so a
4314	* pinned css_set is guaranteed to stay put and we can resume
4315	* iteration afterwards.
4316	*
4317	* Tasks may leave @cset across iteration steps. This is resolved
4318	* by registering each iterator with the css_set currently being
4319	* walked and making css_set_move_task() advance iterators whose
4320	* next task is leaving.
4321	*/
4322	if (it->cur_cset) {
4323	list_del(&it->iters_node);
4324	put_css_set_locked(it->cur_cset);
4325	}
4326	get_css_set(cset);
4327	it->cur_cset = cset;
4328	list_add(&it->iters_node, &cset->task_iters);
4329	}
4330
4331	static void css_task_iter_advance(struct css_task_iter *it)
4332	{
4333	struct list_head *l = it->task_pos;
4334
4335	lockdep_assert_held(&css_set_lock);
4336	WARN_ON_ONCE(!l);
4337
4338	/*
4339	* Advance iterator to find next entry. cset->tasks is consumed
4340	* first and then ->mg_tasks. After ->mg_tasks, we move onto the
4341	* next cset.
4342	*/
4343	l = l->next;
4344
4345	if (l == it->tasks_head)
4346	l = it->mg_tasks_head->next;
4347
4348	if (l == it->mg_tasks_head)
4349	css_task_iter_advance_css_set(it);
4350	else
4351	it->task_pos = l;
4352	}
4353
4354	/**
4355	* css_task_iter_start - initiate task iteration
4356	* @css: the css to walk tasks of
4357	* @it: the task iterator to use
4358	*
4359	* Initiate iteration through the tasks of @css. The caller can call
4360	* css_task_iter_next() to walk through the tasks until the function
4361	* returns NULL. On completion of iteration, css_task_iter_end() must be
4362	* called.
4363	*/
4364	void css_task_iter_start(struct cgroup_subsys_state *css,
4365	struct css_task_iter *it)
4366	{
4367	/* no one should try to iterate before mounting cgroups */
4368	WARN_ON_ONCE(!use_task_css_set_links);
4369
4370	memset(it, 0, sizeof(*it));
4371
4372	spin_lock_irq(&css_set_lock);
4373
4374	it->ss = css->ss;
4375
4376	if (it->ss)
4377	it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4378	else
4379	it->cset_pos = &css->cgroup->cset_links;
4380
4381	it->cset_head = it->cset_pos;
4382
4383	css_task_iter_advance_css_set(it);
4384
4385	spin_unlock_irq(&css_set_lock);
4386	}
4387
4388	/**
4389	* css_task_iter_next - return the next task for the iterator
4390	* @it: the task iterator being iterated
4391	*
4392	* The "next" function for task iteration. @it should have been
4393	* initialized via css_task_iter_start(). Returns NULL when the iteration
4394	* reaches the end.
4395	*/
4396	struct task_struct *css_task_iter_next(struct css_task_iter *it)
4397	{
4398	if (it->cur_task) {
4399	put_task_struct(it->cur_task);
4400	it->cur_task = NULL;
4401	}
4402
4403	spin_lock_irq(&css_set_lock);
4404
4405	if (it->task_pos) {
4406	it->cur_task = list_entry(it->task_pos, struct task_struct,
4407	cg_list);
4408	get_task_struct(it->cur_task);
4409	css_task_iter_advance(it);
4410	}
4411
4412	spin_unlock_irq(&css_set_lock);
4413
4414	return it->cur_task;
4415	}
4416
4417	/**
4418	* css_task_iter_end - finish task iteration
4419	* @it: the task iterator to finish
4420	*
4421	* Finish task iteration started by css_task_iter_start().
4422	*/
4423	void css_task_iter_end(struct css_task_iter *it)
4424	{
4425	if (it->cur_cset) {
4426	spin_lock_irq(&css_set_lock);
4427	list_del(&it->iters_node);
4428	put_css_set_locked(it->cur_cset);
4429	spin_unlock_irq(&css_set_lock);
4430	}
4431
4432	if (it->cur_task)
4433	put_task_struct(it->cur_task);
4434	}
4435
4436	/**
4437	* cgroup_trasnsfer_tasks - move tasks from one cgroup to another
4438	* @to: cgroup to which the tasks will be moved
4439	* @from: cgroup in which the tasks currently reside
4440	*
4441	* Locking rules between cgroup_post_fork() and the migration path
4442	* guarantee that, if a task is forking while being migrated, the new child
4443	* is guaranteed to be either visible in the source cgroup after the
4444	* parent's migration is complete or put into the target cgroup. No task
4445	* can slip out of migration through forking.
4446	*/
4447	int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
4448	{
4449	LIST_HEAD(preloaded_csets);
4450	struct cgrp_cset_link *link;
4451	struct css_task_iter it;
4452	struct task_struct *task;
4453	int ret;
4454
4455	if (!cgroup_may_migrate_to(to))
4456	return -EBUSY;
4457
4458	mutex_lock(&cgroup_mutex);
4459
4460	percpu_down_write(&cgroup_threadgroup_rwsem);
4461
4462	/* all tasks in @from are being moved, all csets are source */
4463	spin_lock_irq(&css_set_lock);
4464	list_for_each_entry(link, &from->cset_links, cset_link)
4465	cgroup_migrate_add_src(link->cset, to, &preloaded_csets);
4466	spin_unlock_irq(&css_set_lock);
4467
4468	ret = cgroup_migrate_prepare_dst(&preloaded_csets);
4469	if (ret)
4470	goto out_err;
4471
4472	/*
4473	* Migrate tasks one-by-one until @from is empty. This fails iff
4474	* ->can_attach() fails.
4475	*/
4476	do {
4477	css_task_iter_start(&from->self, &it);
4478
4479	do {
4480	task = css_task_iter_next(&it);
4481	} while (task && (task->flags & PF_EXITING));
4482
4483	if (task)
4484	get_task_struct(task);
4485	css_task_iter_end(&it);
4486
4487	if (task) {
4488	ret = cgroup_migrate(task, false, to->root);
4489	if (!ret)
4490	trace_cgroup_transfer_tasks(to, task, false);
4491	put_task_struct(task);
4492	}
4493	} while (task && !ret);
4494	out_err:
4495	cgroup_migrate_finish(&preloaded_csets);
4496	percpu_up_write(&cgroup_threadgroup_rwsem);
4497	mutex_unlock(&cgroup_mutex);
4498	return ret;
4499	}
4500
4501	/*
4502	* Stuff for reading the 'tasks'/'procs' files.
4503	*
4504	* Reading this file can return large amounts of data if a cgroup has
4505	* *lots* of attached tasks. So it may need several calls to read(),
4506	* but we cannot guarantee that the information we produce is correct
4507	* unless we produce it entirely atomically.
4508	*
4509	*/
4510
4511	/* which pidlist file are we talking about? */
4512	enum cgroup_filetype {
4513	CGROUP_FILE_PROCS,
4514	CGROUP_FILE_TASKS,
4515	};
4516
4517	/*
4518	* A pidlist is a list of pids that virtually represents the contents of one
4519	* of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4520	* a pair (one each for procs, tasks) for each pid namespace that's relevant
4521	* to the cgroup.
4522	*/
4523	struct cgroup_pidlist {
4524	/*
4525	* used to find which pidlist is wanted. doesn't change as long as
4526	* this particular list stays in the list.
4527	*/
4528	struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
4529	/* array of xids */
4530	pid_t *list;
4531	/* how many elements the above list has */
4532	int length;
4533	/* each of these stored in a list by its cgroup */
4534	struct list_head links;
4535	/* pointer to the cgroup we belong to, for list removal purposes */
4536	struct cgroup *owner;
4537	/* for delayed destruction */
4538	struct delayed_work destroy_dwork;
4539	};
4540
4541	/*
4542	* The following two functions "fix" the issue where there are more pids
4543	* than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4544	* TODO: replace with a kernel-wide solution to this problem
4545	*/
4546	#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4547	static void *pidlist_allocate(int count)
4548	{
4549	if (PIDLIST_TOO_LARGE(count))
4550	return vmalloc(count * sizeof(pid_t));
4551	else
4552	return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
4553	}
4554
4555	static void pidlist_free(void *p)
4556	{
4557	kvfree(p);
4558	}
4559
4560	/*
4561	* Used to destroy all pidlists lingering waiting for destroy timer. None
4562	* should be left afterwards.
4563	*/
4564	static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
4565	{
4566	struct cgroup_pidlist *l, *tmp_l;
4567
4568	mutex_lock(&cgrp->pidlist_mutex);
4569	list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
4570	mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
4571	mutex_unlock(&cgrp->pidlist_mutex);
4572
4573	flush_workqueue(cgroup_pidlist_destroy_wq);
4574	BUG_ON(!list_empty(&cgrp->pidlists));
4575	}
4576
4577	static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
4578	{
4579	struct delayed_work *dwork = to_delayed_work(work);
4580	struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
4581	destroy_dwork);
4582	struct cgroup_pidlist *tofree = NULL;
4583
4584	mutex_lock(&l->owner->pidlist_mutex);
4585
4586	/*
4587	* Destroy iff we didn't get queued again. The state won't change
4588	* as destroy_dwork can only be queued while locked.
4589	*/
4590	if (!delayed_work_pending(dwork)) {
4591	list_del(&l->links);
4592	pidlist_free(l->list);
4593	put_pid_ns(l->key.ns);
4594	tofree = l;
4595	}
4596
4597	mutex_unlock(&l->owner->pidlist_mutex);
4598	kfree(tofree);
4599	}
4600
4601	/*
4602	* pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
4603	* Returns the number of unique elements.
4604	*/
4605	static int pidlist_uniq(pid_t *list, int length)
4606	{
4607	int src, dest = 1;
4608
4609	/*
4610	* we presume the 0th element is unique, so i starts at 1. trivial
4611	* edge cases first; no work needs to be done for either
4612	*/
4613	if (length == 0 || length == 1)
4614	return length;
4615	/* src and dest walk down the list; dest counts unique elements */
4616	for (src = 1; src < length; src++) {
4617	/* find next unique element */
4618	while (list[src] == list[src-1]) {
4619	src++;
4620	if (src == length)
4621	goto after;
4622	}
4623	/* dest always points to where the next unique element goes */
4624	list[dest] = list[src];
4625	dest++;
4626	}
4627	after:
4628	return dest;
4629	}
4630
4631	/*
4632	* The two pid files - task and cgroup.procs - guaranteed that the result
4633	* is sorted, which forced this whole pidlist fiasco. As pid order is
4634	* different per namespace, each namespace needs differently sorted list,
4635	* making it impossible to use, for example, single rbtree of member tasks
4636	* sorted by task pointer. As pidlists can be fairly large, allocating one
4637	* per open file is dangerous, so cgroup had to implement shared pool of
4638	* pidlists keyed by cgroup and namespace.
