linux - kernel/cgroup.c

6564 lines · 174.5 KiB

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 <net/sock.h>
66
67	/*
68	* pidlists linger the following amount before being destroyed. The goal
69	* is avoiding frequent destruction in the middle of consecutive read calls
70	* Expiring in the middle is a performance problem not a correctness one.
71	* 1 sec should be enough.
72	*/
73	#define CGROUP_PIDLIST_DESTROY_DELAY HZ
74
75	#define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
76	MAX_CFTYPE_NAME + 2)
77
78	/*
79	* cgroup_mutex is the master lock. Any modification to cgroup or its
80	* hierarchy must be performed while holding it.
81	*
82	* css_set_lock protects task->cgroups pointer, the list of css_set
83	* objects, and the chain of tasks off each css_set.
84	*
85	* These locks are exported if CONFIG_PROVE_RCU so that accessors in
86	* cgroup.h can use them for lockdep annotations.
87	*/
88	#ifdef CONFIG_PROVE_RCU
89	DEFINE_MUTEX(cgroup_mutex);
90	DEFINE_SPINLOCK(css_set_lock);
91	EXPORT_SYMBOL_GPL(cgroup_mutex);
92	EXPORT_SYMBOL_GPL(css_set_lock);
93	#else
94	static DEFINE_MUTEX(cgroup_mutex);
95	static DEFINE_SPINLOCK(css_set_lock);
96	#endif
97
98	/*
99	* Protects cgroup_idr and css_idr so that IDs can be released without
100	* grabbing cgroup_mutex.
101	*/
102	static DEFINE_SPINLOCK(cgroup_idr_lock);
103
104	/*
105	* Protects cgroup_file->kn for !self csses. It synchronizes notifications
106	* against file removal/re-creation across css hiding.
107	*/
108	static DEFINE_SPINLOCK(cgroup_file_kn_lock);
109
110	/*
111	* Protects cgroup_subsys->release_agent_path. Modifying it also requires
112	* cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
113	*/
114	static DEFINE_SPINLOCK(release_agent_path_lock);
115
116	struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
117
118	#define cgroup_assert_mutex_or_rcu_locked() \
119	RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
120	!lockdep_is_held(&cgroup_mutex), \
121	"cgroup_mutex or RCU read lock required");
122
123	/*
124	* cgroup destruction makes heavy use of work items and there can be a lot
125	* of concurrent destructions. Use a separate workqueue so that cgroup
126	* destruction work items don't end up filling up max_active of system_wq
127	* which may lead to deadlock.
128	*/
129	static struct workqueue_struct *cgroup_destroy_wq;
130
131	/*
132	* pidlist destructions need to be flushed on cgroup destruction. Use a
133	* separate workqueue as flush domain.
134	*/
135	static struct workqueue_struct *cgroup_pidlist_destroy_wq;
136
137	/* generate an array of cgroup subsystem pointers */
138	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
139	static struct cgroup_subsys *cgroup_subsys[] = {
140	#include <linux/cgroup_subsys.h>
141	};
142	#undef SUBSYS
143
144	/* array of cgroup subsystem names */
145	#define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
146	static const char *cgroup_subsys_name[] = {
147	#include <linux/cgroup_subsys.h>
148	};
149	#undef SUBSYS
150
151	/* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
152	#define SUBSYS(_x) \
153	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
154	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
155	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
156	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
157	#include <linux/cgroup_subsys.h>
158	#undef SUBSYS
159
160	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
161	static struct static_key_true *cgroup_subsys_enabled_key[] = {
162	#include <linux/cgroup_subsys.h>
163	};
164	#undef SUBSYS
165
166	#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
167	static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
168	#include <linux/cgroup_subsys.h>
169	};
170	#undef SUBSYS
171
172	/*
173	* The default hierarchy, reserved for the subsystems that are otherwise
174	* unattached - it never has more than a single cgroup, and all tasks are
175	* part of that cgroup.
176	*/
177	struct cgroup_root cgrp_dfl_root;
178	EXPORT_SYMBOL_GPL(cgrp_dfl_root);
179
180	/*
181	* The default hierarchy always exists but is hidden until mounted for the
182	* first time. This is for backward compatibility.
183	*/
184	static bool cgrp_dfl_visible;
185
186	/* Controllers blocked by the commandline in v1 */
187	static u16 cgroup_no_v1_mask;
188
189	/* some controllers are not supported in the default hierarchy */
190	static u16 cgrp_dfl_inhibit_ss_mask;
191
192	/* some controllers are implicitly enabled on the default hierarchy */
193	static unsigned long cgrp_dfl_implicit_ss_mask;
194
195	/* The list of hierarchy roots */
196
197	static LIST_HEAD(cgroup_roots);
198	static int cgroup_root_count;
199
200	/* hierarchy ID allocation and mapping, protected by cgroup_mutex */
201	static DEFINE_IDR(cgroup_hierarchy_idr);
202
203	/*
204	* Assign a monotonically increasing serial number to csses. It guarantees
205	* cgroups with bigger numbers are newer than those with smaller numbers.
206	* Also, as csses are always appended to the parent's ->children list, it
207	* guarantees that sibling csses are always sorted in the ascending serial
208	* number order on the list. Protected by cgroup_mutex.
209	*/
210	static u64 css_serial_nr_next = 1;
211
212	/*
213	* These bitmask flags indicate whether tasks in the fork and exit paths have
214	* fork/exit handlers to call. This avoids us having to do extra work in the
215	* fork/exit path to check which subsystems have fork/exit callbacks.
216	*/
217	static u16 have_fork_callback __read_mostly;
218	static u16 have_exit_callback __read_mostly;
219	static u16 have_free_callback __read_mostly;
220
221	/* cgroup namespace for init task */
222	struct cgroup_namespace init_cgroup_ns = {
223	.count = { .counter = 2, },
224	.user_ns = &init_user_ns,
225	.ns.ops = &cgroupns_operations,
226	.ns.inum = PROC_CGROUP_INIT_INO,
227	.root_cset = &init_css_set,
228	};
229
230	/* Ditto for the can_fork callback. */
231	static u16 have_canfork_callback __read_mostly;
232
233	static struct file_system_type cgroup2_fs_type;
234	static struct cftype cgroup_dfl_base_files[];
235	static struct cftype cgroup_legacy_base_files[];
236
237	static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
238	static void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
239	static int cgroup_apply_control(struct cgroup *cgrp);
240	static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
241	static void css_task_iter_advance(struct css_task_iter *it);
242	static int cgroup_destroy_locked(struct cgroup *cgrp);
243	static struct cgroup_subsys_state css_create(struct cgroup cgrp,
244	struct cgroup_subsys *ss);
245	static void css_release(struct percpu_ref *ref);
246	static void kill_css(struct cgroup_subsys_state *css);
247	static int cgroup_addrm_files(struct cgroup_subsys_state *css,
248	struct cgroup *cgrp, struct cftype cfts[],
249	bool is_add);
250
251	/**
252	* cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
253	* @ssid: subsys ID of interest
254	*
255	* cgroup_subsys_enabled() can only be used with literal subsys names which
256	* is fine for individual subsystems but unsuitable for cgroup core. This
257	* is slower static_key_enabled() based test indexed by @ssid.
258	*/
259	static bool cgroup_ssid_enabled(int ssid)
260	{
261	if (CGROUP_SUBSYS_COUNT == 0)
262	return false;
263
264	return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
265	}
266
267	static bool cgroup_ssid_no_v1(int ssid)
268	{
269	return cgroup_no_v1_mask & (1 << ssid);
270	}
271
272	/**
273	* cgroup_on_dfl - test whether a cgroup is on the default hierarchy
274	* @cgrp: the cgroup of interest
275	*
276	* The default hierarchy is the v2 interface of cgroup and this function
277	* can be used to test whether a cgroup is on the default hierarchy for
278	* cases where a subsystem should behave differnetly depending on the
279	* interface version.
280	*
281	* The set of behaviors which change on the default hierarchy are still
282	* being determined and the mount option is prefixed with __DEVEL__.
283	*
284	* List of changed behaviors:
285	*
286	* - Mount options "noprefix", "xattr", "clone_children", "release_agent"
287	* and "name" are disallowed.
288	*
289	* - When mounting an existing superblock, mount options should match.
290	*
291	* - Remount is disallowed.
292	*
293	* - rename(2) is disallowed.
294	*
295	* - "tasks" is removed. Everything should be at process granularity. Use
296	* "cgroup.procs" instead.
297	*
298	* - "cgroup.procs" is not sorted. pids will be unique unless they got
299	* recycled inbetween reads.
300	*
301	* - "release_agent" and "notify_on_release" are removed. Replacement
302	* notification mechanism will be implemented.
303	*
304	* - "cgroup.clone_children" is removed.
305	*
306	* - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
307	* and its descendants contain no task; otherwise, 1. The file also
308	* generates kernfs notification which can be monitored through poll and
309	* [di]notify when the value of the file changes.
310	*
311	* - cpuset: tasks will be kept in empty cpusets when hotplug happens and
312	* take masks of ancestors with non-empty cpus/mems, instead of being
313	* moved to an ancestor.
314	*
315	* - cpuset: a task can be moved into an empty cpuset, and again it takes
316	* masks of ancestors.
317	*
318	* - memcg: use_hierarchy is on by default and the cgroup file for the flag
319	* is not created.
320	*
321	* - blkcg: blk-throttle becomes properly hierarchical.
322	*
323	* - debug: disallowed on the default hierarchy.
324	*/
325	static bool cgroup_on_dfl(const struct cgroup *cgrp)
326	{
327	return cgrp->root == &cgrp_dfl_root;
328	}
329
330	/* IDR wrappers which synchronize using cgroup_idr_lock */
331	static int cgroup_idr_alloc(struct idr idr, void ptr, int start, int end,
332	gfp_t gfp_mask)
333	{
334	int ret;
335
336	idr_preload(gfp_mask);
337	spin_lock_bh(&cgroup_idr_lock);
338	ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
339	spin_unlock_bh(&cgroup_idr_lock);
340	idr_preload_end();
341	return ret;
342	}
343
344	static void cgroup_idr_replace(struct idr idr, void *ptr, int id)
345	{
346	void *ret;
347
348	spin_lock_bh(&cgroup_idr_lock);
349	ret = idr_replace(idr, ptr, id);
350	spin_unlock_bh(&cgroup_idr_lock);
351	return ret;
352	}
353
354	static void cgroup_idr_remove(struct idr *idr, int id)
355	{
356	spin_lock_bh(&cgroup_idr_lock);
357	idr_remove(idr, id);
358	spin_unlock_bh(&cgroup_idr_lock);
359	}
360
361	static struct cgroup cgroup_parent(struct cgroup cgrp)
362	{
363	struct cgroup_subsys_state *parent_css = cgrp->self.parent;
364
365	if (parent_css)
366	return container_of(parent_css, struct cgroup, self);
367	return NULL;
368	}
369
370	/* subsystems visibly enabled on a cgroup */
371	static u16 cgroup_control(struct cgroup *cgrp)
372	{
373	struct cgroup *parent = cgroup_parent(cgrp);
374	u16 root_ss_mask = cgrp->root->subsys_mask;
375
376	if (parent)
377	return parent->subtree_control;
378
379	if (cgroup_on_dfl(cgrp))
380	root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask \|
381	cgrp_dfl_implicit_ss_mask);
382	return root_ss_mask;
383	}
384
385	/* subsystems enabled on a cgroup */
386	static u16 cgroup_ss_mask(struct cgroup *cgrp)
387	{
388	struct cgroup *parent = cgroup_parent(cgrp);
389
390	if (parent)
391	return parent->subtree_ss_mask;
392
393	return cgrp->root->subsys_mask;
394	}
395
396	/**
397	* cgroup_css - obtain a cgroup's css for the specified subsystem
398	* @cgrp: the cgroup of interest
399	* @ss: the subsystem of interest (%NULL returns @cgrp->self)
400	*
401	* Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
402	* function must be called either under cgroup_mutex or rcu_read_lock() and
403	* the caller is responsible for pinning the returned css if it wants to
404	* keep accessing it outside the said locks. This function may return
405	* %NULL if @cgrp doesn't have @subsys_id enabled.
406	*/
407	static struct cgroup_subsys_state cgroup_css(struct cgroup cgrp,
408	struct cgroup_subsys *ss)
409	{
410	if (ss)
411	return rcu_dereference_check(cgrp->subsys[ss->id],
412	lockdep_is_held(&cgroup_mutex));
413	else
414	return &cgrp->self;
415	}
416
417	/**
418	* cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
419	* @cgrp: the cgroup of interest
420	* @ss: the subsystem of interest (%NULL returns @cgrp->self)
421	*
422	* Similar to cgroup_css() but returns the effective css, which is defined
423	* as the matching css of the nearest ancestor including self which has @ss
424	* enabled. If @ss is associated with the hierarchy @cgrp is on, this
425	* function is guaranteed to return non-NULL css.
426	*/
427	static struct cgroup_subsys_state cgroup_e_css(struct cgroup cgrp,
428	struct cgroup_subsys *ss)
429	{
430	lockdep_assert_held(&cgroup_mutex);
431
432	if (!ss)
433	return &cgrp->self;
434
435	/*
436	* This function is used while updating css associations and thus
437	* can't test the csses directly. Test ss_mask.
438	*/
439	while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
440	cgrp = cgroup_parent(cgrp);
441	if (!cgrp)
442	return NULL;
443	}
444
445	return cgroup_css(cgrp, ss);
446	}
447
448	/**
449	* cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
450	* @cgrp: the cgroup of interest
451	* @ss: the subsystem of interest
452	*
453	* Find and get the effective css of @cgrp for @ss. The effective css is
454	* defined as the matching css of the nearest ancestor including self which
455	* has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
456	* the root css is returned, so this function always returns a valid css.