4639	*
4640	* All this extra complexity was caused by the original implementation
4641	* committing to an entirely unnecessary property. In the long term, we
4642	* want to do away with it. Explicitly scramble sort order if on the
4643	* default hierarchy so that no such expectation exists in the new
4644	* interface.
4645	*
4646	* Scrambling is done by swapping every two consecutive bits, which is
4647	* non-identity one-to-one mapping which disturbs sort order sufficiently.
4648	*/
4649	static pid_t pid_fry(pid_t pid)
4650	{
4651	unsigned a = pid & 0x55555555;
4652	unsigned b = pid & 0xAAAAAAAA;
4653
4654	return (a << 1) | (b >> 1);
4655	}
4656
4657	static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
4658	{
4659	if (cgroup_on_dfl(cgrp))
4660	return pid_fry(pid);
4661	else
4662	return pid;
4663	}
4664
4665	static int cmppid(const void *a, const void *b)
4666	{
4667	return *(pid_t *)a - *(pid_t *)b;
4668	}
4669
4670	static int fried_cmppid(const void *a, const void *b)
4671	{
4672	return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
4673	}
4674
4675	static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
4676	enum cgroup_filetype type)
4677	{
4678	struct cgroup_pidlist *l;
4679	/* don't need task_nsproxy() if we're looking at ourself */
4680	struct pid_namespace *ns = task_active_pid_ns(current);
4681
4682	lockdep_assert_held(&cgrp->pidlist_mutex);
4683
4684	list_for_each_entry(l, &cgrp->pidlists, links)
4685	if (l->key.type == type && l->key.ns == ns)
4686	return l;
4687	return NULL;
4688	}
4689
4690	/*
4691	* find the appropriate pidlist for our purpose (given procs vs tasks)
4692	* returns with the lock on that pidlist already held, and takes care
4693	* of the use count, or returns NULL with no locks held if we're out of
4694	* memory.
4695	*/
4696	static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
4697	enum cgroup_filetype type)
4698	{
4699	struct cgroup_pidlist *l;
4700
4701	lockdep_assert_held(&cgrp->pidlist_mutex);
4702
4703	l = cgroup_pidlist_find(cgrp, type);
4704	if (l)
4705	return l;
4706
4707	/* entry not found; create a new one */
4708	l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
4709	if (!l)
4710	return l;
4711
4712	INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
4713	l->key.type = type;
4714	/* don't need task_nsproxy() if we're looking at ourself */
4715	l->key.ns = get_pid_ns(task_active_pid_ns(current));
4716	l->owner = cgrp;
4717	list_add(&l->links, &cgrp->pidlists);
4718	return l;
4719	}
4720
4721	/*
4722	* Load a cgroup's pidarray with either procs' tgids or tasks' pids
4723	*/
4724	static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
4725	struct cgroup_pidlist **lp)
4726	{
4727	pid_t *array;
4728	int length;
4729	int pid, n = 0; /* used for populating the array */
4730	struct css_task_iter it;
4731	struct task_struct *tsk;
4732	struct cgroup_pidlist *l;
4733
4734	lockdep_assert_held(&cgrp->pidlist_mutex);
4735
4736	/*
4737	* If cgroup gets more users after we read count, we won't have
4738	* enough space - tough. This race is indistinguishable to the
4739	* caller from the case that the additional cgroup users didn't
4740	* show up until sometime later on.
4741	*/
4742	length = cgroup_task_count(cgrp);
4743	array = pidlist_allocate(length);
4744	if (!array)
4745	return -ENOMEM;
4746	/* now, populate the array */
4747	css_task_iter_start(&cgrp->self, &it);
4748	while ((tsk = css_task_iter_next(&it))) {
4749	if (unlikely(n == length))
4750	break;
4751	/* get tgid or pid for procs or tasks file respectively */
4752	if (type == CGROUP_FILE_PROCS)
4753	pid = task_tgid_vnr(tsk);
4754	else
4755	pid = task_pid_vnr(tsk);
4756	if (pid > 0) /* make sure to only use valid results */
4757	array[n++] = pid;
4758	}
4759	css_task_iter_end(&it);
4760	length = n;
4761	/* now sort & (if procs) strip out duplicates */
4762	if (cgroup_on_dfl(cgrp))
4763	sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
4764	else
4765	sort(array, length, sizeof(pid_t), cmppid, NULL);
4766	if (type == CGROUP_FILE_PROCS)
4767	length = pidlist_uniq(array, length);
4768
4769	l = cgroup_pidlist_find_create(cgrp, type);
4770	if (!l) {
4771	pidlist_free(array);
4772	return -ENOMEM;
4773	}
4774
4775	/* store array, freeing old if necessary */
4776	pidlist_free(l->list);
4777	l->list = array;
4778	l->length = length;
4779	*lp = l;
4780	return 0;
4781	}
4782
4783	/**
4784	* cgroupstats_build - build and fill cgroupstats
4785	* @stats: cgroupstats to fill information into
4786	* @dentry: A dentry entry belonging to the cgroup for which stats have
4787	* been requested.
4788	*
4789	* Build and fill cgroupstats so that taskstats can export it to user
4790	* space.
4791	*/
4792	int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
4793	{
4794	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
4795	struct cgroup *cgrp;
4796	struct css_task_iter it;
4797	struct task_struct *tsk;
4798
4799	/* it should be kernfs_node belonging to cgroupfs and is a directory */
4800	if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
4801	kernfs_type(kn) != KERNFS_DIR)
4802	return -EINVAL;
4803
4804	mutex_lock(&cgroup_mutex);
4805
4806	/*
4807	* We aren't being called from kernfs and there's no guarantee on
4808	* @kn->priv's validity. For this and css_tryget_online_from_dir(),
4809	* @kn->priv is RCU safe. Let's do the RCU dancing.
4810	*/
4811	rcu_read_lock();
4812	cgrp = rcu_dereference(kn->priv);
4813	if (!cgrp || cgroup_is_dead(cgrp)) {
4814	rcu_read_unlock();
4815	mutex_unlock(&cgroup_mutex);
4816	return -ENOENT;
4817	}
4818	rcu_read_unlock();
4819
4820	css_task_iter_start(&cgrp->self, &it);
4821	while ((tsk = css_task_iter_next(&it))) {
4822	switch (tsk->state) {
4823	case TASK_RUNNING:
4824	stats->nr_running++;
4825	break;
4826	case TASK_INTERRUPTIBLE:
4827	stats->nr_sleeping++;
4828	break;
4829	case TASK_UNINTERRUPTIBLE:
4830	stats->nr_uninterruptible++;
4831	break;
4832	case TASK_STOPPED:
4833	stats->nr_stopped++;
4834	break;
4835	default:
4836	if (delayacct_is_task_waiting_on_io(tsk))
4837	stats->nr_io_wait++;
4838	break;
4839	}
4840	}
4841	css_task_iter_end(&it);
4842
4843	mutex_unlock(&cgroup_mutex);
4844	return 0;
4845	}
4846
4847
4848	/*
4849	* seq_file methods for the tasks/procs files. The seq_file position is the
4850	* next pid to display; the seq_file iterator is a pointer to the pid
4851	* in the cgroup->l->list array.
4852	*/
4853
4854	static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
4855	{
4856	/*
4857	* Initially we receive a position value that corresponds to
4858	* one more than the last pid shown (or 0 on the first call or
4859	* after a seek to the start). Use a binary-search to find the
4860	* next pid to display, if any
4861	*/
4862	struct kernfs_open_file *of = s->private;
4863	struct cgroup *cgrp = seq_css(s)->cgroup;
4864	struct cgroup_pidlist *l;
4865	enum cgroup_filetype type = seq_cft(s)->private;
4866	int index = 0, pid = *pos;
4867	int *iter, ret;
4868
4869	mutex_lock(&cgrp->pidlist_mutex);
4870
4871	/*
4872	* !NULL @of->priv indicates that this isn't the first start()
4873	* after open. If the matching pidlist is around, we can use that.
4874	* Look for it. Note that @of->priv can't be used directly. It
4875	* could already have been destroyed.
4876	*/
4877	if (of->priv)
4878	of->priv = cgroup_pidlist_find(cgrp, type);
4879
4880	/*
4881	* Either this is the first start() after open or the matching
4882	* pidlist has been destroyed inbetween. Create a new one.
4883	*/
4884	if (!of->priv) {
4885	ret = pidlist_array_load(cgrp, type,
4886	(struct cgroup_pidlist **)&of->priv);
4887	if (ret)
4888	return ERR_PTR(ret);
4889	}
4890	l = of->priv;
4891
4892	if (pid) {
4893	int end = l->length;
4894
4895	while (index < end) {
4896	int mid = (index + end) / 2;
4897	if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
4898	index = mid;
4899	break;
4900	} else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
4901	index = mid + 1;
4902	else
4903	end = mid;
4904	}
4905	}
4906	/* If we're off the end of the array, we're done */
4907	if (index >= l->length)
4908	return NULL;
4909	/* Update the abstract position to be the actual pid that we found */
4910	iter = l->list + index;
4911	*pos = cgroup_pid_fry(cgrp, *iter);
4912	return iter;
4913	}
4914
4915	static void cgroup_pidlist_stop(struct seq_file *s, void *v)
4916	{
4917	struct kernfs_open_file *of = s->private;
4918	struct cgroup_pidlist *l = of->priv;
4919
4920	if (l)
4921	mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
4922	CGROUP_PIDLIST_DESTROY_DELAY);
4923	mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
4924	}
4925
4926	static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
4927	{
4928	struct kernfs_open_file *of = s->private;
4929	struct cgroup_pidlist *l = of->priv;
4930	pid_t *p = v;
4931	pid_t *end = l->list + l->length;
4932	/*
4933	* Advance to the next pid in the array. If this goes off the
4934	* end, we're done
4935	*/
4936	p++;
4937	if (p >= end) {
4938	return NULL;
4939	} else {
4940	*pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
4941	return p;
4942	}
4943	}
4944
4945	static int cgroup_pidlist_show(struct seq_file *s, void *v)
4946	{
4947	seq_printf(s, "%d\n", *(int *)v);
4948
4949	return 0;
4950	}
4951
4952	static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
4953	struct cftype *cft)
4954	{
4955	return notify_on_release(css->cgroup);
4956	}
4957
4958	static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
4959	struct cftype *cft, u64 val)
4960	{
4961	if (val)
4962	set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
4963	else
4964	clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
4965	return 0;
4966	}
4967
4968	static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
4969	struct cftype *cft)
4970	{
4971	return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4972	}
4973
4974	static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
4975	struct cftype *cft, u64 val)
4976	{
4977	if (val)
4978	set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4979	else
4980	clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4981	return 0;
4982	}
4983
4984	/* cgroup core interface files for the default hierarchy */
4985	static struct cftype cgroup_dfl_base_files[] = {
4986	{
4987	.name = "cgroup.procs",
4988	.file_offset = offsetof(struct cgroup, procs_file),
4989	.seq_start = cgroup_pidlist_start,
4990	.seq_next = cgroup_pidlist_next,
4991	.seq_stop = cgroup_pidlist_stop,
4992	.seq_show = cgroup_pidlist_show,
4993	.private = CGROUP_FILE_PROCS,
4994	.write = cgroup_procs_write,
4995	},
4996	{
4997	.name = "cgroup.controllers",
4998	.seq_show = cgroup_controllers_show,
4999	},
5000	{
5001	.name = "cgroup.subtree_control",
5002	.seq_show = cgroup_subtree_control_show,
5003	.write = cgroup_subtree_control_write,
5004	},
5005	{
5006	.name = "cgroup.events",
5007	.flags = CFTYPE_NOT_ON_ROOT,
5008	.file_offset = offsetof(struct cgroup, events_file),
5009	.seq_show = cgroup_events_show,
5010	},
5011	#ifdef CONFIG_PSI
5012	{
5013	.name = "io.pressure",
5014	.flags = CFTYPE_NOT_ON_ROOT,
5015	.seq_show = cgroup_io_pressure_show,
5016	.write = cgroup_io_pressure_write,
5017	.poll = cgroup_pressure_poll,
5018	.release = cgroup_pressure_release,
5019	},
5020	{
5021	.name = "memory.pressure",
5022	.flags = CFTYPE_NOT_ON_ROOT,
5023	.seq_show = cgroup_memory_pressure_show,
5024	.write = cgroup_memory_pressure_write,
5025	.poll = cgroup_pressure_poll,
5026	.release = cgroup_pressure_release,
5027	},
5028	{
5029	.name = "cpu.pressure",
5030	.flags = CFTYPE_NOT_ON_ROOT,
5031	.seq_show = cgroup_cpu_pressure_show,
5032	.write = cgroup_cpu_pressure_write,
5033	.poll = cgroup_pressure_poll,
5034	.release = cgroup_pressure_release,
5035	},
5036	#endif /* CONFIG_PSI */
5037	{ } /* terminate */
5038	};
5039
5040	/* cgroup core interface files for the legacy hierarchies */
5041	static struct cftype cgroup_legacy_base_files[] = {
5042	{
5043	.name = "cgroup.procs",
5044	.seq_start = cgroup_pidlist_start,
5045	.seq_next = cgroup_pidlist_next,
5046	.seq_stop = cgroup_pidlist_stop,
5047	.seq_show = cgroup_pidlist_show,
5048	.private = CGROUP_FILE_PROCS,
5049	.write = cgroup_procs_write,
5050	},
5051	{
5052	.name = "cgroup.clone_children",
5053	.read_u64 = cgroup_clone_children_read,
5054	.write_u64 = cgroup_clone_children_write,
5055	},
5056	{
5057	.name = "cgroup.sane_behavior",
5058	.flags = CFTYPE_ONLY_ON_ROOT,
5059	.seq_show = cgroup_sane_behavior_show,
5060	},
5061	{
5062	.name = "tasks",
5063	.seq_start = cgroup_pidlist_start,
5064	.seq_next = cgroup_pidlist_next,
5065	.seq_stop = cgroup_pidlist_stop,
5066	.seq_show = cgroup_pidlist_show,
5067	.private = CGROUP_FILE_TASKS,
5068	.write = cgroup_tasks_write,
5069	},
5070	{
5071	.name = "notify_on_release",
5072	.read_u64 = cgroup_read_notify_on_release,
5073	.write_u64 = cgroup_write_notify_on_release,
5074	},
5075	{
5076	.name = "release_agent",
5077	.flags = CFTYPE_ONLY_ON_ROOT,
5078	.seq_show = cgroup_release_agent_show,
5079	.write = cgroup_release_agent_write,
5080	.max_write_len = PATH_MAX - 1,
5081	},
5082	{ } /* terminate */
5083	};
5084
5085	/*
5086	* css destruction is four-stage process.
5087	*
5088	* 1. Destruction starts. Killing of the percpu_ref is initiated.
5089	* Implemented in kill_css().
5090	*
5091	* 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5092	* and thus css_tryget_online() is guaranteed to fail, the css can be
5093	* offlined by invoking offline_css(). After offlining, the base ref is
5094	* put. Implemented in css_killed_work_fn().
5095	*
5096	* 3. When the percpu_ref reaches zero, the only possible remaining
5097	* accessors are inside RCU read sections. css_release() schedules the
5098	* RCU callback.
5099	*
5100	* 4. After the grace period, the css can be freed. Implemented in
5101	* css_free_work_fn().
5102	*
5103	* It is actually hairier because both step 2 and 4 require process context
5104	* and thus involve punting to css->destroy_work adding two additional
5105	* steps to the already complex sequence.
5106	*/
5107	static void css_free_work_fn(struct work_struct *work)
5108	{
5109	struct cgroup_subsys_state *css =
5110	container_of(work, struct cgroup_subsys_state, destroy_work);
5111	struct cgroup_subsys *ss = css->ss;
5112	struct cgroup *cgrp = css->cgroup;
5113
5114	percpu_ref_exit(&css->refcnt);
5115
5116	if (ss) {
5117	/* css free path */
5118	struct cgroup_subsys_state *parent = css->parent;
5119	int id = css->id;
5120
5121	ss->css_free(css);
5122	cgroup_idr_remove(&ss->css_idr, id);
5123	cgroup_put(cgrp);
5124
5125	if (parent)
5126	css_put(parent);
5127	} else {
5128	/* cgroup free path */
5129	atomic_dec(&cgrp->root->nr_cgrps);
5130	cgroup_pidlist_destroy_all(cgrp);
5131	cancel_work_sync(&cgrp->release_agent_work);
5132
5133	if (cgroup_parent(cgrp)) {
5134	/*
5135	* We get a ref to the parent, and put the ref when
5136	* this cgroup is being freed, so it's guaranteed
5137	* that the parent won't be destroyed before its
5138	* children.
5139	*/
5140	cgroup_put(cgroup_parent(cgrp));
5141	kernfs_put(cgrp->kn);
5142	if (cgroup_on_dfl(cgrp))
5143	psi_cgroup_free(cgrp);
5144	kfree(cgrp);
5145	} else {
5146	/*
5147	* This is root cgroup's refcnt reaching zero,
5148	* which indicates that the root should be
5149	* released.
5150	*/
5151	cgroup_destroy_root(cgrp->root);
5152	}
5153	}
5154	}
5155
5156	static void css_free_rcu_fn(struct rcu_head *rcu_head)
5157	{
5158	struct cgroup_subsys_state *css =
5159	container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
5160
5161	INIT_WORK(&css->destroy_work, css_free_work_fn);
5162	queue_work(cgroup_destroy_wq, &css->destroy_work);
5163	}
5164
5165	static void css_release_work_fn(struct work_struct *work)
5166	{
5167	struct cgroup_subsys_state *css =
5168	container_of(work, struct cgroup_subsys_state, destroy_work);
5169	struct cgroup_subsys *ss = css->ss;
5170	struct cgroup *cgrp = css->cgroup;
5171
5172	mutex_lock(&cgroup_mutex);
5173
5174	css->flags |= CSS_RELEASED;
5175	list_del_rcu(&css->sibling);
5176
5177	if (ss) {
5178	/* css release path */
5179	cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5180	if (ss->css_released)
5181	ss->css_released(css);
5182	} else {
5183	/* cgroup release path */
5184	trace_cgroup_release(cgrp);
5185
5186	cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
5187	cgrp->id = -1;
5188
5189	/*
5190	* There are two control paths which try to determine
5191	* cgroup from dentry without going through kernfs -
5192	* cgroupstats_build() and css_tryget_online_from_dir().
5193	* Those are supported by RCU protecting clearing of
5194	* cgrp->kn->priv backpointer.