457	* The returned css must be put using css_put().
458	*/
459	struct cgroup_subsys_state cgroup_get_e_css(struct cgroup cgrp,
460	struct cgroup_subsys *ss)
461	{
462	struct cgroup_subsys_state *css;
463
464	rcu_read_lock();
465
466	do {
467	css = cgroup_css(cgrp, ss);
468
469	if (css && css_tryget_online(css))
470	goto out_unlock;
471	cgrp = cgroup_parent(cgrp);
472	} while (cgrp);
473
474	css = init_css_set.subsys[ss->id];
475	css_get(css);
476	out_unlock:
477	rcu_read_unlock();
478	return css;
479	}
480
481	/* convenient tests for these bits */
482	static inline bool cgroup_is_dead(const struct cgroup *cgrp)
483	{
484	return !(cgrp->self.flags & CSS_ONLINE);
485	}
486
487	static void cgroup_get(struct cgroup *cgrp)
488	{
489	WARN_ON_ONCE(cgroup_is_dead(cgrp));
490	css_get(&cgrp->self);
491	}
492
493	static bool cgroup_tryget(struct cgroup *cgrp)
494	{
495	return css_tryget(&cgrp->self);
496	}
497
498	struct cgroup_subsys_state of_css(struct kernfs_open_file of)
499	{
500	struct cgroup *cgrp = of->kn->parent->priv;
501	struct cftype *cft = of_cft(of);
502
503	/*
504	* This is open and unprotected implementation of cgroup_css().
505	* seq_css() is only called from a kernfs file operation which has
506	* an active reference on the file. Because all the subsystem
507	* files are drained before a css is disassociated with a cgroup,
508	* the matching css from the cgroup's subsys table is guaranteed to
509	* be and stay valid until the enclosing operation is complete.
510	*/
511	if (cft->ss)
512	return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
513	else
514	return &cgrp->self;
515	}
516	EXPORT_SYMBOL_GPL(of_css);
517
518	static int notify_on_release(const struct cgroup *cgrp)
519	{
520	return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
521	}
522
523	/**
524	* for_each_css - iterate all css's of a cgroup
525	* @css: the iteration cursor
526	* @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
527	* @cgrp: the target cgroup to iterate css's of
528	*
529	* Should be called under cgroup_[tree_]mutex.
530	*/
531	#define for_each_css(css, ssid, cgrp) \
532	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
533	if (!((css) = rcu_dereference_check( \
534	(cgrp)->subsys[(ssid)], \
535	lockdep_is_held(&cgroup_mutex)))) { } \
536	else
537
538	/**
539	* for_each_e_css - iterate all effective css's of a cgroup
540	* @css: the iteration cursor
541	* @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
542	* @cgrp: the target cgroup to iterate css's of
543	*
544	* Should be called under cgroup_[tree_]mutex.
545	*/
546	#define for_each_e_css(css, ssid, cgrp) \
547	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
548	if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
549	; \
550	else
551
552	/**
553	* for_each_subsys - iterate all enabled cgroup subsystems
554	* @ss: the iteration cursor
555	* @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
556	*/
557	#define for_each_subsys(ss, ssid) \
558	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
559	(((ss) = cgroup_subsys[ssid]) \|\| true); (ssid)++)
560
561	/**
562	* do_each_subsys_mask - filter for_each_subsys with a bitmask
563	* @ss: the iteration cursor
564	* @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
565	* @ss_mask: the bitmask
566	*
567	* The block will only run for cases where the ssid-th bit (1 << ssid) of
568	* @ss_mask is set.
569	*/
570	#define do_each_subsys_mask(ss, ssid, ss_mask) do { \
571	unsigned long __ss_mask = (ss_mask); \
572	if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
573	(ssid) = 0; \
574	break; \
575	} \
576	for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
577	(ss) = cgroup_subsys[ssid]; \
578	{
579
580	#define while_each_subsys_mask() \
581	} \
582	} \
583	} while (false)
584
585	/* iterate across the hierarchies */
586	#define for_each_root(root) \
587	list_for_each_entry((root), &cgroup_roots, root_list)
588
589	/* iterate over child cgrps, lock should be held throughout iteration */
590	#define cgroup_for_each_live_child(child, cgrp) \
591	list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
592	if (({ lockdep_assert_held(&cgroup_mutex); \
593	cgroup_is_dead(child); })) \
594	; \
595	else
596
597	/* walk live descendants in preorder */
598	#define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
599	css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
600	if (({ lockdep_assert_held(&cgroup_mutex); \
601	(dsct) = (d_css)->cgroup; \
602	cgroup_is_dead(dsct); })) \
603	; \
604	else
605
606	/* walk live descendants in postorder */
607	#define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
608	css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
609	if (({ lockdep_assert_held(&cgroup_mutex); \
610	(dsct) = (d_css)->cgroup; \
611	cgroup_is_dead(dsct); })) \
612	; \
613	else
614
615	static void cgroup_release_agent(struct work_struct *work);
616	static void check_for_release(struct cgroup *cgrp);
617
618	/*
619	* A cgroup can be associated with multiple css_sets as different tasks may
620	* belong to different cgroups on different hierarchies. In the other
621	* direction, a css_set is naturally associated with multiple cgroups.
622	* This M:N relationship is represented by the following link structure
623	* which exists for each association and allows traversing the associations
624	* from both sides.
625	*/
626	struct cgrp_cset_link {
627	/* the cgroup and css_set this link associates */
628	struct cgroup *cgrp;
629	struct css_set *cset;
630
631	/* list of cgrp_cset_links anchored at cgrp->cset_links */
632	struct list_head cset_link;
633
634	/* list of cgrp_cset_links anchored at css_set->cgrp_links */
635	struct list_head cgrp_link;
636	};
637
638	/*
639	* The default css_set - used by init and its children prior to any
640	* hierarchies being mounted. It contains a pointer to the root state
641	* for each subsystem. Also used to anchor the list of css_sets. Not
642	* reference-counted, to improve performance when child cgroups
643	* haven't been created.
644	*/
645	struct css_set init_css_set = {
646	.refcount = ATOMIC_INIT(1),
647	.cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
648	.tasks = LIST_HEAD_INIT(init_css_set.tasks),
649	.mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
650	.mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
651	.mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
652	.task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
653	};
654
655	static int css_set_count = 1; /* 1 for init_css_set */
656
657	/**
658	* css_set_populated - does a css_set contain any tasks?
659	* @cset: target css_set
660	*/
661	static bool css_set_populated(struct css_set *cset)
662	{
663	lockdep_assert_held(&css_set_lock);
664
665	return !list_empty(&cset->tasks) \|\| !list_empty(&cset->mg_tasks);
666	}
667
668	/**
669	* cgroup_update_populated - updated populated count of a cgroup
670	* @cgrp: the target cgroup
671	* @populated: inc or dec populated count
672	*
673	* One of the css_sets associated with @cgrp is either getting its first
674	* task or losing the last. Update @cgrp->populated_cnt accordingly. The
675	* count is propagated towards root so that a given cgroup's populated_cnt
676	* is zero iff the cgroup and all its descendants don't contain any tasks.
677	*
678	* @cgrp's interface file "cgroup.populated" is zero if
679	* @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
680	* changes from or to zero, userland is notified that the content of the
681	* interface file has changed. This can be used to detect when @cgrp and
682	* its descendants become populated or empty.
683	*/
684	static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
685	{
686	lockdep_assert_held(&css_set_lock);
687
688	do {
689	bool trigger;
690
691	if (populated)
692	trigger = !cgrp->populated_cnt++;
693	else
694	trigger = !--cgrp->populated_cnt;
695
696	if (!trigger)
697	break;
698
699	check_for_release(cgrp);
700	cgroup_file_notify(&cgrp->events_file);
701
702	cgrp = cgroup_parent(cgrp);
703	} while (cgrp);
704	}
705
706	/**
707	* css_set_update_populated - update populated state of a css_set
708	* @cset: target css_set
709	* @populated: whether @cset is populated or depopulated
710	*
711	* @cset is either getting the first task or losing the last. Update the
712	* ->populated_cnt of all associated cgroups accordingly.
713	*/
714	static void css_set_update_populated(struct css_set *cset, bool populated)
715	{
716	struct cgrp_cset_link *link;
717
718	lockdep_assert_held(&css_set_lock);
719
720	list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
721	cgroup_update_populated(link->cgrp, populated);
722	}
723
724	/**
725	* css_set_move_task - move a task from one css_set to another
726	* @task: task being moved
727	* @from_cset: css_set @task currently belongs to (may be NULL)
728	* @to_cset: new css_set @task is being moved to (may be NULL)
729	* @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
730	*
731	* Move @task from @from_cset to @to_cset. If @task didn't belong to any
732	* css_set, @from_cset can be NULL. If @task is being disassociated
733	* instead of moved, @to_cset can be NULL.
734	*
735	* This function automatically handles populated_cnt updates and
736	* css_task_iter adjustments but the caller is responsible for managing
737	* @from_cset and @to_cset's reference counts.
738	*/
739	static void css_set_move_task(struct task_struct *task,
740	struct css_set from_cset, struct css_set to_cset,
741	bool use_mg_tasks)
742	{
743	lockdep_assert_held(&css_set_lock);
744
745	if (to_cset && !css_set_populated(to_cset))
746	css_set_update_populated(to_cset, true);
747
748	if (from_cset) {
749	struct css_task_iter it, pos;
750
751	WARN_ON_ONCE(list_empty(&task->cg_list));
752
753	/*
754	* @task is leaving, advance task iterators which are
755	* pointing to it so that they can resume at the next
756	* position. Advancing an iterator might remove it from
757	* the list, use safe walk. See css_task_iter_advance*()
758	* for details.
759	*/
760	list_for_each_entry_safe(it, pos, &from_cset->task_iters,
761	iters_node)
762	if (it->task_pos == &task->cg_list)
763	css_task_iter_advance(it);
764
765	list_del_init(&task->cg_list);
766	if (!css_set_populated(from_cset))
767	css_set_update_populated(from_cset, false);
768	} else {
769	WARN_ON_ONCE(!list_empty(&task->cg_list));
770	}
771
772	if (to_cset) {
773	/*
774	* We are synchronized through cgroup_threadgroup_rwsem
775	* against PF_EXITING setting such that we can't race
776	* against cgroup_exit() changing the css_set to
777	* init_css_set and dropping the old one.
778	*/
779	WARN_ON_ONCE(task->flags & PF_EXITING);
780
781	rcu_assign_pointer(task->cgroups, to_cset);
782	list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
783	&to_cset->tasks);
784	}
785	}
786
787	/*
788	* hash table for cgroup groups. This improves the performance to find
789	* an existing css_set. This hash doesn't (currently) take into
790	* account cgroups in empty hierarchies.
791	*/
792	#define CSS_SET_HASH_BITS 7
793	static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
794
795	static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
796	{
797	unsigned long key = 0UL;
798	struct cgroup_subsys *ss;
799	int i;
800
801	for_each_subsys(ss, i)
802	key += (unsigned long)css[i];
803	key = (key >> 16) ^ key;
804
805	return key;
806	}
807
808	static void put_css_set_locked(struct css_set *cset)
809	{
810	struct cgrp_cset_link link, tmp_link;
811	struct cgroup_subsys *ss;
812	int ssid;
813
814	lockdep_assert_held(&css_set_lock);
815
816	if (!atomic_dec_and_test(&cset->refcount))
817	return;
818
819	/* This css_set is dead. unlink it and release cgroup and css refs */
820	for_each_subsys(ss, ssid) {
821	list_del(&cset->e_cset_node[ssid]);
822	css_put(cset->subsys[ssid]);
823	}
824	hash_del(&cset->hlist);
825	css_set_count--;
826
827	list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
828	list_del(&link->cset_link);
829	list_del(&link->cgrp_link);
830	if (cgroup_parent(link->cgrp))
831	cgroup_put(link->cgrp);
832	kfree(link);
833	}
834
835	kfree_rcu(cset, rcu_head);
836	}
837
838	static void put_css_set(struct css_set *cset)
839	{
840	unsigned long flags;
841
842	/*
843	* Ensure that the refcount doesn't hit zero while any readers
844	* can see it. Similar to atomic_dec_and_lock(), but for an
845	* rwlock
846	*/
847	if (atomic_add_unless(&cset->refcount, -1, 1))
848	return;
849
850	spin_lock_irqsave(&css_set_lock, flags);
851	put_css_set_locked(cset);
852	spin_unlock_irqrestore(&css_set_lock, flags);
853	}
854
855	/*
856	* refcounted get/put for css_set objects
857	*/
858	static inline void get_css_set(struct css_set *cset)
859	{
860	atomic_inc(&cset->refcount);
861	}
862
863	/**
864	* compare_css_sets - helper function for find_existing_css_set().
865	* @cset: candidate css_set being tested
866	* @old_cset: existing css_set for a task
867	* @new_cgrp: cgroup that's being entered by the task
868	* @template: desired set of css pointers in css_set (pre-calculated)
869	*
870	* Returns true if "cset" matches "old_cset" except for the hierarchy
871	* which "new_cgrp" belongs to, for which it should match "new_cgrp".
872	*/
873	static bool compare_css_sets(struct css_set *cset,
874	struct css_set *old_cset,
875	struct cgroup *new_cgrp,
876	struct cgroup_subsys_state *template[])
877	{
878	struct list_head l1, l2;
879
880	/*
881	* On the default hierarchy, there can be csets which are
882	* associated with the same set of cgroups but different csses.
883	* Let's first ensure that csses match.
884	*/
885	if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
886	return false;
887
888	/*
889	* Compare cgroup pointers in order to distinguish between
890	* different cgroups in hierarchies. As different cgroups may
891	* share the same effective css, this comparison is always
892	* necessary.