5195	*/
5196	if (cgrp->kn)
5197	RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5198	NULL);
5199
5200	cgroup_bpf_put(cgrp);
5201	}
5202
5203	mutex_unlock(&cgroup_mutex);
5204
5205	call_rcu(&css->rcu_head, css_free_rcu_fn);
5206	}
5207
5208	static void css_release(struct percpu_ref *ref)
5209	{
5210	struct cgroup_subsys_state *css =
5211	container_of(ref, struct cgroup_subsys_state, refcnt);
5212
5213	INIT_WORK(&css->destroy_work, css_release_work_fn);
5214	queue_work(cgroup_destroy_wq, &css->destroy_work);
5215	}
5216
5217	static void init_and_link_css(struct cgroup_subsys_state *css,
5218	struct cgroup_subsys *ss, struct cgroup *cgrp)
5219	{
5220	lockdep_assert_held(&cgroup_mutex);
5221
5222	cgroup_get(cgrp);
5223
5224	memset(css, 0, sizeof(*css));
5225	css->cgroup = cgrp;
5226	css->ss = ss;
5227	css->id = -1;
5228	INIT_LIST_HEAD(&css->sibling);
5229	INIT_LIST_HEAD(&css->children);
5230	css->serial_nr = css_serial_nr_next++;
5231	atomic_set(&css->online_cnt, 0);
5232
5233	if (cgroup_parent(cgrp)) {
5234	css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5235	css_get(css->parent);
5236	}
5237
5238	BUG_ON(cgroup_css(cgrp, ss));
5239	}
5240
5241	/* invoke ->css_online() on a new CSS and mark it online if successful */
5242	static int online_css(struct cgroup_subsys_state *css)
5243	{
5244	struct cgroup_subsys *ss = css->ss;
5245	int ret = 0;
5246
5247	lockdep_assert_held(&cgroup_mutex);
5248
5249	if (ss->css_online)
5250	ret = ss->css_online(css);
5251	if (!ret) {
5252	css->flags |= CSS_ONLINE;
5253	rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5254
5255	atomic_inc(&css->online_cnt);
5256	if (css->parent)
5257	atomic_inc(&css->parent->online_cnt);
5258	}
5259	return ret;
5260	}
5261
5262	/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5263	static void offline_css(struct cgroup_subsys_state *css)
5264	{
5265	struct cgroup_subsys *ss = css->ss;
5266
5267	lockdep_assert_held(&cgroup_mutex);
5268
5269	if (!(css->flags & CSS_ONLINE))
5270	return;
5271
5272	if (ss->css_reset)
5273	ss->css_reset(css);
5274
5275	if (ss->css_offline)
5276	ss->css_offline(css);
5277
5278	css->flags &= ~CSS_ONLINE;
5279	RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5280
5281	wake_up_all(&css->cgroup->offline_waitq);
5282	}
5283
5284	/**
5285	* css_create - create a cgroup_subsys_state
5286	* @cgrp: the cgroup new css will be associated with
5287	* @ss: the subsys of new css
5288	*
5289	* Create a new css associated with @cgrp - @ss pair. On success, the new
5290	* css is online and installed in @cgrp. This function doesn't create the
5291	* interface files. Returns 0 on success, -errno on failure.
5292	*/
5293	static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5294	struct cgroup_subsys *ss)
5295	{
5296	struct cgroup *parent = cgroup_parent(cgrp);
5297	struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5298	struct cgroup_subsys_state *css;
5299	int err;
5300
5301	lockdep_assert_held(&cgroup_mutex);
5302
5303	css = ss->css_alloc(parent_css);
5304	if (!css)
5305	css = ERR_PTR(-ENOMEM);
5306	if (IS_ERR(css))
5307	return css;
5308
5309	init_and_link_css(css, ss, cgrp);
5310
5311	err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5312	if (err)
5313	goto err_free_css;
5314
5315	err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5316	if (err < 0)
5317	goto err_free_css;
5318	css->id = err;
5319
5320	/* @css is ready to be brought online now, make it visible */
5321	list_add_tail_rcu(&css->sibling, &parent_css->children);
5322	cgroup_idr_replace(&ss->css_idr, css, css->id);
5323
5324	err = online_css(css);
5325	if (err)
5326	goto err_list_del;
5327
5328	if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
5329	cgroup_parent(parent)) {
5330	pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5331	current->comm, current->pid, ss->name);
5332	if (!strcmp(ss->name, "memory"))
5333	pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5334	ss->warned_broken_hierarchy = true;
5335	}
5336
5337	return css;
5338
5339	err_list_del:
5340	list_del_rcu(&css->sibling);
5341	err_free_css:
5342	call_rcu(&css->rcu_head, css_free_rcu_fn);
5343	return ERR_PTR(err);
5344	}
5345
5346	/*
5347	* The returned cgroup is fully initialized including its control mask, but
5348	* it isn't associated with its kernfs_node and doesn't have the control
5349	* mask applied.
5350	*/
5351	static struct cgroup *cgroup_create(struct cgroup *parent)
5352	{
5353	struct cgroup_root *root = parent->root;
5354	struct cgroup *cgrp, *tcgrp;
5355	int level = parent->level + 1;
5356	int ret;
5357
5358	/* allocate the cgroup and its ID, 0 is reserved for the root */
5359	cgrp = kzalloc(sizeof(*cgrp) +
5360	sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
5361	if (!cgrp)
5362	return ERR_PTR(-ENOMEM);
5363
5364	ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5365	if (ret)
5366	goto out_free_cgrp;
5367
5368	/*
5369	* Temporarily set the pointer to NULL, so idr_find() won't return
5370	* a half-baked cgroup.
5371	*/
5372	cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
5373	if (cgrp->id < 0) {
5374	ret = -ENOMEM;
5375	goto out_cancel_ref;
5376	}
5377
5378	init_cgroup_housekeeping(cgrp);
5379
5380	cgrp->self.parent = &parent->self;
5381	cgrp->root = root;
5382	cgrp->level = level;
5383
5384	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp))
5385	cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
5386
5387	if (notify_on_release(parent))
5388	set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5389
5390	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5391	set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5392
5393	cgrp->self.serial_nr = css_serial_nr_next++;
5394
5395	/* allocation complete, commit to creation */
5396	list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5397	atomic_inc(&root->nr_cgrps);
5398	cgroup_get(parent);
5399
5400	/*
5401	* @cgrp is now fully operational. If something fails after this
5402	* point, it'll be released via the normal destruction path.
5403	*/
5404	cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
5405
5406	/*
5407	* On the default hierarchy, a child doesn't automatically inherit
5408	* subtree_control from the parent. Each is configured manually.
5409	*/
5410	if (!cgroup_on_dfl(cgrp))
5411	cgrp->subtree_control = cgroup_control(cgrp);
5412
5413	if (cgroup_on_dfl(cgrp)) {
5414	ret = psi_cgroup_alloc(cgrp);
5415	if (ret)
5416	goto out_idr_free;
5417	}
5418
5419	if (parent)
5420	cgroup_bpf_inherit(cgrp, parent);
5421
5422	cgroup_propagate_control(cgrp);
5423
5424	return cgrp;
5425
5426	out_idr_free:
5427	cgroup_idr_remove(&root->cgroup_idr, cgrp->id);
5428	out_cancel_ref:
5429	percpu_ref_exit(&cgrp->self.refcnt);
5430	out_free_cgrp:
5431	kfree(cgrp);
5432	return ERR_PTR(ret);
5433	}
5434
5435	static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
5436	umode_t mode)
5437	{
5438	struct cgroup *parent, *cgrp;
5439	struct kernfs_node *kn;
5440	int ret;
5441
5442	/* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5443	if (strchr(name, '\n'))
5444	return -EINVAL;
5445
5446	parent = cgroup_kn_lock_live(parent_kn, false);
5447	if (!parent)
5448	return -ENODEV;
5449
5450	cgrp = cgroup_create(parent);
5451	if (IS_ERR(cgrp)) {
5452	ret = PTR_ERR(cgrp);
5453	goto out_unlock;
5454	}
5455
5456	/* create the directory */
5457	kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5458	if (IS_ERR(kn)) {
5459	ret = PTR_ERR(kn);
5460	goto out_destroy;
5461	}
5462	cgrp->kn = kn;
5463
5464	/*
5465	* This extra ref will be put in cgroup_free_fn() and guarantees
5466	* that @cgrp->kn is always accessible.
5467	*/
5468	kernfs_get(kn);
5469
5470	ret = cgroup_kn_set_ugid(kn);
5471	if (ret)
5472	goto out_destroy;
5473
5474	ret = css_populate_dir(&cgrp->self);
5475	if (ret)
5476	goto out_destroy;
5477
5478	ret = cgroup_apply_control_enable(cgrp);
5479	if (ret)
5480	goto out_destroy;
5481
5482	trace_cgroup_mkdir(cgrp);
5483
5484	/* let's create and online css's */
5485	kernfs_activate(kn);
5486
5487	ret = 0;
5488	goto out_unlock;
5489
5490	out_destroy:
5491	cgroup_destroy_locked(cgrp);
5492	out_unlock:
5493	cgroup_kn_unlock(parent_kn);
5494	return ret;
5495	}
5496
5497	/*
5498	* This is called when the refcnt of a css is confirmed to be killed.
5499	* css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5500	* initate destruction and put the css ref from kill_css().
5501	*/
5502	static void css_killed_work_fn(struct work_struct *work)
5503	{
5504	struct cgroup_subsys_state *css =
5505	container_of(work, struct cgroup_subsys_state, destroy_work);
5506
5507	mutex_lock(&cgroup_mutex);
5508
5509	do {
5510	offline_css(css);
5511	css_put(css);
5512	/* @css can't go away while we're holding cgroup_mutex */
5513	css = css->parent;
5514	} while (css && atomic_dec_and_test(&css->online_cnt));
5515
5516	mutex_unlock(&cgroup_mutex);
5517	}
5518
5519	/* css kill confirmation processing requires process context, bounce */
5520	static void css_killed_ref_fn(struct percpu_ref *ref)
5521	{
5522	struct cgroup_subsys_state *css =
5523	container_of(ref, struct cgroup_subsys_state, refcnt);
5524
5525	if (atomic_dec_and_test(&css->online_cnt)) {
5526	INIT_WORK(&css->destroy_work, css_killed_work_fn);
5527	queue_work(cgroup_destroy_wq, &css->destroy_work);
5528	}
5529	}
5530
5531	/**
5532	* kill_css - destroy a css
5533	* @css: css to destroy
5534	*
5535	* This function initiates destruction of @css by removing cgroup interface
5536	* files and putting its base reference. ->css_offline() will be invoked
5537	* asynchronously once css_tryget_online() is guaranteed to fail and when
5538	* the reference count reaches zero, @css will be released.
5539	*/
5540	static void kill_css(struct cgroup_subsys_state *css)
5541	{
5542	lockdep_assert_held(&cgroup_mutex);
5543
5544	if (css->flags & CSS_DYING)
5545	return;
5546
5547	css->flags |= CSS_DYING;
5548
5549	/*
5550	* This must happen before css is disassociated with its cgroup.
5551	* See seq_css() for details.
5552	*/
5553	css_clear_dir(css);
5554
5555	/*
5556	* Killing would put the base ref, but we need to keep it alive
5557	* until after ->css_offline().
5558	*/
5559	css_get(css);
5560
5561	/*
5562	* cgroup core guarantees that, by the time ->css_offline() is
5563	* invoked, no new css reference will be given out via
5564	* css_tryget_online(). We can't simply call percpu_ref_kill() and
5565	* proceed to offlining css's because percpu_ref_kill() doesn't
5566	* guarantee that the ref is seen as killed on all CPUs on return.
5567	*
5568	* Use percpu_ref_kill_and_confirm() to get notifications as each
5569	* css is confirmed to be seen as killed on all CPUs.
5570	*/
5571	percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5572	}
5573
5574	/**
5575	* cgroup_destroy_locked - the first stage of cgroup destruction
5576	* @cgrp: cgroup to be destroyed
5577	*
5578	* css's make use of percpu refcnts whose killing latency shouldn't be
5579	* exposed to userland and are RCU protected. Also, cgroup core needs to
5580	* guarantee that css_tryget_online() won't succeed by the time
5581	* ->css_offline() is invoked. To satisfy all the requirements,
5582	* destruction is implemented in the following two steps.
5583	*
5584	* s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5585	* userland visible parts and start killing the percpu refcnts of
5586	* css's. Set up so that the next stage will be kicked off once all
5587	* the percpu refcnts are confirmed to be killed.
5588	*
5589	* s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5590	* rest of destruction. Once all cgroup references are gone, the
5591	* cgroup is RCU-freed.
5592	*
5593	* This function implements s1. After this step, @cgrp is gone as far as
5594	* the userland is concerned and a new cgroup with the same name may be
5595	* created. As cgroup doesn't care about the names internally, this
5596	* doesn't cause any problem.
5597	*/
5598	static int cgroup_destroy_locked(struct cgroup *cgrp)
5599	__releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5600	{
5601	struct cgroup_subsys_state *css;
5602	struct cgrp_cset_link *link;
5603	int ssid;
5604
5605	lockdep_assert_held(&cgroup_mutex);
5606
5607	/*
5608	* Only migration can raise populated from zero and we're already
5609	* holding cgroup_mutex.
5610	*/
5611	if (cgroup_is_populated(cgrp))
5612	return -EBUSY;
5613
5614	/*
5615	* Make sure there's no live children. We can't test emptiness of
5616	* ->self.children as dead children linger on it while being
5617	* drained; otherwise, "rmdir parent/child parent" may fail.
5618	*/
5619	if (css_has_online_children(&cgrp->self))
5620	return -EBUSY;
5621
5622	/*
5623	* Mark @cgrp and the associated csets dead. The former prevents
5624	* further task migration and child creation by disabling
5625	* cgroup_lock_live_group(). The latter makes the csets ignored by
5626	* the migration path.
5627	*/
5628	cgrp->self.flags &= ~CSS_ONLINE;
5629
5630	spin_lock_irq(&css_set_lock);
5631	list_for_each_entry(link, &cgrp->cset_links, cset_link)
5632	link->cset->dead = true;
5633	spin_unlock_irq(&css_set_lock);
5634
5635	/* initiate massacre of all css's */
5636	for_each_css(css, ssid, cgrp)
5637	kill_css(css);
5638
5639	/*
5640	* Remove @cgrp directory along with the base files. @cgrp has an
5641	* extra ref on its kn.
5642	*/
5643	kernfs_remove(cgrp->kn);
5644
5645	check_for_release(cgroup_parent(cgrp));
5646
5647	/* put the base reference */
5648	percpu_ref_kill(&cgrp->self.refcnt);
5649
5650	return 0;
5651	};
5652
5653	static int cgroup_rmdir(struct kernfs_node *kn)
5654	{
5655	struct cgroup *cgrp;
5656	int ret = 0;
5657
5658	cgrp = cgroup_kn_lock_live(kn, false);
5659	if (!cgrp)
5660	return 0;
5661
5662	ret = cgroup_destroy_locked(cgrp);
5663
5664	if (!ret)
5665	trace_cgroup_rmdir(cgrp);
5666
5667	cgroup_kn_unlock(kn);
5668	return ret;
5669	}
5670
5671	static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5672	.remount_fs = cgroup_remount,
5673	.show_options = cgroup_show_options,
5674	.mkdir = cgroup_mkdir,
5675	.rmdir = cgroup_rmdir,
5676	.rename = cgroup_rename,
5677	.show_path = cgroup_show_path,
5678	};
5679
5680	static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5681	{
5682	struct cgroup_subsys_state *css;
5683
5684	pr_debug("Initializing cgroup subsys %s\n", ss->name);
5685
5686	mutex_lock(&cgroup_mutex);
5687
5688	idr_init(&ss->css_idr);
5689	INIT_LIST_HEAD(&ss->cfts);
5690
5691	/* Create the root cgroup state for this subsystem */
5692	ss->root = &cgrp_dfl_root;
5693	css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5694	/* We don't handle early failures gracefully */
5695	BUG_ON(IS_ERR(css));
5696	init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5697
5698	/*
5699	* Root csses are never destroyed and we can't initialize
5700	* percpu_ref during early init. Disable refcnting.
5701	*/
5702	css->flags |= CSS_NO_REF;
5703
5704	if (early) {
5705	/* allocation can't be done safely during early init */
5706	css->id = 1;
5707	} else {
5708	css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5709	BUG_ON(css->id < 0);
5710	}
5711
5712	/* Update the init_css_set to contain a subsys
5713	* pointer to this state - since the subsystem is
5714	* newly registered, all tasks and hence the
5715	* init_css_set is in the subsystem's root cgroup. */
5716	init_css_set.subsys[ss->id] = css;
5717
5718	have_fork_callback |= (bool)ss->fork << ss->id;
5719	have_exit_callback |= (bool)ss->exit << ss->id;
5720	have_free_callback |= (bool)ss->free << ss->id;
5721	have_canfork_callback |= (bool)ss->can_fork << ss->id;
5722
5723	/* At system boot, before all subsystems have been
5724	* registered, no tasks have been forked, so we don't
5725	* need to invoke fork callbacks here. */
5726	BUG_ON(!list_empty(&init_task.tasks));
5727
5728	BUG_ON(online_css(css));
5729
5730	mutex_unlock(&cgroup_mutex);
5731	}
5732
5733	/**
5734	* cgroup_init_early - cgroup initialization at system boot
5735	*
5736	* Initialize cgroups at system boot, and initialize any
5737	* subsystems that request early init.
5738	*/
5739	int __init cgroup_init_early(void)
5740	{
5741	static struct cgroup_sb_opts __initdata opts;
5742	struct cgroup_subsys *ss;
5743	int i;
5744
5745	init_cgroup_root(&cgrp_dfl_root, &opts);
5746	cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5747
5748	RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5749
5750	for_each_subsys(ss, i) {
5751	WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5752	"invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5753	i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5754	ss->id, ss->name);
5755	WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5756	"cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5757
5758	ss->id = i;
5759	ss->name = cgroup_subsys_name[i];
5760	if (!ss->legacy_name)
5761	ss->legacy_name = cgroup_subsys_name[i];
5762
5763	if (ss->early_init)
5764	cgroup_init_subsys(ss, true);
5765	}
5766	return 0;
5767	}
5768
5769	static u16 cgroup_disable_mask __initdata;
5770
5771	/**
5772	* cgroup_init - cgroup initialization
5773	*
5774	* Register cgroup filesystem and /proc file, and initialize
5775	* any subsystems that didn't request early init.
5776	*/
5777	int __init cgroup_init(void)
5778	{
5779	struct cgroup_subsys *ss;
5780	int ssid;
5781
5782	BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5783	BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
5784	BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files));
5785	BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files));
5786
5787	/*
5788	* The latency of the synchronize_sched() is too high for cgroups,
5789	* avoid it at the cost of forcing all readers into the slow path.
5790	*/
5791	rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5792
5793	get_user_ns(init_cgroup_ns.user_ns);
5794
5795	mutex_lock(&cgroup_mutex);
5796
5797	/*
5798	* Add init_css_set to the hash table so that dfl_root can link to
5799	* it during init.
5800	*/
5801	hash_add(css_set_table, &init_css_set.hlist,
5802	css_set_hash(init_css_set.subsys));
5803
5804	BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5805
5806	mutex_unlock(&cgroup_mutex);
5807
5808	for_each_subsys(ss, ssid) {
5809	if (ss->early_init) {
5810	struct cgroup_subsys_state *css =
5811	init_css_set.subsys[ss->id];
5812
5813	css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5814	GFP_KERNEL);
5815	BUG_ON(css->id < 0);
5816	} else {
5817	cgroup_init_subsys(ss, false);
5818	}
5819
5820	list_add_tail(&init_css_set.e_cset_node[ssid],
5821	&cgrp_dfl_root.cgrp.e_csets[ssid]);
5822
5823	/*
5824	* Setting dfl_root subsys_mask needs to consider the
5825	* disabled flag and cftype registration needs kmalloc,
5826	* both of which aren't available during early_init.
5827	*/
5828	if (cgroup_disable_mask & (1 << ssid)) {
5829	static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5830	printk(KERN_INFO "Disabling %s control group subsystem\n",
5831	ss->name);
5832	continue;
5833	}
5834
5835	if (cgroup_ssid_no_v1(ssid))
5836	printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5837	ss->name);
5838
5839	cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5840
5841	if (ss->implicit_on_dfl)
5842	cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5843	else if (!ss->dfl_cftypes)
5844	cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5845
5846	if (ss->dfl_cftypes == ss->legacy_cftypes) {
5847	WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5848	} else {
5849	WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5850	WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5851	}
5852
5853	if (ss->bind)
5854	ss->bind(init_css_set.subsys[ssid]);
5855
5856	mutex_lock(&cgroup_mutex);
5857	css_populate_dir(init_css_set.subsys[ssid]);
5858	mutex_unlock(&cgroup_mutex);
5859	}
5860
5861	/* init_css_set.subsys[] has been updated, re-hash */
5862	hash_del(&init_css_set.hlist);
5863	hash_add(css_set_table, &init_css_set.hlist,
5864	css_set_hash(init_css_set.subsys));
5865
5866	WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5867	WARN_ON(register_filesystem(&cgroup_fs_type));
5868	WARN_ON(register_filesystem(&cgroup2_fs_type));
5869	WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
5870
5871	return 0;
5872	}
5873
5874	static int __init cgroup_wq_init(void)
5875	{
5876	/*
5877	* There isn't much point in executing destruction path in
5878	* parallel. Good chunk is serialized with cgroup_mutex anyway.