893	*/
894	l1 = &cset->cgrp_links;
895	l2 = &old_cset->cgrp_links;
896	while (1) {
897	struct cgrp_cset_link link1, link2;
898	struct cgroup cgrp1, cgrp2;
899
900	l1 = l1->next;
901	l2 = l2->next;
902	/* See if we reached the end - both lists are equal length. */
903	if (l1 == &cset->cgrp_links) {
904	BUG_ON(l2 != &old_cset->cgrp_links);
905	break;
906	} else {
907	BUG_ON(l2 == &old_cset->cgrp_links);
908	}
909	/* Locate the cgroups associated with these links. */
910	link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
911	link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
912	cgrp1 = link1->cgrp;
913	cgrp2 = link2->cgrp;
914	/* Hierarchies should be linked in the same order. */
915	BUG_ON(cgrp1->root != cgrp2->root);
916
917	/*
918	* If this hierarchy is the hierarchy of the cgroup
919	* that's changing, then we need to check that this
920	* css_set points to the new cgroup; if it's any other
921	* hierarchy, then this css_set should point to the
922	* same cgroup as the old css_set.
923	*/
924	if (cgrp1->root == new_cgrp->root) {
925	if (cgrp1 != new_cgrp)
926	return false;
927	} else {
928	if (cgrp1 != cgrp2)
929	return false;
930	}
931	}
932	return true;
933	}
934
935	/**
936	* find_existing_css_set - init css array and find the matching css_set
937	* @old_cset: the css_set that we're using before the cgroup transition
938	* @cgrp: the cgroup that we're moving into
939	* @template: out param for the new set of csses, should be clear on entry
940	*/
941	static struct css_set find_existing_css_set(struct css_set old_cset,
942	struct cgroup *cgrp,
943	struct cgroup_subsys_state *template[])
944	{
945	struct cgroup_root *root = cgrp->root;
946	struct cgroup_subsys *ss;
947	struct css_set *cset;
948	unsigned long key;
949	int i;
950
951	/*
952	* Build the set of subsystem state objects that we want to see in the
953	* new css_set. while subsystems can change globally, the entries here
954	* won't change, so no need for locking.
955	*/
956	for_each_subsys(ss, i) {
957	if (root->subsys_mask & (1UL << i)) {
958	/*
959	* @ss is in this hierarchy, so we want the
960	* effective css from @cgrp.
961	*/
962	template[i] = cgroup_e_css(cgrp, ss);
963	} else {
964	/*
965	* @ss is not in this hierarchy, so we don't want
966	* to change the css.
967	*/
968	template[i] = old_cset->subsys[i];
969	}
970	}
971
972	key = css_set_hash(template);
973	hash_for_each_possible(css_set_table, cset, hlist, key) {
974	if (!compare_css_sets(cset, old_cset, cgrp, template))
975	continue;
976
977	/* This css_set matches what we need */
978	return cset;
979	}
980
981	/* No existing cgroup group matched */
982	return NULL;
983	}
984
985	static void free_cgrp_cset_links(struct list_head *links_to_free)
986	{
987	struct cgrp_cset_link link, tmp_link;
988
989	list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
990	list_del(&link->cset_link);
991	kfree(link);
992	}
993	}
994
995	/**
996	* allocate_cgrp_cset_links - allocate cgrp_cset_links
997	* @count: the number of links to allocate
998	* @tmp_links: list_head the allocated links are put on
999	*
1000	* Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1001	* through ->cset_link. Returns 0 on success or -errno.
1002	*/
1003	static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1004	{
1005	struct cgrp_cset_link *link;
1006	int i;
1007
1008	INIT_LIST_HEAD(tmp_links);
1009
1010	for (i = 0; i < count; i++) {
1011	link = kzalloc(sizeof(*link), GFP_KERNEL);
1012	if (!link) {
1013	free_cgrp_cset_links(tmp_links);
1014	return -ENOMEM;
1015	}
1016	list_add(&link->cset_link, tmp_links);
1017	}
1018	return 0;
1019	}
1020
1021	/**
1022	* link_css_set - a helper function to link a css_set to a cgroup
1023	* @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1024	* @cset: the css_set to be linked
1025	* @cgrp: the destination cgroup
1026	*/
1027	static void link_css_set(struct list_head tmp_links, struct css_set cset,
1028	struct cgroup *cgrp)
1029	{
1030	struct cgrp_cset_link *link;
1031
1032	BUG_ON(list_empty(tmp_links));
1033
1034	if (cgroup_on_dfl(cgrp))
1035	cset->dfl_cgrp = cgrp;
1036
1037	link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1038	link->cset = cset;
1039	link->cgrp = cgrp;
1040
1041	/*
1042	* Always add links to the tail of the lists so that the lists are
1043	* in choronological order.
1044	*/
1045	list_move_tail(&link->cset_link, &cgrp->cset_links);
1046	list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1047
1048	if (cgroup_parent(cgrp))
1049	cgroup_get(cgrp);
1050	}
1051
1052	/**
1053	* find_css_set - return a new css_set with one cgroup updated
1054	* @old_cset: the baseline css_set
1055	* @cgrp: the cgroup to be updated
1056	*
1057	* Return a new css_set that's equivalent to @old_cset, but with @cgrp
1058	* substituted into the appropriate hierarchy.
1059	*/
1060	static struct css_set find_css_set(struct css_set old_cset,
1061	struct cgroup *cgrp)
1062	{
1063	struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1064	struct css_set *cset;
1065	struct list_head tmp_links;
1066	struct cgrp_cset_link *link;
1067	struct cgroup_subsys *ss;
1068	unsigned long key;
1069	int ssid;
1070
1071	lockdep_assert_held(&cgroup_mutex);
1072
1073	/* First see if we already have a cgroup group that matches
1074	* the desired set */
1075	spin_lock_irq(&css_set_lock);
1076	cset = find_existing_css_set(old_cset, cgrp, template);
1077	if (cset)
1078	get_css_set(cset);
1079	spin_unlock_irq(&css_set_lock);
1080
1081	if (cset)
1082	return cset;
1083
1084	cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1085	if (!cset)
1086	return NULL;
1087
1088	/* Allocate all the cgrp_cset_link objects that we'll need */
1089	if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1090	kfree(cset);
1091	return NULL;
1092	}
1093
1094	atomic_set(&cset->refcount, 1);
1095	INIT_LIST_HEAD(&cset->cgrp_links);
1096	INIT_LIST_HEAD(&cset->tasks);
1097	INIT_LIST_HEAD(&cset->mg_tasks);
1098	INIT_LIST_HEAD(&cset->mg_preload_node);
1099	INIT_LIST_HEAD(&cset->mg_node);
1100	INIT_LIST_HEAD(&cset->task_iters);
1101	INIT_HLIST_NODE(&cset->hlist);
1102
1103	/* Copy the set of subsystem state objects generated in
1104	* find_existing_css_set() */
1105	memcpy(cset->subsys, template, sizeof(cset->subsys));
1106
1107	spin_lock_irq(&css_set_lock);
1108	/* Add reference counts and links from the new css_set. */
1109	list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1110	struct cgroup *c = link->cgrp;
1111
1112	if (c->root == cgrp->root)
1113	c = cgrp;
1114	link_css_set(&tmp_links, cset, c);
1115	}
1116
1117	BUG_ON(!list_empty(&tmp_links));
1118
1119	css_set_count++;
1120
1121	/* Add @cset to the hash table */
1122	key = css_set_hash(cset->subsys);
1123	hash_add(css_set_table, &cset->hlist, key);
1124
1125	for_each_subsys(ss, ssid) {
1126	struct cgroup_subsys_state *css = cset->subsys[ssid];
1127
1128	list_add_tail(&cset->e_cset_node[ssid],
1129	&css->cgroup->e_csets[ssid]);
1130	css_get(css);
1131	}
1132
1133	spin_unlock_irq(&css_set_lock);
1134
1135	return cset;
1136	}
1137
1138	static struct cgroup_root cgroup_root_from_kf(struct kernfs_root kf_root)
1139	{
1140	struct cgroup *root_cgrp = kf_root->kn->priv;
1141
1142	return root_cgrp->root;
1143	}
1144
1145	static int cgroup_init_root_id(struct cgroup_root *root)
1146	{
1147	int id;
1148
1149	lockdep_assert_held(&cgroup_mutex);
1150
1151	id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1152	if (id < 0)
1153	return id;
1154
1155	root->hierarchy_id = id;
1156	return 0;
1157	}
1158
1159	static void cgroup_exit_root_id(struct cgroup_root *root)
1160	{
1161	lockdep_assert_held(&cgroup_mutex);
1162
1163	idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1164	}
1165
1166	static void cgroup_free_root(struct cgroup_root *root)
1167	{
1168	if (root) {
1169	idr_destroy(&root->cgroup_idr);
1170	kfree(root);
1171	}
1172	}
1173
1174	static void cgroup_destroy_root(struct cgroup_root *root)
1175	{
1176	struct cgroup *cgrp = &root->cgrp;
1177	struct cgrp_cset_link link, tmp_link;
1178
1179	cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1180
1181	BUG_ON(atomic_read(&root->nr_cgrps));
1182	BUG_ON(!list_empty(&cgrp->self.children));
1183
1184	/* Rebind all subsystems back to the default hierarchy */
1185	WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1186
1187	/*
1188	* Release all the links from cset_links to this hierarchy's
1189	* root cgroup
1190	*/
1191	spin_lock_irq(&css_set_lock);
1192
1193	list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1194	list_del(&link->cset_link);
1195	list_del(&link->cgrp_link);
1196	kfree(link);
1197	}
1198
1199	spin_unlock_irq(&css_set_lock);
1200
1201	if (!list_empty(&root->root_list)) {
1202	list_del(&root->root_list);
1203	cgroup_root_count--;
1204	}
1205
1206	cgroup_exit_root_id(root);
1207
1208	mutex_unlock(&cgroup_mutex);
1209
1210	kernfs_destroy_root(root->kf_root);
1211	cgroup_free_root(root);
1212	}
1213
1214	/*
1215	* look up cgroup associated with current task's cgroup namespace on the
1216	* specified hierarchy
1217	*/
1218	static struct cgroup *
1219	current_cgns_cgroup_from_root(struct cgroup_root *root)
1220	{
1221	struct cgroup *res = NULL;
1222	struct css_set *cset;
1223
1224	lockdep_assert_held(&css_set_lock);
1225
1226	rcu_read_lock();
1227
1228	cset = current->nsproxy->cgroup_ns->root_cset;
1229	if (cset == &init_css_set) {
1230	res = &root->cgrp;
1231	} else {
1232	struct cgrp_cset_link *link;
1233
1234	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1235	struct cgroup *c = link->cgrp;
1236
1237	if (c->root == root) {
1238	res = c;
1239	break;
1240	}
1241	}
1242	}
1243	rcu_read_unlock();
1244
1245	BUG_ON(!res);
1246	return res;
1247	}
1248
1249	/* look up cgroup associated with given css_set on the specified hierarchy */
1250	static struct cgroup cset_cgroup_from_root(struct css_set cset,
1251	struct cgroup_root *root)
1252	{
1253	struct cgroup *res = NULL;
1254
1255	lockdep_assert_held(&cgroup_mutex);
1256	lockdep_assert_held(&css_set_lock);
1257
1258	if (cset == &init_css_set) {
1259	res = &root->cgrp;
1260	} else {
1261	struct cgrp_cset_link *link;
1262
1263	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1264	struct cgroup *c = link->cgrp;
1265
1266	if (c->root == root) {
1267	res = c;
1268	break;
1269	}
1270	}
1271	}
1272
1273	BUG_ON(!res);
1274	return res;
1275	}
1276
1277	/*
1278	* Return the cgroup for "task" from the given hierarchy. Must be
1279	* called with cgroup_mutex and css_set_lock held.
1280	*/
1281	static struct cgroup task_cgroup_from_root(struct task_struct task,
1282	struct cgroup_root *root)
1283	{
1284	/*
1285	* No need to lock the task - since we hold cgroup_mutex the
1286	* task can't change groups, so the only thing that can happen
1287	* is that it exits and its css is set back to init_css_set.
1288	*/
1289	return cset_cgroup_from_root(task_css_set(task), root);
1290	}
1291
1292	/*
1293	* A task must hold cgroup_mutex to modify cgroups.
1294	*
1295	* Any task can increment and decrement the count field without lock.
1296	* So in general, code holding cgroup_mutex can't rely on the count
1297	* field not changing. However, if the count goes to zero, then only
1298	* cgroup_attach_task() can increment it again. Because a count of zero
1299	* means that no tasks are currently attached, therefore there is no
1300	* way a task attached to that cgroup can fork (the other way to
1301	* increment the count). So code holding cgroup_mutex can safely
1302	* assume that if the count is zero, it will stay zero. Similarly, if
1303	* a task holds cgroup_mutex on a cgroup with zero count, it
1304	* knows that the cgroup won't be removed, as cgroup_rmdir()
1305	* needs that mutex.
1306	*
1307	* A cgroup can only be deleted if both its 'count' of using tasks
1308	* is zero, and its list of 'children' cgroups is empty. Since all
1309	* tasks in the system use _some_ cgroup, and since there is always at
1310	* least one task in the system (init, pid == 1), therefore, root cgroup
1311	* always has either children cgroups and/or using tasks. So we don't
1312	* need a special hack to ensure that root cgroup cannot be deleted.
1313	*
1314	* P.S. One more locking exception. RCU is used to guard the
1315	* update of a tasks cgroup pointer by cgroup_attach_task()
1316	*/
1317
1318	static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1319	static const struct file_operations proc_cgroupstats_operations;
1320
1321	static char cgroup_file_name(struct cgroup cgrp, const struct cftype *cft,
1322	char *buf)
1323	{
1324	struct cgroup_subsys *ss = cft->ss;
1325
1326	if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1327	!(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1328	snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1329	cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1330	cft->name);
1331	else
1332	strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1333	return buf;
1334	}
1335
1336	/**
1337	* cgroup_file_mode - deduce file mode of a control file
1338	* @cft: the control file in question
1339	*
1340	* S_IRUGO for read, S_IWUSR for write.