5879	* Use 1 for @max_active.
5880	*
5881	* We would prefer to do this in cgroup_init() above, but that
5882	* is called before init_workqueues(): so leave this until after.
5883	*/
5884	cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5885	BUG_ON(!cgroup_destroy_wq);
5886
5887	/*
5888	* Used to destroy pidlists and separate to serve as flush domain.
5889	* Cap @max_active to 1 too.
5890	*/
5891	cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
5892	0, 1);
5893	BUG_ON(!cgroup_pidlist_destroy_wq);
5894
5895	return 0;
5896	}
5897	core_initcall(cgroup_wq_init);
5898
5899	/*
5900	* proc_cgroup_show()
5901	* - Print task's cgroup paths into seq_file, one line for each hierarchy
5902	* - Used for /proc/<pid>/cgroup.
5903	*/
5904	int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5905	struct pid *pid, struct task_struct *tsk)
5906	{
5907	char *buf;
5908	int retval;
5909	struct cgroup_root *root;
5910
5911	retval = -ENOMEM;
5912	buf = kmalloc(PATH_MAX, GFP_KERNEL);
5913	if (!buf)
5914	goto out;
5915
5916	mutex_lock(&cgroup_mutex);
5917	spin_lock_irq(&css_set_lock);
5918
5919	for_each_root(root) {
5920	struct cgroup_subsys *ss;
5921	struct cgroup *cgrp;
5922	int ssid, count = 0;
5923
5924	if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5925	continue;
5926
5927	seq_printf(m, "%d:", root->hierarchy_id);
5928	if (root != &cgrp_dfl_root)
5929	for_each_subsys(ss, ssid)
5930	if (root->subsys_mask & (1 << ssid))
5931	seq_printf(m, "%s%s", count++ ? "," : "",
5932	ss->legacy_name);
5933	if (strlen(root->name))
5934	seq_printf(m, "%sname=%s", count ? "," : "",
5935	root->name);
5936	seq_putc(m, ':');
5937
5938	cgrp = task_cgroup_from_root(tsk, root);
5939
5940	/*
5941	* On traditional hierarchies, all zombie tasks show up as
5942	* belonging to the root cgroup. On the default hierarchy,
5943	* while a zombie doesn't show up in "cgroup.procs" and
5944	* thus can't be migrated, its /proc/PID/cgroup keeps
5945	* reporting the cgroup it belonged to before exiting. If
5946	* the cgroup is removed before the zombie is reaped,
5947	* " (deleted)" is appended to the cgroup path.
5948	*/
5949	if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5950	retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5951	current->nsproxy->cgroup_ns);
5952	if (retval >= PATH_MAX)
5953	retval = -ENAMETOOLONG;
5954	if (retval < 0)
5955	goto out_unlock;
5956
5957	seq_puts(m, buf);
5958	} else {
5959	seq_puts(m, "/");
5960	}
5961
5962	if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5963	seq_puts(m, " (deleted)\n");
5964	else
5965	seq_putc(m, '\n');
5966	}
5967
5968	retval = 0;
5969	out_unlock:
5970	spin_unlock_irq(&css_set_lock);
5971	mutex_unlock(&cgroup_mutex);
5972	kfree(buf);
5973	out:
5974	return retval;
5975	}
5976
5977	/* Display information about each subsystem and each hierarchy */
5978	static int proc_cgroupstats_show(struct seq_file *m, void *v)
5979	{
5980	struct cgroup_subsys *ss;
5981	int i;
5982
5983	seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5984	/*
5985	* ideally we don't want subsystems moving around while we do this.
5986	* cgroup_mutex is also necessary to guarantee an atomic snapshot of
5987	* subsys/hierarchy state.
5988	*/
5989	mutex_lock(&cgroup_mutex);
5990
5991	for_each_subsys(ss, i)
5992	seq_printf(m, "%s\t%d\t%d\t%d\n",
5993	ss->legacy_name, ss->root->hierarchy_id,
5994	atomic_read(&ss->root->nr_cgrps),
5995	cgroup_ssid_enabled(i));
5996
5997	mutex_unlock(&cgroup_mutex);
5998	return 0;
5999	}
6000
6001	static int cgroupstats_open(struct inode *inode, struct file *file)
6002	{
6003	return single_open(file, proc_cgroupstats_show, NULL);
6004	}
6005
6006	static const struct file_operations proc_cgroupstats_operations = {
6007	.open = cgroupstats_open,
6008	.read = seq_read,
6009	.llseek = seq_lseek,
6010	.release = single_release,
6011	};
6012
6013	/**
6014	* cgroup_fork - initialize cgroup related fields during copy_process()
6015	* @child: pointer to task_struct of forking parent process.
6016	*
6017	* A task is associated with the init_css_set until cgroup_post_fork()
6018	* attaches it to the parent's css_set. Empty cg_list indicates that
6019	* @child isn't holding reference to its css_set.
6020	*/
6021	void cgroup_fork(struct task_struct *child)
6022	{
6023	RCU_INIT_POINTER(child->cgroups, &init_css_set);
6024	INIT_LIST_HEAD(&child->cg_list);
6025	}
6026
6027	/**
6028	* cgroup_can_fork - called on a new task before the process is exposed
6029	* @child: the task in question.
6030	*
6031	* This calls the subsystem can_fork() callbacks. If the can_fork() callback
6032	* returns an error, the fork aborts with that error code. This allows for
6033	* a cgroup subsystem to conditionally allow or deny new forks.
6034	*/
6035	int cgroup_can_fork(struct task_struct *child)
6036	{
6037	struct cgroup_subsys *ss;
6038	int i, j, ret;
6039
6040	do_each_subsys_mask(ss, i, have_canfork_callback) {
6041	ret = ss->can_fork(child);
6042	if (ret)
6043	goto out_revert;
6044	} while_each_subsys_mask();
6045
6046	return 0;
6047
6048	out_revert:
6049	for_each_subsys(ss, j) {
6050	if (j >= i)
6051	break;
6052	if (ss->cancel_fork)
6053	ss->cancel_fork(child);
6054	}
6055
6056	return ret;
6057	}
6058
6059	/**
6060	* cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6061	* @child: the task in question
6062	*
6063	* This calls the cancel_fork() callbacks if a fork failed *after*
6064	* cgroup_can_fork() succeded.
6065	*/
6066	void cgroup_cancel_fork(struct task_struct *child)
6067	{
6068	struct cgroup_subsys *ss;
6069	int i;
6070
6071	for_each_subsys(ss, i)
6072	if (ss->cancel_fork)
6073	ss->cancel_fork(child);
6074	}
6075
6076	/**
6077	* cgroup_post_fork - called on a new task after adding it to the task list
6078	* @child: the task in question
6079	*
6080	* Adds the task to the list running through its css_set if necessary and
6081	* call the subsystem fork() callbacks. Has to be after the task is
6082	* visible on the task list in case we race with the first call to
6083	* cgroup_task_iter_start() - to guarantee that the new task ends up on its
6084	* list.
6085	*/
6086	void cgroup_post_fork(struct task_struct *child)
6087	{
6088	struct cgroup_subsys *ss;
6089	int i;
6090
6091	/*
6092	* This may race against cgroup_enable_task_cg_lists(). As that
6093	* function sets use_task_css_set_links before grabbing
6094	* tasklist_lock and we just went through tasklist_lock to add
6095	* @child, it's guaranteed that either we see the set
6096	* use_task_css_set_links or cgroup_enable_task_cg_lists() sees
6097	* @child during its iteration.
6098	*
6099	* If we won the race, @child is associated with %current's
6100	* css_set. Grabbing css_set_lock guarantees both that the
6101	* association is stable, and, on completion of the parent's
6102	* migration, @child is visible in the source of migration or
6103	* already in the destination cgroup. This guarantee is necessary
6104	* when implementing operations which need to migrate all tasks of
6105	* a cgroup to another.
6106	*
6107	* Note that if we lose to cgroup_enable_task_cg_lists(), @child
6108	* will remain in init_css_set. This is safe because all tasks are
6109	* in the init_css_set before cg_links is enabled and there's no
6110	* operation which transfers all tasks out of init_css_set.
6111	*/
6112	if (use_task_css_set_links) {
6113	struct css_set *cset;
6114
6115	spin_lock_irq(&css_set_lock);
6116	cset = task_css_set(current);
6117	if (list_empty(&child->cg_list)) {
6118	get_css_set(cset);
6119	css_set_move_task(child, NULL, cset, false);
6120	}
6121	spin_unlock_irq(&css_set_lock);
6122	}
6123
6124	/*
6125	* Call ss->fork(). This must happen after @child is linked on
6126	* css_set; otherwise, @child might change state between ->fork()
6127	* and addition to css_set.
6128	*/
6129	do_each_subsys_mask(ss, i, have_fork_callback) {
6130	ss->fork(child);
6131	} while_each_subsys_mask();
6132	}
6133
6134	/**
6135	* cgroup_exit - detach cgroup from exiting task
6136	* @tsk: pointer to task_struct of exiting process
6137	*
6138	* Description: Detach cgroup from @tsk and release it.
6139	*
6140	* Note that cgroups marked notify_on_release force every task in
6141	* them to take the global cgroup_mutex mutex when exiting.
6142	* This could impact scaling on very large systems. Be reluctant to
6143	* use notify_on_release cgroups where very high task exit scaling
6144	* is required on large systems.
6145	*
6146	* We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
6147	* call cgroup_exit() while the task is still competent to handle
6148	* notify_on_release(), then leave the task attached to the root cgroup in
6149	* each hierarchy for the remainder of its exit. No need to bother with
6150	* init_css_set refcnting. init_css_set never goes away and we can't race
6151	* with migration path - PF_EXITING is visible to migration path.
6152	*/
6153	void cgroup_exit(struct task_struct *tsk)
6154	{
6155	struct cgroup_subsys *ss;
6156	struct css_set *cset;
6157	int i;
6158
6159	/*
6160	* Unlink from @tsk from its css_set. As migration path can't race
6161	* with us, we can check css_set and cg_list without synchronization.