1341	*/
1342	static umode_t cgroup_file_mode(const struct cftype *cft)
1343	{
1344	umode_t mode = 0;
1345
1346	if (cft->read_u64 \|\| cft->read_s64 \|\| cft->seq_show)
1347	mode \|= S_IRUGO;
1348
1349	if (cft->write_u64 \|\| cft->write_s64 \|\| cft->write) {
1350	if (cft->flags & CFTYPE_WORLD_WRITABLE)
1351	mode \|= S_IWUGO;
1352	else
1353	mode \|= S_IWUSR;
1354	}
1355
1356	return mode;
1357	}
1358
1359	/**
1360	* cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1361	* @subtree_control: the new subtree_control mask to consider
1362	* @this_ss_mask: available subsystems
1363	*
1364	* On the default hierarchy, a subsystem may request other subsystems to be
1365	* enabled together through its ->depends_on mask. In such cases, more
1366	* subsystems than specified in "cgroup.subtree_control" may be enabled.
1367	*
1368	* This function calculates which subsystems need to be enabled if
1369	* @subtree_control is to be applied while restricted to @this_ss_mask.
1370	*/
1371	static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1372	{
1373	u16 cur_ss_mask = subtree_control;
1374	struct cgroup_subsys *ss;
1375	int ssid;
1376
1377	lockdep_assert_held(&cgroup_mutex);
1378
1379	cur_ss_mask \|= cgrp_dfl_implicit_ss_mask;
1380
1381	while (true) {
1382	u16 new_ss_mask = cur_ss_mask;
1383
1384	do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1385	new_ss_mask \|= ss->depends_on;
1386	} while_each_subsys_mask();
1387
1388	/*
1389	* Mask out subsystems which aren't available. This can
1390	* happen only if some depended-upon subsystems were bound
1391	* to non-default hierarchies.
1392	*/
1393	new_ss_mask &= this_ss_mask;
1394
1395	if (new_ss_mask == cur_ss_mask)
1396	break;
1397	cur_ss_mask = new_ss_mask;
1398	}
1399
1400	return cur_ss_mask;
1401	}
1402
1403	/**
1404	* cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1405	* @kn: the kernfs_node being serviced
1406	*
1407	* This helper undoes cgroup_kn_lock_live() and should be invoked before
1408	* the method finishes if locking succeeded. Note that once this function
1409	* returns the cgroup returned by cgroup_kn_lock_live() may become
1410	* inaccessible any time. If the caller intends to continue to access the
1411	* cgroup, it should pin it before invoking this function.
1412	*/
1413	static void cgroup_kn_unlock(struct kernfs_node *kn)
1414	{
1415	struct cgroup *cgrp;
1416
1417	if (kernfs_type(kn) == KERNFS_DIR)
1418	cgrp = kn->priv;
1419	else
1420	cgrp = kn->parent->priv;
1421
1422	mutex_unlock(&cgroup_mutex);
1423
1424	kernfs_unbreak_active_protection(kn);
1425	cgroup_put(cgrp);
1426	}
1427
1428	/**
1429	* cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1430	* @kn: the kernfs_node being serviced
1431	* @drain_offline: perform offline draining on the cgroup
1432	*
1433	* This helper is to be used by a cgroup kernfs method currently servicing
1434	* @kn. It breaks the active protection, performs cgroup locking and
1435	* verifies that the associated cgroup is alive. Returns the cgroup if
1436	* alive; otherwise, %NULL. A successful return should be undone by a
1437	* matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1438	* cgroup is drained of offlining csses before return.
1439	*
1440	* Any cgroup kernfs method implementation which requires locking the
1441	* associated cgroup should use this helper. It avoids nesting cgroup
1442	* locking under kernfs active protection and allows all kernfs operations
1443	* including self-removal.
1444	*/
1445	static struct cgroup cgroup_kn_lock_live(struct kernfs_node kn,
1446	bool drain_offline)
1447	{
1448	struct cgroup *cgrp;
1449
1450	if (kernfs_type(kn) == KERNFS_DIR)
1451	cgrp = kn->priv;
1452	else
1453	cgrp = kn->parent->priv;
1454
1455	/*
1456	* We're gonna grab cgroup_mutex which nests outside kernfs
1457	* active_ref. cgroup liveliness check alone provides enough
1458	* protection against removal. Ensure @cgrp stays accessible and
1459	* break the active_ref protection.
1460	*/
1461	if (!cgroup_tryget(cgrp))
1462	return NULL;
1463	kernfs_break_active_protection(kn);
1464
1465	if (drain_offline)
1466	cgroup_lock_and_drain_offline(cgrp);
1467	else
1468	mutex_lock(&cgroup_mutex);
1469
1470	if (!cgroup_is_dead(cgrp))
1471	return cgrp;
1472
1473	cgroup_kn_unlock(kn);
1474	return NULL;
1475	}
1476
1477	static void cgroup_rm_file(struct cgroup cgrp, const struct cftype cft)
1478	{
1479	char name[CGROUP_FILE_NAME_MAX];
1480
1481	lockdep_assert_held(&cgroup_mutex);
1482
1483	if (cft->file_offset) {
1484	struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1485	struct cgroup_file cfile = (void )css + cft->file_offset;
1486
1487	spin_lock_irq(&cgroup_file_kn_lock);
1488	cfile->kn = NULL;
1489	spin_unlock_irq(&cgroup_file_kn_lock);
1490	}
1491
1492	kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1493	}
1494
1495	/**
1496	* css_clear_dir - remove subsys files in a cgroup directory
1497	* @css: taget css
1498	*/
1499	static void css_clear_dir(struct cgroup_subsys_state *css)
1500	{
1501	struct cgroup *cgrp = css->cgroup;
1502	struct cftype *cfts;
1503
1504	if (!(css->flags & CSS_VISIBLE))
1505	return;
1506
1507	css->flags &= ~CSS_VISIBLE;
1508
1509	list_for_each_entry(cfts, &css->ss->cfts, node)
1510	cgroup_addrm_files(css, cgrp, cfts, false);
1511	}
1512
1513	/**
1514	* css_populate_dir - create subsys files in a cgroup directory
1515	* @css: target css
1516	*
1517	* On failure, no file is added.
1518	*/
1519	static int css_populate_dir(struct cgroup_subsys_state *css)
1520	{
1521	struct cgroup *cgrp = css->cgroup;
1522	struct cftype cfts, failed_cfts;
1523	int ret;
1524
1525	if ((css->flags & CSS_VISIBLE) \|\| !cgrp->kn)
1526	return 0;
1527
1528	if (!css->ss) {
1529	if (cgroup_on_dfl(cgrp))
1530	cfts = cgroup_dfl_base_files;
1531	else
1532	cfts = cgroup_legacy_base_files;
1533
1534	return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1535	}
1536
1537	list_for_each_entry(cfts, &css->ss->cfts, node) {
1538	ret = cgroup_addrm_files(css, cgrp, cfts, true);
1539	if (ret < 0) {
1540	failed_cfts = cfts;
1541	goto err;
1542	}
1543	}
1544
1545	css->flags \|= CSS_VISIBLE;
1546
1547	return 0;
1548	err:
1549	list_for_each_entry(cfts, &css->ss->cfts, node) {
1550	if (cfts == failed_cfts)
1551	break;
1552	cgroup_addrm_files(css, cgrp, cfts, false);
1553	}
1554	return ret;
1555	}
1556
1557	static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1558	{
1559	struct cgroup *dcgrp = &dst_root->cgrp;
1560	struct cgroup_subsys *ss;
1561	int ssid, i, ret;
1562
1563	lockdep_assert_held(&cgroup_mutex);
1564
1565	do_each_subsys_mask(ss, ssid, ss_mask) {
1566	/*
1567	* If @ss has non-root csses attached to it, can't move.
1568	* If @ss is an implicit controller, it is exempt from this
1569	* rule and can be stolen.
1570	*/
1571	if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1572	!ss->implicit_on_dfl)
1573	return -EBUSY;
1574
1575	/* can't move between two non-dummy roots either */
1576	if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1577	return -EBUSY;
1578	} while_each_subsys_mask();
1579
1580	do_each_subsys_mask(ss, ssid, ss_mask) {
1581	struct cgroup_root *src_root = ss->root;
1582	struct cgroup *scgrp = &src_root->cgrp;
1583	struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1584	struct css_set *cset;
1585
1586	WARN_ON(!css \|\| cgroup_css(dcgrp, ss));
1587
1588	/* disable from the source */
1589	src_root->subsys_mask &= ~(1 << ssid);
1590	WARN_ON(cgroup_apply_control(scgrp));
1591	cgroup_finalize_control(scgrp, 0);
1592
1593	/* rebind */
1594	RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1595	rcu_assign_pointer(dcgrp->subsys[ssid], css);
1596	ss->root = dst_root;
1597	css->cgroup = dcgrp;
1598
1599	spin_lock_irq(&css_set_lock);
1600	hash_for_each(css_set_table, i, cset, hlist)
1601	list_move_tail(&cset->e_cset_node[ss->id],
1602	&dcgrp->e_csets[ss->id]);
1603	spin_unlock_irq(&css_set_lock);
1604
1605	/* default hierarchy doesn't enable controllers by default */
1606	dst_root->subsys_mask \|= 1 << ssid;
1607	if (dst_root == &cgrp_dfl_root) {
1608	static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1609	} else {
1610	dcgrp->subtree_control \|= 1 << ssid;
1611	static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1612	}
1613
1614	ret = cgroup_apply_control(dcgrp);
1615	if (ret)
1616	pr_warn("partial failure to rebind %s controller (err=%d)\n",
1617	ss->name, ret);
1618
1619	if (ss->bind)
1620	ss->bind(css);
1621	} while_each_subsys_mask();
1622
1623	kernfs_activate(dcgrp->kn);
1624	return 0;
1625	}
1626
1627	static int cgroup_show_path(struct seq_file sf, struct kernfs_node kf_node,
1628	struct kernfs_root *kf_root)
1629	{
1630	int len = 0;
1631	char *buf = NULL;
1632	struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1633	struct cgroup *ns_cgroup;
1634
1635	buf = kmalloc(PATH_MAX, GFP_KERNEL);
1636	if (!buf)
1637	return -ENOMEM;
1638
1639	spin_lock_irq(&css_set_lock);
1640	ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1641	len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1642	spin_unlock_irq(&css_set_lock);
1643
1644	if (len >= PATH_MAX)
1645	len = -ERANGE;
1646	else if (len > 0) {
1647	seq_escape(sf, buf, " \t\n\\");
1648	len = 0;
1649	}
1650	kfree(buf);
1651	return len;
1652	}
1653
1654	static int cgroup_show_options(struct seq_file *seq,
1655	struct kernfs_root *kf_root)
1656	{
1657	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1658	struct cgroup_subsys *ss;
1659	int ssid;
1660
1661	if (root != &cgrp_dfl_root)
1662	for_each_subsys(ss, ssid)
1663	if (root->subsys_mask & (1 << ssid))
1664	seq_show_option(seq, ss->legacy_name, NULL);
1665	if (root->flags & CGRP_ROOT_NOPREFIX)
1666	seq_puts(seq, ",noprefix");
1667	if (root->flags & CGRP_ROOT_XATTR)
1668	seq_puts(seq, ",xattr");
1669
1670	spin_lock(&release_agent_path_lock);
1671	if (strlen(root->release_agent_path))
1672	seq_show_option(seq, "release_agent",
1673	root->release_agent_path);
1674	spin_unlock(&release_agent_path_lock);
1675
1676	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
1677	seq_puts(seq, ",clone_children");
1678	if (strlen(root->name))
1679	seq_show_option(seq, "name", root->name);
1680	return 0;
1681	}
1682
1683	struct cgroup_sb_opts {
1684	u16 subsys_mask;
1685	unsigned int flags;
1686	char *release_agent;
1687	bool cpuset_clone_children;
1688	char *name;
1689	/* User explicitly requested empty subsystem */
1690	bool none;
1691	};
1692
1693	static int parse_cgroupfs_options(char data, struct cgroup_sb_opts opts)
1694	{
1695	char token, o = data;
1696	bool all_ss = false, one_ss = false;
1697	u16 mask = U16_MAX;
1698	struct cgroup_subsys *ss;
1699	int nr_opts = 0;
1700	int i;
1701
1702	#ifdef CONFIG_CPUSETS
1703	mask = ~((u16)1 << cpuset_cgrp_id);
1704	#endif
1705
1706	memset(opts, 0, sizeof(*opts));
1707
1708	while ((token = strsep(&o, ",")) != NULL) {
1709	nr_opts++;
1710
1711	if (!*token)
1712	return -EINVAL;
1713	if (!strcmp(token, "none")) {
1714	/* Explicitly have no subsystems */
1715	opts->none = true;
1716	continue;
1717	}
1718	if (!strcmp(token, "all")) {
1719	/* Mutually exclusive option 'all' + subsystem name */
1720	if (one_ss)
1721	return -EINVAL;
1722	all_ss = true;
1723	continue;
1724	}
1725	if (!strcmp(token, "noprefix")) {
1726	opts->flags \|= CGRP_ROOT_NOPREFIX;
1727	continue;
1728	}
1729	if (!strcmp(token, "clone_children")) {
1730	opts->cpuset_clone_children = true;
1731	continue;
1732	}
1733	if (!strcmp(token, "xattr")) {
1734	opts->flags \|= CGRP_ROOT_XATTR;
1735	continue;
1736	}
1737	if (!strncmp(token, "release_agent=", 14)) {
1738	/* Specifying two release agents is forbidden */
1739	if (opts->release_agent)
1740	return -EINVAL;
1741	opts->release_agent =
1742	kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1743	if (!opts->release_agent)
1744	return -ENOMEM;
1745	continue;
1746	}
1747	if (!strncmp(token, "name=", 5)) {
1748	const char *name = token + 5;
1749	/* Can't specify an empty name */
1750	if (!strlen(name))
1751	return -EINVAL;
1752	/* Must match [\w.-]+ */
1753	for (i = 0; i < strlen(name); i++) {
1754	char c = name[i];
1755	if (isalnum(c))
1756	continue;
1757	if ((c == '.') \|\| (c == '-') \|\| (c == '_'))
1758	continue;
1759	return -EINVAL;
1760	}
1761	/* Specifying two names is forbidden */
1762	if (opts->name)
1763	return -EINVAL;
1764	opts->name = kstrndup(name,
1765	MAX_CGROUP_ROOT_NAMELEN - 1,
1766	GFP_KERNEL);
1767	if (!opts->name)
1768	return -ENOMEM;
1769
1770	continue;
1771	}
1772
1773	for_each_subsys(ss, i) {
1774	if (strcmp(token, ss->legacy_name))
1775	continue;
1776	if (!cgroup_ssid_enabled(i))
1777	continue;
1778	if (cgroup_ssid_no_v1(i))
1779	continue;
1780
1781	/* Mutually exclusive option 'all' + subsystem name */
1782	if (all_ss)
1783	return -EINVAL;
1784	opts->subsys_mask \|= (1 << i);
1785	one_ss = true;
1786
1787	break;
1788	}
1789	if (i == CGROUP_SUBSYS_COUNT)
1790	return -ENOENT;
1791	}
1792
1793	/*
1794	* If the 'all' option was specified select all the subsystems,
1795	* otherwise if 'none', 'name=' and a subsystem name options were
1796	* not specified, let's default to 'all'
1797	*/
1798	if (all_ss \|\| (!one_ss && !opts->none && !opts->name))
1799	for_each_subsys(ss, i)
1800	if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i))
1801	opts->subsys_mask \|= (1 << i);
1802
1803	/*
1804	* We either have to specify by name or by subsystems. (So all
1805	* empty hierarchies must have a name).
1806	*/
1807	if (!opts->subsys_mask && !opts->name)
1808	return -EINVAL;
1809
1810	/*
1811	* Option noprefix was introduced just for backward compatibility
1812	* with the old cpuset, so we allow noprefix only if mounting just
1813	* the cpuset subsystem.
1814	*/
1815	if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1816	return -EINVAL;
1817
1818	/* Can't specify "none" and some subsystems */
1819	if (opts->subsys_mask && opts->none)
1820	return -EINVAL;
1821
1822	return 0;
1823	}
1824
1825	static int cgroup_remount(struct kernfs_root kf_root, int flags, char *data)
1826	{
1827	int ret = 0;
1828	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1829	struct cgroup_sb_opts opts;
1830	u16 added_mask, removed_mask;
1831
1832	if (root == &cgrp_dfl_root) {
1833	pr_err("remount is not allowed\n");
1834	return -EINVAL;
1835	}
1836
1837	cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1838
1839	/* See what subsystems are wanted */
1840	ret = parse_cgroupfs_options(data, &opts);
1841	if (ret)
1842	goto out_unlock;
1843
1844	if (opts.subsys_mask != root->subsys_mask \|\| opts.release_agent)
1845	pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1846	task_tgid_nr(current), current->comm);
1847
1848	added_mask = opts.subsys_mask & ~root->subsys_mask;
1849	removed_mask = root->subsys_mask & ~opts.subsys_mask;
1850
1851	/* Don't allow flags or name to change at remount */
1852	if ((opts.flags ^ root->flags) \|\|
1853	(opts.name && strcmp(opts.name, root->name))) {
1854	pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1855	opts.flags, opts.name ?: "", root->flags, root->name);
1856	ret = -EINVAL;
1857	goto out_unlock;
1858	}
1859
1860	/* remounting is not allowed for populated hierarchies */
1861	if (!list_empty(&root->cgrp.self.children)) {
1862	ret = -EBUSY;
1863	goto out_unlock;
1864	}
1865
1866	ret = rebind_subsystems(root, added_mask);
1867	if (ret)
1868	goto out_unlock;
1869
1870	WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1871
1872	if (opts.release_agent) {
1873	spin_lock(&release_agent_path_lock);
1874	strcpy(root->release_agent_path, opts.release_agent);
1875	spin_unlock(&release_agent_path_lock);
1876	}
1877	out_unlock:
1878	kfree(opts.release_agent);
1879	kfree(opts.name);
1880	mutex_unlock(&cgroup_mutex);
1881	return ret;
1882	}
1883
1884	/*
1885	* To reduce the fork() overhead for systems that are not actually using
1886	* their cgroups capability, we don't maintain the lists running through
1887	* each css_set to its tasks until we see the list actually used - in other
1888	* words after the first mount.
1889	*/
1890	static bool use_task_css_set_links __read_mostly;
1891
1892	static void cgroup_enable_task_cg_lists(void)
1893	{
1894	struct task_struct p, g;
1895
1896	spin_lock_irq(&css_set_lock);
1897
1898	if (use_task_css_set_links)
1899	goto out_unlock;
1900
1901	use_task_css_set_links = true;
1902
1903	/*
1904	* We need tasklist_lock because RCU is not safe against
1905	* while_each_thread(). Besides, a forking task that has passed
1906	* cgroup_post_fork() without seeing use_task_css_set_links = 1
1907	* is not guaranteed to have its child immediately visible in the
1908	* tasklist if we walk through it with RCU.
1909	*/
1910	read_lock(&tasklist_lock);
1911	do_each_thread(g, p) {
1912	WARN_ON_ONCE(!list_empty(&p->cg_list) \|\|
1913	task_css_set(p) != &init_css_set);
1914
1915	/*
1916	* We should check if the process is exiting, otherwise
1917	* it will race with cgroup_exit() in that the list
1918	* entry won't be deleted though the process has exited.
1919	* Do it while holding siglock so that we don't end up
1920	* racing against cgroup_exit().
1921	*
1922	* Interrupts were already disabled while acquiring
1923	* the css_set_lock, so we do not need to disable it
1924	* again when acquiring the sighand->siglock here.
1925	*/
1926	spin_lock(&p->sighand->siglock);
1927	if (!(p->flags & PF_EXITING)) {
1928	struct css_set *cset = task_css_set(p);
1929
1930	if (!css_set_populated(cset))
1931	css_set_update_populated(cset, true);
1932	list_add_tail(&p->cg_list, &cset->tasks);
1933	get_css_set(cset);
1934	}
1935	spin_unlock(&p->sighand->siglock);
1936	} while_each_thread(g, p);
1937	read_unlock(&tasklist_lock);
1938	out_unlock:
1939	spin_unlock_irq(&css_set_lock);
1940	}
1941
1942	static void init_cgroup_housekeeping(struct cgroup *cgrp)
1943	{
1944	struct cgroup_subsys *ss;
1945	int ssid;
1946
1947	INIT_LIST_HEAD(&cgrp->self.sibling);
1948	INIT_LIST_HEAD(&cgrp->self.children);
1949	INIT_LIST_HEAD(&cgrp->cset_links);
1950	INIT_LIST_HEAD(&cgrp->pidlists);
1951	mutex_init(&cgrp->pidlist_mutex);
1952	cgrp->self.cgroup = cgrp;
1953	cgrp->self.flags \|= CSS_ONLINE;
1954
1955	for_each_subsys(ss, ssid)
1956	INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1957
1958	init_waitqueue_head(&cgrp->offline_waitq);
1959	INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent);
1960	}
1961
1962	static void init_cgroup_root(struct cgroup_root *root,
1963	struct cgroup_sb_opts *opts)
1964	{
1965	struct cgroup *cgrp = &root->cgrp;
1966
1967	INIT_LIST_HEAD(&root->root_list);
1968	atomic_set(&root->nr_cgrps, 1);
1969	cgrp->root = root;
1970	init_cgroup_housekeeping(cgrp);
1971	idr_init(&root->cgroup_idr);
1972
1973	root->flags = opts->flags;
1974	if (opts->release_agent)
1975	strcpy(root->release_agent_path, opts->release_agent);
1976	if (opts->name)
1977	strcpy(root->name, opts->name);
1978	if (opts->cpuset_clone_children)
1979	set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1980	}
1981
1982	static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1983	{
1984	LIST_HEAD(tmp_links);
1985	struct cgroup *root_cgrp = &root->cgrp;
1986	struct css_set *cset;
1987	int i, ret;
1988
1989	lockdep_assert_held(&cgroup_mutex);
1990
1991	ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1992	if (ret < 0)
1993	goto out;
1994	root_cgrp->id = ret;
1995	root_cgrp->ancestor_ids[0] = ret;
1996
1997	ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1998	GFP_KERNEL);
1999	if (ret)
2000	goto out;
2001
2002	/*
2003	* We're accessing css_set_count without locking css_set_lock here,
2004	* but that's OK - it can only be increased by someone holding
2005	* cgroup_lock, and that's us. Later rebinding may disable
2006	* controllers on the default hierarchy and thus create new csets,
2007	* which can't be more than the existing ones. Allocate 2x.
2008	*/
2009	ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
2010	if (ret)
2011	goto cancel_ref;
2012
2013	ret = cgroup_init_root_id(root);
2014	if (ret)
2015	goto cancel_ref;
2016
2017	root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
2018	KERNFS_ROOT_CREATE_DEACTIVATED,
2019	root_cgrp);
2020	if (IS_ERR(root->kf_root)) {
2021	ret = PTR_ERR(root->kf_root);
2022	goto exit_root_id;
2023	}
2024	root_cgrp->kn = root->kf_root->kn;
2025
2026	ret = css_populate_dir(&root_cgrp->self);
2027	if (ret)
2028	goto destroy_root;
2029
2030	ret = rebind_subsystems(root, ss_mask);
2031	if (ret)
2032	goto destroy_root;
2033
2034	/*
2035	* There must be no failure case after here, since rebinding takes
2036	* care of subsystems' refcounts, which are explicitly dropped in
2037	* the failure exit path.
2038	*/
2039	list_add(&root->root_list, &cgroup_roots);
2040	cgroup_root_count++;
2041
2042	/*
2043	* Link the root cgroup in this hierarchy into all the css_set
2044	* objects.
2045	*/
2046	spin_lock_irq(&css_set_lock);
2047	hash_for_each(css_set_table, i, cset, hlist) {
2048	link_css_set(&tmp_links, cset, root_cgrp);
2049	if (css_set_populated(cset))
2050	cgroup_update_populated(root_cgrp, true);
2051	}
2052	spin_unlock_irq(&css_set_lock);
2053
2054	BUG_ON(!list_empty(&root_cgrp->self.children));
2055	BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2056
2057	kernfs_activate(root_cgrp->kn);
2058	ret = 0;
2059	goto out;
2060
2061	destroy_root:
2062	kernfs_destroy_root(root->kf_root);
2063	root->kf_root = NULL;
2064	exit_root_id:
2065	cgroup_exit_root_id(root);
2066	cancel_ref:
2067	percpu_ref_exit(&root_cgrp->self.refcnt);
2068	out:
2069	free_cgrp_cset_links(&tmp_links);
2070	return ret;
2071	}
2072
2073	static struct dentry cgroup_mount(struct file_system_type fs_type,
2074	int flags, const char *unused_dev_name,
2075	void *data)
2076	{
2077	bool is_v2 = fs_type == &cgroup2_fs_type;
2078	struct super_block *pinned_sb = NULL;
2079	struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
2080	struct cgroup_subsys *ss;
2081	struct cgroup_root *root;
2082	struct cgroup_sb_opts opts;
2083	struct dentry *dentry;
2084	int ret;
2085	int i;
2086	bool new_sb;
2087
2088	get_cgroup_ns(ns);
2089
2090	/* Check if the caller has permission to mount. */
2091	if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
2092	put_cgroup_ns(ns);
2093	return ERR_PTR(-EPERM);
2094	}
2095
2096	/*
2097	* The first time anyone tries to mount a cgroup, enable the list
2098	* linking each css_set to its tasks and fix up all existing tasks.
2099	*/
2100	if (!use_task_css_set_links)
2101	cgroup_enable_task_cg_lists();
2102
2103	if (is_v2) {
2104	if (data) {
2105	pr_err("cgroup2: unknown option \"%s\"\n", (char *)data);
2106	put_cgroup_ns(ns);
2107	return ERR_PTR(-EINVAL);
2108	}
2109	cgrp_dfl_visible = true;
2110	root = &cgrp_dfl_root;
2111	cgroup_get(&root->cgrp);
2112	goto out_mount;
2113	}
2114
2115	cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
2116
2117	/* First find the desired set of subsystems */
2118	ret = parse_cgroupfs_options(data, &opts);
2119	if (ret)
2120	goto out_unlock;
2121
2122	/*
2123	* Destruction of cgroup root is asynchronous, so subsystems may
2124	* still be dying after the previous unmount. Let's drain the
2125	* dying subsystems. We just need to ensure that the ones
2126	* unmounted previously finish dying and don't care about new ones
2127	* starting. Testing ref liveliness is good enough.
2128	*/
2129	for_each_subsys(ss, i) {
2130	if (!(opts.subsys_mask & (1 << i)) \|\|
2131	ss->root == &cgrp_dfl_root)
2132	continue;
2133
2134	if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) {
2135	mutex_unlock(&cgroup_mutex);
2136	msleep(10);
2137	ret = restart_syscall();
2138	goto out_free;
2139	}
2140	cgroup_put(&ss->root->cgrp);
2141	}
2142
2143	for_each_root(root) {
2144	bool name_match = false;
2145
2146	if (root == &cgrp_dfl_root)
2147	continue;
2148
2149	/*
2150	* If we asked for a name then it must match. Also, if
2151	* name matches but sybsys_mask doesn't, we should fail.