6162	*/
6163	cset = task_css_set(tsk);
6164
6165	if (!list_empty(&tsk->cg_list)) {
6166	spin_lock_irq(&css_set_lock);
6167	css_set_move_task(tsk, cset, NULL, false);
6168	spin_unlock_irq(&css_set_lock);
6169	} else {
6170	get_css_set(cset);
6171	}
6172
6173	/* see cgroup_post_fork() for details */
6174	do_each_subsys_mask(ss, i, have_exit_callback) {
6175	ss->exit(tsk);
6176	} while_each_subsys_mask();
6177	}
6178
6179	void cgroup_free(struct task_struct *task)
6180	{
6181	struct css_set *cset = task_css_set(task);
6182	struct cgroup_subsys *ss;
6183	int ssid;
6184
6185	do_each_subsys_mask(ss, ssid, have_free_callback) {
6186	ss->free(task);
6187	} while_each_subsys_mask();
6188
6189	put_css_set(cset);
6190	}
6191
6192	static void check_for_release(struct cgroup *cgrp)
6193	{
6194	if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
6195	!css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
6196	schedule_work(&cgrp->release_agent_work);
6197	}
6198
6199	/*
6200	* Notify userspace when a cgroup is released, by running the
6201	* configured release agent with the name of the cgroup (path
6202	* relative to the root of cgroup file system) as the argument.
6203	*
6204	* Most likely, this user command will try to rmdir this cgroup.
6205	*
6206	* This races with the possibility that some other task will be
6207	* attached to this cgroup before it is removed, or that some other
6208	* user task will 'mkdir' a child cgroup of this cgroup. That's ok.
6209	* The presumed 'rmdir' will fail quietly if this cgroup is no longer
6210	* unused, and this cgroup will be reprieved from its death sentence,
6211	* to continue to serve a useful existence. Next time it's released,
6212	* we will get notified again, if it still has 'notify_on_release' set.
6213	*
6214	* The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
6215	* means only wait until the task is successfully execve()'d. The
6216	* separate release agent task is forked by call_usermodehelper(),
6217	* then control in this thread returns here, without waiting for the
6218	* release agent task. We don't bother to wait because the caller of
6219	* this routine has no use for the exit status of the release agent
6220	* task, so no sense holding our caller up for that.
6221	*/
6222	static void cgroup_release_agent(struct work_struct *work)
6223	{
6224	struct cgroup *cgrp =
6225	container_of(work, struct cgroup, release_agent_work);
6226	char *pathbuf = NULL, *agentbuf = NULL;
6227	char *argv[3], *envp[3];
6228	int ret;
6229
6230	mutex_lock(&cgroup_mutex);
6231
6232	pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
6233	agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
6234	if (!pathbuf || !agentbuf)
6235	goto out;
6236
6237	spin_lock_irq(&css_set_lock);
6238	ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
6239	spin_unlock_irq(&css_set_lock);
6240	if (ret < 0 || ret >= PATH_MAX)
6241	goto out;
6242
6243	argv[0] = agentbuf;
6244	argv[1] = pathbuf;
6245	argv[2] = NULL;
6246
6247	/* minimal command environment */
6248	envp[0] = "HOME=/";
6249	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
6250	envp[2] = NULL;
6251
6252	mutex_unlock(&cgroup_mutex);
6253	call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
6254	goto out_free;
6255	out:
6256	mutex_unlock(&cgroup_mutex);
6257	out_free:
6258	kfree(agentbuf);
6259	kfree(pathbuf);
6260	}
6261
6262	static int __init cgroup_disable(char *str)
6263	{
6264	struct cgroup_subsys *ss;
6265	char *token;
6266	int i;
6267
6268	while ((token = strsep(&str, ",")) != NULL) {
6269	if (!*token)
6270	continue;
6271
6272	for_each_subsys(ss, i) {
6273	if (strcmp(token, ss->name) &&
6274	strcmp(token, ss->legacy_name))
6275	continue;
6276	cgroup_disable_mask |= 1 << i;
6277	}
6278	}
6279	return 1;
6280	}
6281	__setup("cgroup_disable=", cgroup_disable);
6282
6283	static int __init cgroup_no_v1(char *str)
6284	{
6285	struct cgroup_subsys *ss;
6286	char *token;
6287	int i;
6288
6289	while ((token = strsep(&str, ",")) != NULL) {
6290	if (!*token)
6291	continue;
6292
6293	if (!strcmp(token, "all")) {
6294	cgroup_no_v1_mask = U16_MAX;
6295	break;
6296	}
6297
6298	for_each_subsys(ss, i) {
6299	if (strcmp(token, ss->name) &&
6300	strcmp(token, ss->legacy_name))
6301	continue;
6302
6303	cgroup_no_v1_mask |= 1 << i;
6304	}
6305	}
6306	return 1;
6307	}
6308	__setup("cgroup_no_v1=", cgroup_no_v1);
6309
6310	/**
6311	* css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6312	* @dentry: directory dentry of interest
6313	* @ss: subsystem of interest
6314	*
6315	* If @dentry is a directory for a cgroup which has @ss enabled on it, try
6316	* to get the corresponding css and return it. If such css doesn't exist
6317	* or can't be pinned, an ERR_PTR value is returned.
6318	*/
6319	struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6320	struct cgroup_subsys *ss)
6321	{
6322	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6323	struct file_system_type *s_type = dentry->d_sb->s_type;
6324	struct cgroup_subsys_state *css = NULL;
6325	struct cgroup *cgrp;
6326
6327	/* is @dentry a cgroup dir? */
6328	if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6329	!kn || kernfs_type(kn) != KERNFS_DIR)
6330	return ERR_PTR(-EBADF);
6331
6332	rcu_read_lock();
6333
6334	/*
6335	* This path doesn't originate from kernfs and @kn could already
6336	* have been or be removed at any point. @kn->priv is RCU
6337	* protected for this access. See css_release_work_fn() for details.
6338	*/
6339	cgrp = rcu_dereference(kn->priv);
6340	if (cgrp)
6341	css = cgroup_css(cgrp, ss);
6342
6343	if (!css || !css_tryget_online(css))
6344	css = ERR_PTR(-ENOENT);
6345
6346	rcu_read_unlock();
6347	return css;
6348	}
6349
6350	/**
6351	* css_from_id - lookup css by id
6352	* @id: the cgroup id
6353	* @ss: cgroup subsys to be looked into
6354	*
6355	* Returns the css if there's valid one with @id, otherwise returns NULL.
6356	* Should be called under rcu_read_lock().
6357	*/
6358	struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6359	{
6360	WARN_ON_ONCE(!rcu_read_lock_held());
6361	return idr_find(&ss->css_idr, id);
6362	}
6363
6364	/**
6365	* cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6366	* @path: path on the default hierarchy
6367	*
6368	* Find the cgroup at @path on the default hierarchy, increment its
6369	* reference count and return it. Returns pointer to the found cgroup on
6370	* success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6371	* if @path points to a non-directory.
6372	*/
6373	struct cgroup *cgroup_get_from_path(const char *path)
6374	{
6375	struct kernfs_node *kn;
6376	struct cgroup *cgrp;
6377
6378	mutex_lock(&cgroup_mutex);
6379
6380	kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6381	if (kn) {
6382	if (kernfs_type(kn) == KERNFS_DIR) {
6383	cgrp = kn->priv;
6384	cgroup_get(cgrp);
6385	} else {
6386	cgrp = ERR_PTR(-ENOTDIR);
6387	}
6388	kernfs_put(kn);
6389	} else {
6390	cgrp = ERR_PTR(-ENOENT);
6391	}
6392
6393	mutex_unlock(&cgroup_mutex);
6394	return cgrp;
6395	}
6396	EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6397
6398	/**
6399	* cgroup_get_from_fd - get a cgroup pointer from a fd
6400	* @fd: fd obtained by open(cgroup2_dir)
6401	*
6402	* Find the cgroup from a fd which should be obtained
6403	* by opening a cgroup directory. Returns a pointer to the
6404	* cgroup on success. ERR_PTR is returned if the cgroup
6405	* cannot be found.
6406	*/
6407	struct cgroup *cgroup_get_from_fd(int fd)
6408	{
6409	struct cgroup_subsys_state *css;
6410	struct cgroup *cgrp;
6411	struct file *f;
6412
6413	f = fget_raw(fd);
6414	if (!f)
6415	return ERR_PTR(-EBADF);
6416
6417	css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6418	fput(f);
6419	if (IS_ERR(css))
6420	return ERR_CAST(css);
6421
6422	cgrp = css->cgroup;
6423	if (!cgroup_on_dfl(cgrp)) {
6424	cgroup_put(cgrp);
6425	return ERR_PTR(-EBADF);
6426	}
6427
6428	return cgrp;
6429	}
6430	EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6431
6432	/*
6433	* sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6434	* definition in cgroup-defs.h.