2152	* Remember whether name matched.
2153	*/
2154	if (opts.name) {
2155	if (strcmp(opts.name, root->name))
2156	continue;
2157	name_match = true;
2158	}
2159
2160	/*
2161	* If we asked for subsystems (or explicitly for no
2162	* subsystems) then they must match.
2163	*/
2164	if ((opts.subsys_mask \|\| opts.none) &&
2165	(opts.subsys_mask != root->subsys_mask)) {
2166	if (!name_match)
2167	continue;
2168	ret = -EBUSY;
2169	goto out_unlock;
2170	}
2171
2172	if (root->flags ^ opts.flags)
2173	pr_warn("new mount options do not match the existing superblock, will be ignored\n");
2174
2175	/*
2176	* We want to reuse @root whose lifetime is governed by its
2177	* ->cgrp. Let's check whether @root is alive and keep it
2178	* that way. As cgroup_kill_sb() can happen anytime, we
2179	* want to block it by pinning the sb so that @root doesn't
2180	* get killed before mount is complete.
2181	*
2182	* With the sb pinned, tryget_live can reliably indicate
2183	* whether @root can be reused. If it's being killed,
2184	* drain it. We can use wait_queue for the wait but this
2185	* path is super cold. Let's just sleep a bit and retry.
2186	*/
2187	pinned_sb = kernfs_pin_sb(root->kf_root, NULL);
2188	if (IS_ERR(pinned_sb) \|\|
2189	!percpu_ref_tryget_live(&root->cgrp.self.refcnt)) {
2190	mutex_unlock(&cgroup_mutex);
2191	if (!IS_ERR_OR_NULL(pinned_sb))
2192	deactivate_super(pinned_sb);
2193	msleep(10);
2194	ret = restart_syscall();
2195	goto out_free;
2196	}
2197
2198	ret = 0;
2199	goto out_unlock;
2200	}
2201
2202	/*
2203	* No such thing, create a new one. name= matching without subsys
2204	* specification is allowed for already existing hierarchies but we
2205	* can't create new one without subsys specification.
2206	*/
2207	if (!opts.subsys_mask && !opts.none) {
2208	ret = -EINVAL;
2209	goto out_unlock;
2210	}
2211
2212	/* Hierarchies may only be created in the initial cgroup namespace. */
2213	if (ns != &init_cgroup_ns) {
2214	ret = -EPERM;
2215	goto out_unlock;
2216	}
2217
2218	root = kzalloc(sizeof(*root), GFP_KERNEL);
2219	if (!root) {
2220	ret = -ENOMEM;
2221	goto out_unlock;
2222	}
2223
2224	init_cgroup_root(root, &opts);
2225
2226	ret = cgroup_setup_root(root, opts.subsys_mask);
2227	if (ret)
2228	cgroup_free_root(root);
2229
2230	out_unlock:
2231	mutex_unlock(&cgroup_mutex);
2232	out_free:
2233	kfree(opts.release_agent);
2234	kfree(opts.name);
2235
2236	if (ret) {
2237	put_cgroup_ns(ns);
2238	return ERR_PTR(ret);
2239	}
2240	out_mount:
2241	dentry = kernfs_mount(fs_type, flags, root->kf_root,
2242	is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC,
2243	&new_sb);
2244
2245	/*
2246	* In non-init cgroup namespace, instead of root cgroup's
2247	* dentry, we return the dentry corresponding to the
2248	* cgroupns->root_cgrp.
2249	*/
2250	if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
2251	struct dentry *nsdentry;
2252	struct cgroup *cgrp;
2253
2254	mutex_lock(&cgroup_mutex);
2255	spin_lock_irq(&css_set_lock);
2256
2257	cgrp = cset_cgroup_from_root(ns->root_cset, root);
2258
2259	spin_unlock_irq(&css_set_lock);
2260	mutex_unlock(&cgroup_mutex);
2261
2262	nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
2263	dput(dentry);
2264	dentry = nsdentry;
2265	}
2266
2267	if (IS_ERR(dentry) \|\| !new_sb)
2268	cgroup_put(&root->cgrp);
2269
2270	/*
2271	* If @pinned_sb, we're reusing an existing root and holding an
2272	* extra ref on its sb. Mount is complete. Put the extra ref.
2273	*/
2274	if (pinned_sb) {
2275	WARN_ON(new_sb);
2276	deactivate_super(pinned_sb);
2277	}
2278
2279	put_cgroup_ns(ns);
2280	return dentry;
2281	}
2282
2283	static void cgroup_kill_sb(struct super_block *sb)
2284	{
2285	struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2286	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2287
2288	/*
2289	* If @root doesn't have any mounts or children, start killing it.
2290	* This prevents new mounts by disabling percpu_ref_tryget_live().
2291	* cgroup_mount() may wait for @root's release.
2292	*
2293	* And don't kill the default root.
2294	*/
2295	if (!list_empty(&root->cgrp.self.children) \|\|
2296	root == &cgrp_dfl_root)
2297	cgroup_put(&root->cgrp);
2298	else
2299	percpu_ref_kill(&root->cgrp.self.refcnt);
2300
2301	kernfs_kill_sb(sb);
2302	}
2303
2304	static struct file_system_type cgroup_fs_type = {
2305	.name = "cgroup",
2306	.mount = cgroup_mount,
2307	.kill_sb = cgroup_kill_sb,
2308	.fs_flags = FS_USERNS_MOUNT,
2309	};
2310
2311	static struct file_system_type cgroup2_fs_type = {
2312	.name = "cgroup2",
2313	.mount = cgroup_mount,
2314	.kill_sb = cgroup_kill_sb,
2315	.fs_flags = FS_USERNS_MOUNT,
2316	};
2317
2318	static char cgroup_path_ns_locked(struct cgroup cgrp, char *buf, size_t buflen,
2319	struct cgroup_namespace *ns)
2320	{
2321	struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2322	int ret;
2323
2324	ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2325	if (ret < 0 \|\| ret >= buflen)
2326	return NULL;
2327	return buf;
2328	}
2329
2330	char cgroup_path_ns(struct cgroup cgrp, char *buf, size_t buflen,
2331	struct cgroup_namespace *ns)
2332	{
2333	char *ret;
2334
2335	mutex_lock(&cgroup_mutex);
2336	spin_lock_irq(&css_set_lock);
2337
2338	ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2339
2340	spin_unlock_irq(&css_set_lock);
2341	mutex_unlock(&cgroup_mutex);
2342
2343	return ret;
2344	}
2345	EXPORT_SYMBOL_GPL(cgroup_path_ns);
2346
2347	/**
2348	* task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2349	* @task: target task
2350	* @buf: the buffer to write the path into
2351	* @buflen: the length of the buffer
2352	*
2353	* Determine @task's cgroup on the first (the one with the lowest non-zero
2354	* hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2355	* function grabs cgroup_mutex and shouldn't be used inside locks used by
2356	* cgroup controller callbacks.
2357	*
2358	* Return value is the same as kernfs_path().
2359	*/
2360	char task_cgroup_path(struct task_struct task, char *buf, size_t buflen)
2361	{
2362	struct cgroup_root *root;
2363	struct cgroup *cgrp;
2364	int hierarchy_id = 1;
2365	char *path = NULL;
2366
2367	mutex_lock(&cgroup_mutex);
2368	spin_lock_irq(&css_set_lock);
2369
2370	root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2371
2372	if (root) {
2373	cgrp = task_cgroup_from_root(task, root);
2374	path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2375	} else {
2376	/* if no hierarchy exists, everyone is in "/" */
2377	if (strlcpy(buf, "/", buflen) < buflen)
2378	path = buf;
2379	}
2380
2381	spin_unlock_irq(&css_set_lock);
2382	mutex_unlock(&cgroup_mutex);
2383	return path;
2384	}
2385	EXPORT_SYMBOL_GPL(task_cgroup_path);
2386
2387	/* used to track tasks and other necessary states during migration */
2388	struct cgroup_taskset {
2389	/* the src and dst cset list running through cset->mg_node */
2390	struct list_head src_csets;
2391	struct list_head dst_csets;
2392
2393	/* the subsys currently being processed */
2394	int ssid;
2395
2396	/*
2397	* Fields for cgroup_taskset_*() iteration.
2398	*
2399	* Before migration is committed, the target migration tasks are on
2400	* ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2401	* the csets on ->dst_csets. ->csets point to either ->src_csets
2402	* or ->dst_csets depending on whether migration is committed.
2403	*
2404	* ->cur_csets and ->cur_task point to the current task position
2405	* during iteration.
2406	*/
2407	struct list_head *csets;
2408	struct css_set *cur_cset;
2409	struct task_struct *cur_task;
2410	};
2411
2412	#define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2413	.src_csets = LIST_HEAD_INIT(tset.src_csets), \
2414	.dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2415	.csets = &tset.src_csets, \
2416	}
2417
2418	/**
2419	* cgroup_taskset_add - try to add a migration target task to a taskset
2420	* @task: target task
2421	* @tset: target taskset
2422	*
2423	* Add @task, which is a migration target, to @tset. This function becomes
2424	* noop if @task doesn't need to be migrated. @task's css_set should have
2425	* been added as a migration source and @task->cg_list will be moved from
2426	* the css_set's tasks list to mg_tasks one.
2427	*/
2428	static void cgroup_taskset_add(struct task_struct *task,
2429	struct cgroup_taskset *tset)
2430	{
2431	struct css_set *cset;
2432
2433	lockdep_assert_held(&css_set_lock);
2434
2435	/* @task either already exited or can't exit until the end */
2436	if (task->flags & PF_EXITING)
2437	return;
2438
2439	/* leave @task alone if post_fork() hasn't linked it yet */
2440	if (list_empty(&task->cg_list))
2441	return;
2442
2443	cset = task_css_set(task);
2444	if (!cset->mg_src_cgrp)
2445	return;
2446
2447	list_move_tail(&task->cg_list, &cset->mg_tasks);
2448	if (list_empty(&cset->mg_node))
2449	list_add_tail(&cset->mg_node, &tset->src_csets);
2450	if (list_empty(&cset->mg_dst_cset->mg_node))
2451	list_move_tail(&cset->mg_dst_cset->mg_node,
2452	&tset->dst_csets);
2453	}
2454
2455	/**
2456	* cgroup_taskset_first - reset taskset and return the first task
2457	* @tset: taskset of interest
2458	* @dst_cssp: output variable for the destination css
2459	*
2460	* @tset iteration is initialized and the first task is returned.
2461	*/
2462	struct task_struct cgroup_taskset_first(struct cgroup_taskset tset,
2463	struct cgroup_subsys_state **dst_cssp)
2464	{
2465	tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2466	tset->cur_task = NULL;
2467
2468	return cgroup_taskset_next(tset, dst_cssp);
2469	}
2470
2471	/**
2472	* cgroup_taskset_next - iterate to the next task in taskset
2473	* @tset: taskset of interest
2474	* @dst_cssp: output variable for the destination css
2475	*
2476	* Return the next task in @tset. Iteration must have been initialized
2477	* with cgroup_taskset_first().
2478	*/
2479	struct task_struct cgroup_taskset_next(struct cgroup_taskset tset,
2480	struct cgroup_subsys_state **dst_cssp)
2481	{
2482	struct css_set *cset = tset->cur_cset;
2483	struct task_struct *task = tset->cur_task;
2484
2485	while (&cset->mg_node != tset->csets) {
2486	if (!task)
2487	task = list_first_entry(&cset->mg_tasks,
2488	struct task_struct, cg_list);
2489	else
2490	task = list_next_entry(task, cg_list);
2491
2492	if (&task->cg_list != &cset->mg_tasks) {
2493	tset->cur_cset = cset;
2494	tset->cur_task = task;
2495
2496	/*
2497	* This function may be called both before and
2498	* after cgroup_taskset_migrate(). The two cases
2499	* can be distinguished by looking at whether @cset
2500	* has its ->mg_dst_cset set.
2501	*/
2502	if (cset->mg_dst_cset)
2503	*dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2504	else
2505	*dst_cssp = cset->subsys[tset->ssid];
2506
2507	return task;
2508	}
2509
2510	cset = list_next_entry(cset, mg_node);
2511	task = NULL;
2512	}
2513
2514	return NULL;
2515	}
2516
2517	/**
2518	* cgroup_taskset_migrate - migrate a taskset
2519	* @tset: taget taskset
2520	* @root: cgroup root the migration is taking place on
2521	*
2522	* Migrate tasks in @tset as setup by migration preparation functions.
2523	* This function fails iff one of the ->can_attach callbacks fails and
2524	* guarantees that either all or none of the tasks in @tset are migrated.
2525	* @tset is consumed regardless of success.
2526	*/
2527	static int cgroup_taskset_migrate(struct cgroup_taskset *tset,
2528	struct cgroup_root *root)
2529	{
2530	struct cgroup_subsys *ss;
2531	struct task_struct task, tmp_task;
2532	struct css_set cset, tmp_cset;
2533	int ssid, failed_ssid, ret;
2534
2535	/* methods shouldn't be called if no task is actually migrating */
2536	if (list_empty(&tset->src_csets))
2537	return 0;
2538
2539	/* check that we can legitimately attach to the cgroup */
2540	do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2541	if (ss->can_attach) {
2542	tset->ssid = ssid;
2543	ret = ss->can_attach(tset);
2544	if (ret) {
2545	failed_ssid = ssid;
2546	goto out_cancel_attach;
2547	}
2548	}
2549	} while_each_subsys_mask();
2550
2551	/*
2552	* Now that we're guaranteed success, proceed to move all tasks to
2553	* the new cgroup. There are no failure cases after here, so this
2554	* is the commit point.