6435	*/
6436	#ifdef CONFIG_SOCK_CGROUP_DATA
6437
6438	#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6439
6440	DEFINE_SPINLOCK(cgroup_sk_update_lock);
6441	static bool cgroup_sk_alloc_disabled __read_mostly;
6442
6443	void cgroup_sk_alloc_disable(void)
6444	{
6445	if (cgroup_sk_alloc_disabled)
6446	return;
6447	pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6448	cgroup_sk_alloc_disabled = true;
6449	}
6450
6451	#else
6452
6453	#define cgroup_sk_alloc_disabled false
6454
6455	#endif
6456
6457	void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6458	{
6459	if (cgroup_sk_alloc_disabled)
6460	return;
6461
6462	/* Socket clone path */
6463	if (skcd->val) {
6464	cgroup_get(sock_cgroup_ptr(skcd));
6465	return;
6466	}
6467
6468	rcu_read_lock();
6469
6470	while (true) {
6471	struct css_set *cset;
6472
6473	cset = task_css_set(current);
6474	if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6475	skcd->val = (unsigned long)cset->dfl_cgrp;
6476	break;
6477	}
6478	cpu_relax();
6479	}
6480
6481	rcu_read_unlock();
6482	}
6483
6484	void cgroup_sk_free(struct sock_cgroup_data *skcd)
6485	{
6486	cgroup_put(sock_cgroup_ptr(skcd));
6487	}
6488
6489	#endif /* CONFIG_SOCK_CGROUP_DATA */
6490
6491	/* cgroup namespaces */
6492
6493	static struct ucounts *inc_cgroup_namespaces(struct user_namespace *ns)
6494	{
6495	return inc_ucount(ns, current_euid(), UCOUNT_CGROUP_NAMESPACES);
6496	}
6497
6498	static void dec_cgroup_namespaces(struct ucounts *ucounts)
6499	{
6500	dec_ucount(ucounts, UCOUNT_CGROUP_NAMESPACES);
6501	}
6502
6503	static struct cgroup_namespace *alloc_cgroup_ns(void)
6504	{
6505	struct cgroup_namespace *new_ns;
6506	int ret;
6507
6508	new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL);
6509	if (!new_ns)
6510	return ERR_PTR(-ENOMEM);
6511	ret = ns_alloc_inum(&new_ns->ns);
6512	if (ret) {
6513	kfree(new_ns);
6514	return ERR_PTR(ret);
6515	}
6516	atomic_set(&new_ns->count, 1);
6517	new_ns->ns.ops = &cgroupns_operations;
6518	return new_ns;
6519	}
6520
6521	void free_cgroup_ns(struct cgroup_namespace *ns)
6522	{
6523	put_css_set(ns->root_cset);
6524	dec_cgroup_namespaces(ns->ucounts);
6525	put_user_ns(ns->user_ns);
6526	ns_free_inum(&ns->ns);
6527	kfree(ns);
6528	}
6529	EXPORT_SYMBOL(free_cgroup_ns);
6530
6531	struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
6532	struct user_namespace *user_ns,
6533	struct cgroup_namespace *old_ns)
6534	{
6535	struct cgroup_namespace *new_ns;
6536	struct ucounts *ucounts;
6537	struct css_set *cset;
6538
6539	BUG_ON(!old_ns);
6540
6541	if (!(flags & CLONE_NEWCGROUP)) {
6542	get_cgroup_ns(old_ns);
6543	return old_ns;
6544	}
6545
6546	/* Allow only sysadmin to create cgroup namespace. */
6547	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
6548	return ERR_PTR(-EPERM);
6549
6550	ucounts = inc_cgroup_namespaces(user_ns);
6551	if (!ucounts)
6552	return ERR_PTR(-ENOSPC);
6553
6554	/* It is not safe to take cgroup_mutex here */
6555	spin_lock_irq(&css_set_lock);
6556	cset = task_css_set(current);
6557	get_css_set(cset);
6558	spin_unlock_irq(&css_set_lock);
6559
6560	new_ns = alloc_cgroup_ns();
6561	if (IS_ERR(new_ns)) {
6562	put_css_set(cset);
6563	dec_cgroup_namespaces(ucounts);
6564	return new_ns;
6565	}
6566
6567	new_ns->user_ns = get_user_ns(user_ns);
6568	new_ns->ucounts = ucounts;
6569	new_ns->root_cset = cset;
6570
6571	return new_ns;
6572	}
6573
6574	static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns)
6575	{
6576	return container_of(ns, struct cgroup_namespace, ns);
6577	}
6578
6579	static int cgroupns_install(struct nsproxy *nsproxy, struct ns_common *ns)
6580	{
6581	struct cgroup_namespace *cgroup_ns = to_cg_ns(ns);
6582
6583	if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) ||
6584	!ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN))
6585	return -EPERM;
6586
6587	/* Don't need to do anything if we are attaching to our own cgroupns. */
6588	if (cgroup_ns == nsproxy->cgroup_ns)
6589	return 0;
6590
6591	get_cgroup_ns(cgroup_ns);
6592	put_cgroup_ns(nsproxy->cgroup_ns);
6593	nsproxy->cgroup_ns = cgroup_ns;
6594
6595	return 0;
6596	}
6597
6598	static struct ns_common *cgroupns_get(struct task_struct *task)
6599	{
6600	struct cgroup_namespace *ns = NULL;
6601	struct nsproxy *nsproxy;
6602
6603	task_lock(task);
6604	nsproxy = task->nsproxy;
6605	if (nsproxy) {
6606	ns = nsproxy->cgroup_ns;
6607	get_cgroup_ns(ns);
6608	}
6609	task_unlock(task);
6610
6611	return ns ? &ns->ns : NULL;
6612	}
6613
6614	static void cgroupns_put(struct ns_common *ns)
6615	{
6616	put_cgroup_ns(to_cg_ns(ns));
6617	}
6618
6619	static struct user_namespace *cgroupns_owner(struct ns_common *ns)
6620	{
6621	return to_cg_ns(ns)->user_ns;
6622	}
6623
6624	const struct proc_ns_operations cgroupns_operations = {
6625	.name = "cgroup",
6626	.type = CLONE_NEWCGROUP,
6627	.get = cgroupns_get,
6628	.put = cgroupns_put,
6629	.install = cgroupns_install,
6630	.owner = cgroupns_owner,
6631	};
6632
6633	static __init int cgroup_namespaces_init(void)
6634	{
6635	return 0;
6636	}
6637	subsys_initcall(cgroup_namespaces_init);
6638
6639	#ifdef CONFIG_CGROUP_BPF
6640	int cgroup_bpf_update(struct cgroup *cgrp, struct bpf_prog *prog,
6641	enum bpf_attach_type type, bool overridable)
6642	{
6643	struct cgroup *parent = cgroup_parent(cgrp);
6644	int ret;
6645
6646	mutex_lock(&cgroup_mutex);
6647	ret = __cgroup_bpf_update(cgrp, parent, prog, type, overridable);
6648	mutex_unlock(&cgroup_mutex);
6649	return ret;
6650	}
6651	#endif /* CONFIG_CGROUP_BPF */
6652
6653	#ifdef CONFIG_CGROUP_DEBUG
6654	static struct cgroup_subsys_state *
6655	debug_css_alloc(struct cgroup_subsys_state *parent_css)
6656	{
6657	struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
6658
6659	if (!css)
6660	return ERR_PTR(-ENOMEM);
6661
6662	return css;
6663	}
6664
6665	static void debug_css_free(struct cgroup_subsys_state *css)
6666	{
6667	kfree(css);
6668	}
6669
6670	static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
6671	struct cftype *cft)
6672	{
6673	return cgroup_task_count(css->cgroup);
6674	}
6675
6676	static u64 current_css_set_read(struct cgroup_subsys_state *css,
6677	struct cftype *cft)
6678	{
6679	return (u64)(unsigned long)current->cgroups;
6680	}
6681
6682	static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
6683	struct cftype *cft)
6684	{
6685	u64 count;
6686
6687	rcu_read_lock();
6688	count = atomic_read(&task_css_set(current)->refcount);
6689	rcu_read_unlock();
6690	return count;
6691	}
6692
6693	static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
6694	{
6695	struct cgrp_cset_link *link;
6696	struct css_set *cset;
6697	char *name_buf;
6698
6699	name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
6700	if (!name_buf)
6701	return -ENOMEM;
6702
6703	spin_lock_irq(&css_set_lock);
6704	rcu_read_lock();
6705	cset = rcu_dereference(current->cgroups);
6706	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
6707	struct cgroup *c = link->cgrp;
6708
6709	cgroup_name(c, name_buf, NAME_MAX + 1);
6710	seq_printf(seq, "Root %d group %s\n",
6711	c->root->hierarchy_id, name_buf);
6712	}
6713	rcu_read_unlock();
6714	spin_unlock_irq(&css_set_lock);
6715	kfree(name_buf);
6716	return 0;
6717	}
6718
6719	#define MAX_TASKS_SHOWN_PER_CSS 25
6720	static int cgroup_css_links_read(struct seq_file *seq, void *v)
6721	{
6722	struct cgroup_subsys_state *css = seq_css(seq);
6723	struct cgrp_cset_link *link;
6724
6725	spin_lock_irq(&css_set_lock);
6726	list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
6727	struct css_set *cset = link->cset;
6728	struct task_struct *task;
6729	int count = 0;
6730
6731	/*
6732	* Fix for android.security.sts.Poc16_11#testPocCVE_2016_6753
6733	* We should not expose kernel address info to user space
6734	*/
6735	#ifdef CONFIG_AMLOGIC_MODIFY
6736	seq_puts(seq, "css_set (____ptrval____)\n");
6737	#else
6738	seq_printf(seq, "css_set %p\n", cset);
6739	#endif
6740
6741	list_for_each_entry(task, &cset->tasks, cg_list) {
6742	if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6743	goto overflow;
6744	seq_printf(seq, " task %d\n", task_pid_vnr(task));
6745	}
6746
6747	list_for_each_entry(task, &cset->mg_tasks, cg_list) {
6748	if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6749	goto overflow;
6750	seq_printf(seq, " task %d\n", task_pid_vnr(task));
6751	}
6752	continue;
6753	overflow:
6754	seq_puts(seq, " ...\n");
6755	}
6756	spin_unlock_irq(&css_set_lock);
6757	return 0;
6758	}
6759
6760	static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
6761	{
6762	return (!cgroup_is_populated(css->cgroup) &&
6763	!css_has_online_children(&css->cgroup->self));
6764	}
6765
6766	static struct cftype debug_files[] = {
6767	{
6768	.name = "taskcount",
6769	.read_u64 = debug_taskcount_read,
6770	},
6771
6772	{
6773	.name = "current_css_set",
6774	.read_u64 = current_css_set_read,
6775	},
6776
6777	{
6778	.name = "current_css_set_refcount",
6779	.read_u64 = current_css_set_refcount_read,
6780	},
6781
6782	{
6783	.name = "current_css_set_cg_links",
6784	.seq_show = current_css_set_cg_links_read,
6785	},
6786
6787	{
6788	.name = "cgroup_css_links",
6789	.seq_show = cgroup_css_links_read,
6790	},
6791
6792	{
6793	.name = "releasable",
6794	.read_u64 = releasable_read,
6795	},
6796
6797	{ } /* terminate */
6798	};
6799
6800	struct cgroup_subsys debug_cgrp_subsys = {
6801	.css_alloc = debug_css_alloc,
6802	.css_free = debug_css_free,
6803	.legacy_cftypes = debug_files,
6804	};
6805	#endif /* CONFIG_CGROUP_DEBUG */
6806