2555	*/
2556	spin_lock_irq(&css_set_lock);
2557	list_for_each_entry(cset, &tset->src_csets, mg_node) {
2558	list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2559	struct css_set *from_cset = task_css_set(task);
2560	struct css_set *to_cset = cset->mg_dst_cset;
2561
2562	get_css_set(to_cset);
2563	css_set_move_task(task, from_cset, to_cset, true);
2564	put_css_set_locked(from_cset);
2565	}
2566	}
2567	spin_unlock_irq(&css_set_lock);
2568
2569	/*
2570	* Migration is committed, all target tasks are now on dst_csets.
2571	* Nothing is sensitive to fork() after this point. Notify
2572	* controllers that migration is complete.
2573	*/
2574	tset->csets = &tset->dst_csets;
2575
2576	do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2577	if (ss->attach) {
2578	tset->ssid = ssid;
2579	ss->attach(tset);
2580	}
2581	} while_each_subsys_mask();
2582
2583	ret = 0;
2584	goto out_release_tset;
2585
2586	out_cancel_attach:
2587	do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2588	if (ssid == failed_ssid)
2589	break;
2590	if (ss->cancel_attach) {
2591	tset->ssid = ssid;
2592	ss->cancel_attach(tset);
2593	}
2594	} while_each_subsys_mask();
2595	out_release_tset:
2596	spin_lock_irq(&css_set_lock);
2597	list_splice_init(&tset->dst_csets, &tset->src_csets);
2598	list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2599	list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2600	list_del_init(&cset->mg_node);
2601	}
2602	spin_unlock_irq(&css_set_lock);
2603	return ret;
2604	}
2605
2606	/**
2607	* cgroup_may_migrate_to - verify whether a cgroup can be migration destination
2608	* @dst_cgrp: destination cgroup to test
2609	*
2610	* On the default hierarchy, except for the root, subtree_control must be
2611	* zero for migration destination cgroups with tasks so that child cgroups
2612	* don't compete against tasks.
2613	*/
2614	static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp)
2615	{
2616	return !cgroup_on_dfl(dst_cgrp) \|\| !cgroup_parent(dst_cgrp) \|\|
2617	!dst_cgrp->subtree_control;
2618	}
2619
2620	/**
2621	* cgroup_migrate_finish - cleanup after attach
2622	* @preloaded_csets: list of preloaded css_sets
2623	*
2624	* Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2625	* those functions for details.
2626	*/
2627	static void cgroup_migrate_finish(struct list_head *preloaded_csets)
2628	{
2629	struct css_set cset, tmp_cset;
2630
2631	lockdep_assert_held(&cgroup_mutex);
2632
2633	spin_lock_irq(&css_set_lock);
2634	list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
2635	cset->mg_src_cgrp = NULL;
2636	cset->mg_dst_cgrp = NULL;
2637	cset->mg_dst_cset = NULL;
2638	list_del_init(&cset->mg_preload_node);
2639	put_css_set_locked(cset);
2640	}
2641	spin_unlock_irq(&css_set_lock);
2642	}
2643
2644	/**
2645	* cgroup_migrate_add_src - add a migration source css_set
2646	* @src_cset: the source css_set to add
2647	* @dst_cgrp: the destination cgroup
2648	* @preloaded_csets: list of preloaded css_sets
2649	*
2650	* Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2651	* @src_cset and add it to @preloaded_csets, which should later be cleaned
2652	* up by cgroup_migrate_finish().
2653	*
2654	* This function may be called without holding cgroup_threadgroup_rwsem
2655	* even if the target is a process. Threads may be created and destroyed
2656	* but as long as cgroup_mutex is not dropped, no new css_set can be put
2657	* into play and the preloaded css_sets are guaranteed to cover all
2658	* migrations.
2659	*/
2660	static void cgroup_migrate_add_src(struct css_set *src_cset,
2661	struct cgroup *dst_cgrp,
2662	struct list_head *preloaded_csets)
2663	{
2664	struct cgroup *src_cgrp;
2665
2666	lockdep_assert_held(&cgroup_mutex);
2667	lockdep_assert_held(&css_set_lock);
2668
2669	/*
2670	* If ->dead, @src_set is associated with one or more dead cgroups
2671	* and doesn't contain any migratable tasks. Ignore it early so
2672	* that the rest of migration path doesn't get confused by it.
2673	*/
2674	if (src_cset->dead)
2675	return;
2676
2677	src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2678
2679	if (!list_empty(&src_cset->mg_preload_node))
2680	return;
2681
2682	WARN_ON(src_cset->mg_src_cgrp);
2683	WARN_ON(src_cset->mg_dst_cgrp);
2684	WARN_ON(!list_empty(&src_cset->mg_tasks));
2685	WARN_ON(!list_empty(&src_cset->mg_node));
2686
2687	src_cset->mg_src_cgrp = src_cgrp;
2688	src_cset->mg_dst_cgrp = dst_cgrp;
2689	get_css_set(src_cset);
2690	list_add(&src_cset->mg_preload_node, preloaded_csets);
2691	}
2692
2693	/**
2694	* cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2695	* @preloaded_csets: list of preloaded source css_sets
2696	*
2697	* Tasks are about to be moved and all the source css_sets have been
2698	* preloaded to @preloaded_csets. This function looks up and pins all
2699	* destination css_sets, links each to its source, and append them to
2700	* @preloaded_csets.
2701	*
2702	* This function must be called after cgroup_migrate_add_src() has been
2703	* called on each migration source css_set. After migration is performed
2704	* using cgroup_migrate(), cgroup_migrate_finish() must be called on
2705	* @preloaded_csets.
2706	*/
2707	static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets)
2708	{
2709	LIST_HEAD(csets);
2710	struct css_set src_cset, tmp_cset;
2711
2712	lockdep_assert_held(&cgroup_mutex);
2713
2714	/* look up the dst cset for each src cset and link it to src */
2715	list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) {
2716	struct css_set *dst_cset;
2717
2718	dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2719	if (!dst_cset)
2720	goto err;
2721
2722	WARN_ON_ONCE(src_cset->mg_dst_cset \|\| dst_cset->mg_dst_cset);
2723
2724	/*
2725	* If src cset equals dst, it's noop. Drop the src.
2726	* cgroup_migrate() will skip the cset too. Note that we
2727	* can't handle src == dst as some nodes are used by both.
2728	*/
2729	if (src_cset == dst_cset) {
2730	src_cset->mg_src_cgrp = NULL;
2731	src_cset->mg_dst_cgrp = NULL;
2732	list_del_init(&src_cset->mg_preload_node);
2733	put_css_set(src_cset);
2734	put_css_set(dst_cset);
2735	continue;
2736	}
2737
2738	src_cset->mg_dst_cset = dst_cset;
2739
2740	if (list_empty(&dst_cset->mg_preload_node))
2741	list_add(&dst_cset->mg_preload_node, &csets);
2742	else
2743	put_css_set(dst_cset);
2744	}
2745
2746	list_splice_tail(&csets, preloaded_csets);
2747	return 0;
2748	err:
2749	cgroup_migrate_finish(&csets);
2750	return -ENOMEM;
2751	}
2752
2753	/**
2754	* cgroup_migrate - migrate a process or task to a cgroup
2755	* @leader: the leader of the process or the task to migrate
2756	* @threadgroup: whether @leader points to the whole process or a single task
2757	* @root: cgroup root migration is taking place on
2758	*
2759	* Migrate a process or task denoted by @leader. If migrating a process,
2760	* the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2761	* responsible for invoking cgroup_migrate_add_src() and
2762	* cgroup_migrate_prepare_dst() on the targets before invoking this
2763	* function and following up with cgroup_migrate_finish().
2764	*
2765	* As long as a controller's ->can_attach() doesn't fail, this function is
2766	* guaranteed to succeed. This means that, excluding ->can_attach()
2767	* failure, when migrating multiple targets, the success or failure can be
2768	* decided for all targets by invoking group_migrate_prepare_dst() before
2769	* actually starting migrating.
2770	*/
2771	static int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2772	struct cgroup_root *root)
2773	{
2774	struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2775	struct task_struct *task;
2776
2777	/*
2778	* Prevent freeing of tasks while we take a snapshot. Tasks that are
2779	* already PF_EXITING could be freed from underneath us unless we
2780	* take an rcu_read_lock.
2781	*/
2782	spin_lock_irq(&css_set_lock);
2783	rcu_read_lock();
2784	task = leader;
2785	do {
2786	cgroup_taskset_add(task, &tset);
2787	if (!threadgroup)
2788	break;
2789	} while_each_thread(leader, task);
2790	rcu_read_unlock();
2791	spin_unlock_irq(&css_set_lock);
2792
2793	return cgroup_taskset_migrate(&tset, root);
2794	}
2795
2796	/**
2797	* cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2798	* @dst_cgrp: the cgroup to attach to
2799	* @leader: the task or the leader of the threadgroup to be attached
2800	* @threadgroup: attach the whole threadgroup?
2801	*
2802	* Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2803	*/
2804	static int cgroup_attach_task(struct cgroup *dst_cgrp,
2805	struct task_struct *leader, bool threadgroup)
2806	{
2807	LIST_HEAD(preloaded_csets);
2808	struct task_struct *task;
2809	int ret;
2810
2811	if (!cgroup_may_migrate_to(dst_cgrp))
2812	return -EBUSY;
2813
2814	/* look up all src csets */
2815	spin_lock_irq(&css_set_lock);
2816	rcu_read_lock();
2817	task = leader;
2818	do {
2819	cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2820	&preloaded_csets);
2821	if (!threadgroup)
2822	break;
2823	} while_each_thread(leader, task);
2824	rcu_read_unlock();
2825	spin_unlock_irq(&css_set_lock);
2826
2827	/* prepare dst csets and commit */
2828	ret = cgroup_migrate_prepare_dst(&preloaded_csets);
2829	if (!ret)
2830	ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root);
2831
2832	cgroup_migrate_finish(&preloaded_csets);
2833	return ret;
2834	}
2835
2836	static int cgroup_procs_write_permission(struct task_struct *task,
2837	struct cgroup *dst_cgrp,
2838	struct kernfs_open_file *of)
2839	{
2840	const struct cred *cred = current_cred();
2841	const struct cred *tcred = get_task_cred(task);
2842	int ret = 0;
2843
2844	/*
2845	* even if we're attaching all tasks in the thread group, we only
2846	* need to check permissions on one of them.
2847	*/
2848	if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2849	!uid_eq(cred->euid, tcred->uid) &&
2850	!uid_eq(cred->euid, tcred->suid))
2851	ret = -EACCES;
2852
2853	if (!ret && cgroup_on_dfl(dst_cgrp)) {
2854	struct super_block *sb = of->file->f_path.dentry->d_sb;
2855	struct cgroup *cgrp;
2856	struct inode *inode;
2857
2858	spin_lock_irq(&css_set_lock);
2859	cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
2860	spin_unlock_irq(&css_set_lock);
2861
2862	while (!cgroup_is_descendant(dst_cgrp, cgrp))
2863	cgrp = cgroup_parent(cgrp);
2864
2865	ret = -ENOMEM;
2866	inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
2867	if (inode) {
2868	ret = inode_permission(inode, MAY_WRITE);
2869	iput(inode);
2870	}
2871	}
2872
2873	put_cred(tcred);
2874	return ret;
2875	}
2876
2877	/*
2878	* Find the task_struct of the task to attach by vpid and pass it along to the
2879	* function to attach either it or all tasks in its threadgroup. Will lock
2880	* cgroup_mutex and threadgroup.
2881	*/
2882	static ssize_t __cgroup_procs_write(struct kernfs_open_file of, char buf,
2883	size_t nbytes, loff_t off, bool threadgroup)
2884	{
2885	struct task_struct *tsk;
2886	struct cgroup_subsys *ss;
2887	struct cgroup *cgrp;
2888	pid_t pid;
2889	int ssid, ret;
2890
2891	if (kstrtoint(strstrip(buf), 0, &pid) \|\| pid < 0)
2892	return -EINVAL;
2893
2894	cgrp = cgroup_kn_lock_live(of->kn, false);
2895	if (!cgrp)
2896	return -ENODEV;
2897
2898	percpu_down_write(&cgroup_threadgroup_rwsem);
2899	rcu_read_lock();
2900	if (pid) {
2901	tsk = find_task_by_vpid(pid);
2902	if (!tsk) {
2903	ret = -ESRCH;
2904	goto out_unlock_rcu;
2905	}
2906	} else {
2907	tsk = current;
2908	}
2909
2910	if (threadgroup)
2911	tsk = tsk->group_leader;
2912
2913	/*
2914	* Workqueue threads may acquire PF_NO_SETAFFINITY and become
2915	* trapped in a cpuset, or RT worker may be born in a cgroup
2916	* with no rt_runtime allocated. Just say no.
2917	*/
2918	if (tsk == kthreadd_task \|\| (tsk->flags & PF_NO_SETAFFINITY)) {
2919	ret = -EINVAL;
2920	goto out_unlock_rcu;
2921	}
2922
2923	get_task_struct(tsk);
2924	rcu_read_unlock();
2925
2926	ret = cgroup_procs_write_permission(tsk, cgrp, of);
2927	if (!ret)
2928	ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2929
2930	put_task_struct(tsk);
2931	goto out_unlock_threadgroup;
2932
2933	out_unlock_rcu:
2934	rcu_read_unlock();
2935	out_unlock_threadgroup:
2936	percpu_up_write(&cgroup_threadgroup_rwsem);
2937	for_each_subsys(ss, ssid)
2938	if (ss->post_attach)
2939	ss->post_attach();
2940	cgroup_kn_unlock(of->kn);
2941	return ret ?: nbytes;
2942	}
2943
2944	/**
2945	* cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2946	* @from: attach to all cgroups of a given task
2947	* @tsk: the task to be attached
2948	*/
2949	int cgroup_attach_task_all(struct task_struct from, struct task_struct tsk)
2950	{
2951	struct cgroup_root *root;
2952	int retval = 0;
2953
2954	mutex_lock(&cgroup_mutex);
2955	percpu_down_write(&cgroup_threadgroup_rwsem);
2956	for_each_root(root) {
2957	struct cgroup *from_cgrp;
2958
2959	if (root == &cgrp_dfl_root)
2960	continue;
2961
2962	spin_lock_irq(&css_set_lock);
2963	from_cgrp = task_cgroup_from_root(from, root);
2964	spin_unlock_irq(&css_set_lock);
2965
2966	retval = cgroup_attach_task(from_cgrp, tsk, false);
2967	if (retval)
2968	break;
2969	}
2970	percpu_up_write(&cgroup_threadgroup_rwsem);
2971	mutex_unlock(&cgroup_mutex);
2972
2973	return retval;
2974	}
2975	EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2976
2977	static ssize_t cgroup_tasks_write(struct kernfs_open_file *of,
2978	char *buf, size_t nbytes, loff_t off)
2979	{
2980	return __cgroup_procs_write(of, buf, nbytes, off, false);
2981	}
2982
2983	static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
2984	char *buf, size_t nbytes, loff_t off)
2985	{
2986	return __cgroup_procs_write(of, buf, nbytes, off, true);
2987	}
2988
2989	static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
2990	char *buf, size_t nbytes, loff_t off)
2991	{
2992	struct cgroup *cgrp;
2993
2994	BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2995
2996	cgrp = cgroup_kn_lock_live(of->kn, false);
2997	if (!cgrp)
2998	return -ENODEV;
2999	spin_lock(&release_agent_path_lock);
3000	strlcpy(cgrp->root->release_agent_path, strstrip(buf),
3001	sizeof(cgrp->root->release_agent_path));
3002	spin_unlock(&release_agent_path_lock);
3003	cgroup_kn_unlock(of->kn);
3004	return nbytes;
3005	}
3006
3007	static int cgroup_release_agent_show(struct seq_file seq, void v)
3008	{
3009	struct cgroup *cgrp = seq_css(seq)->cgroup;
3010
3011	spin_lock(&release_agent_path_lock);
3012	seq_puts(seq, cgrp->root->release_agent_path);
3013	spin_unlock(&release_agent_path_lock);
3014	seq_putc(seq, '\n');
3015	return 0;
3016	}
3017
3018	static int cgroup_sane_behavior_show(struct seq_file seq, void v)
3019	{
3020	seq_puts(seq, "0\n");
3021	return 0;
3022	}
3023
3024	static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
3025	{
3026	struct cgroup_subsys *ss;
3027	bool printed = false;
3028	int ssid;
3029
3030	do_each_subsys_mask(ss, ssid, ss_mask) {
3031	if (printed)
3032	seq_putc(seq, ' ');
3033	seq_printf(seq, "%s", ss->name);
3034	printed = true;
3035	} while_each_subsys_mask();
3036	if (printed)
3037	seq_putc(seq, '\n');
3038	}
3039
3040	/* show controllers which are enabled from the parent */
3041	static int cgroup_controllers_show(struct seq_file seq, void v)
3042	{
3043	struct cgroup *cgrp = seq_css(seq)->cgroup;
3044
3045	cgroup_print_ss_mask(seq, cgroup_control(cgrp));
3046	return 0;
3047	}
3048
3049	/* show controllers which are enabled for a given cgroup's children */
3050	static int cgroup_subtree_control_show(struct seq_file seq, void v)
3051	{
3052	struct cgroup *cgrp = seq_css(seq)->cgroup;
3053
3054	cgroup_print_ss_mask(seq, cgrp->subtree_control);
3055	return 0;
3056	}
3057
3058	/**
3059	* cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
3060	* @cgrp: root of the subtree to update csses for
3061	*
3062	* @cgrp's control masks have changed and its subtree's css associations
3063	* need to be updated accordingly. This function looks up all css_sets
3064	* which are attached to the subtree, creates the matching updated css_sets
3065	* and migrates the tasks to the new ones.
3066	*/
3067	static int cgroup_update_dfl_csses(struct cgroup *cgrp)
3068	{
3069	LIST_HEAD(preloaded_csets);
3070	struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
3071	struct cgroup_subsys_state *d_css;
3072	struct cgroup *dsct;
3073	struct css_set *src_cset;
3074	int ret;
3075
3076	lockdep_assert_held(&cgroup_mutex);
3077
3078	percpu_down_write(&cgroup_threadgroup_rwsem);
3079
3080	/* look up all csses currently attached to @cgrp's subtree */
3081	spin_lock_irq(&css_set_lock);
3082	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3083	struct cgrp_cset_link *link;
3084
3085	list_for_each_entry(link, &dsct->cset_links, cset_link)
3086	cgroup_migrate_add_src(link->cset, dsct,
3087	&preloaded_csets);
3088	}
3089	spin_unlock_irq(&css_set_lock);
3090
3091	/* NULL dst indicates self on default hierarchy */
3092	ret = cgroup_migrate_prepare_dst(&preloaded_csets);
3093	if (ret)
3094	goto out_finish;
3095
3096	spin_lock_irq(&css_set_lock);
3097	list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) {
3098	struct task_struct task, ntask;
3099
3100	/* src_csets precede dst_csets, break on the first dst_cset */
3101	if (!src_cset->mg_src_cgrp)
3102	break;
3103
3104	/* all tasks in src_csets need to be migrated */
3105	list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
3106	cgroup_taskset_add(task, &tset);
3107	}
3108	spin_unlock_irq(&css_set_lock);
3109
3110	ret = cgroup_taskset_migrate(&tset, cgrp->root);
3111	out_finish:
3112	cgroup_migrate_finish(&preloaded_csets);
3113	percpu_up_write(&cgroup_threadgroup_rwsem);
3114	return ret;
3115	}
3116
3117	/**
3118	* cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3119	* @cgrp: root of the target subtree
3120	*
3121	* Because css offlining is asynchronous, userland may try to re-enable a
3122	* controller while the previous css is still around. This function grabs
3123	* cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3124	*/
3125	static void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
3126	__acquires(&cgroup_mutex)
3127	{
3128	struct cgroup *dsct;
3129	struct cgroup_subsys_state *d_css;
3130	struct cgroup_subsys *ss;
3131	int ssid;
3132
3133	restart:
3134	mutex_lock(&cgroup_mutex);
3135
3136	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3137	for_each_subsys(ss, ssid) {
3138	struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3139	DEFINE_WAIT(wait);
3140
3141	if (!css \|\| !percpu_ref_is_dying(&css->refcnt))
3142	continue;
3143
3144	cgroup_get(dsct);
3145	prepare_to_wait(&dsct->offline_waitq, &wait,
3146	TASK_UNINTERRUPTIBLE);
3147
3148	mutex_unlock(&cgroup_mutex);
3149	schedule();
3150	finish_wait(&dsct->offline_waitq, &wait);
3151
3152	cgroup_put(dsct);
3153	goto restart;
3154	}
3155	}
3156	}
3157
3158	/**
3159	* cgroup_save_control - save control masks of a subtree
3160	* @cgrp: root of the target subtree
3161	*
3162	* Save ->subtree_control and ->subtree_ss_mask to the respective old_
3163	* prefixed fields for @cgrp's subtree including @cgrp itself.
3164	*/
3165	static void cgroup_save_control(struct cgroup *cgrp)
3166	{
3167	struct cgroup *dsct;
3168	struct cgroup_subsys_state *d_css;
3169
3170	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3171	dsct->old_subtree_control = dsct->subtree_control;
3172	dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3173	}
3174	}
3175
3176	/**
3177	* cgroup_propagate_control - refresh control masks of a subtree
3178	* @cgrp: root of the target subtree
3179	*
3180	* For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3181	* ->subtree_control and propagate controller availability through the
3182	* subtree so that descendants don't have unavailable controllers enabled.
3183	*/
3184	static void cgroup_propagate_control(struct cgroup *cgrp)
3185	{
3186	struct cgroup *dsct;
3187	struct cgroup_subsys_state *d_css;
3188
3189	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3190	dsct->subtree_control &= cgroup_control(dsct);
3191	dsct->subtree_ss_mask =
3192	cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3193	cgroup_ss_mask(dsct));
3194	}
3195	}
3196
3197	/**
3198	* cgroup_restore_control - restore control masks of a subtree
3199	* @cgrp: root of the target subtree
3200	*
3201	* Restore ->subtree_control and ->subtree_ss_mask from the respective old_
3202	* prefixed fields for @cgrp's subtree including @cgrp itself.
3203	*/
3204	static void cgroup_restore_control(struct cgroup *cgrp)
3205	{
3206	struct cgroup *dsct;
3207	struct cgroup_subsys_state *d_css;
3208
3209	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3210	dsct->subtree_control = dsct->old_subtree_control;
3211	dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3212	}
3213	}
3214
3215	static bool css_visible(struct cgroup_subsys_state *css)
3216	{
3217	struct cgroup_subsys *ss = css->ss;
3218	struct cgroup *cgrp = css->cgroup;
3219
3220	if (cgroup_control(cgrp) & (1 << ss->id))
3221	return true;
3222	if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3223	return false;
3224	return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3225	}
3226
3227	/**
3228	* cgroup_apply_control_enable - enable or show csses according to control
3229	* @cgrp: root of the target subtree
3230	*
3231	* Walk @cgrp's subtree and create new csses or make the existing ones
3232	* visible. A css is created invisible if it's being implicitly enabled
3233	* through dependency. An invisible css is made visible when the userland
3234	* explicitly enables it.
3235	*
3236	* Returns 0 on success, -errno on failure. On failure, csses which have
3237	* been processed already aren't cleaned up. The caller is responsible for
3238	* cleaning up with cgroup_apply_control_disble().
3239	*/
3240	static int cgroup_apply_control_enable(struct cgroup *cgrp)
3241	{
3242	struct cgroup *dsct;
3243	struct cgroup_subsys_state *d_css;
3244	struct cgroup_subsys *ss;
3245	int ssid, ret;
3246
3247	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3248	for_each_subsys(ss, ssid) {
3249	struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3250
3251	WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3252
3253	if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3254	continue;
3255
3256	if (!css) {
3257	css = css_create(dsct, ss);
3258	if (IS_ERR(css))
3259	return PTR_ERR(css);
3260	}
3261
3262	if (css_visible(css)) {
3263	ret = css_populate_dir(css);
3264	if (ret)
3265	return ret;
3266	}
3267	}
3268	}
3269
3270	return 0;
3271	}
3272
3273	/**
3274	* cgroup_apply_control_disable - kill or hide csses according to control
3275	* @cgrp: root of the target subtree
3276	*
3277	* Walk @cgrp's subtree and kill and hide csses so that they match
3278	* cgroup_ss_mask() and cgroup_visible_mask().
3279	*
3280	* A css is hidden when the userland requests it to be disabled while other
3281	* subsystems are still depending on it. The css must not actively control
3282	* resources and be in the vanilla state if it's made visible again later.
3283	* Controllers which may be depended upon should provide ->css_reset() for
3284	* this purpose.
3285	*/
3286	static void cgroup_apply_control_disable(struct cgroup *cgrp)
3287	{
3288	struct cgroup *dsct;
3289	struct cgroup_subsys_state *d_css;
3290	struct cgroup_subsys *ss;
3291	int ssid;
3292
3293	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3294	for_each_subsys(ss, ssid) {
3295	struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3296
3297	WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
3298
3299	if (!css)
3300	continue;
3301
3302	if (css->parent &&
3303	!(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3304	kill_css(css);
3305	} else if (!css_visible(css)) {
3306	css_clear_dir(css);
3307	if (ss->css_reset)
3308	ss->css_reset(css);
3309	}
3310	}
3311	}
3312	}
3313
3314	/**
3315	* cgroup_apply_control - apply control mask updates to the subtree
3316	* @cgrp: root of the target subtree
3317	*
3318	* subsystems can be enabled and disabled in a subtree using the following
3319	* steps.
3320	*
3321	* 1. Call cgroup_save_control() to stash the current state.
3322	* 2. Update ->subtree_control masks in the subtree as desired.
3323	* 3. Call cgroup_apply_control() to apply the changes.
3324	* 4. Optionally perform other related operations.
3325	* 5. Call cgroup_finalize_control() to finish up.
3326	*
3327	* This function implements step 3 and propagates the mask changes
3328	* throughout @cgrp's subtree, updates csses accordingly and perform
3329	* process migrations.
3330	*/
3331	static int cgroup_apply_control(struct cgroup *cgrp)
3332	{
3333	int ret;
3334
3335	cgroup_propagate_control(cgrp);
3336
3337	ret = cgroup_apply_control_enable(cgrp);
3338	if (ret)
3339	return ret;
3340
3341	/*
3342	* At this point, cgroup_e_css() results reflect the new csses
3343	* making the following cgroup_update_dfl_csses() properly update
3344	* css associations of all tasks in the subtree.
3345	*/
3346	ret = cgroup_update_dfl_csses(cgrp);
3347	if (ret)
3348	return ret;
3349
3350	return 0;
3351	}
3352
3353	/**
3354	* cgroup_finalize_control - finalize control mask update
3355	* @cgrp: root of the target subtree
3356	* @ret: the result of the update
3357	*
3358	* Finalize control mask update. See cgroup_apply_control() for more info.
3359	*/
3360	static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3361	{
3362	if (ret) {
3363	cgroup_restore_control(cgrp);
3364	cgroup_propagate_control(cgrp);
3365	}
3366
3367	cgroup_apply_control_disable(cgrp);
3368	}
3369
3370	/* change the enabled child controllers for a cgroup in the default hierarchy */
3371	static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3372	char *buf, size_t nbytes,
3373	loff_t off)
3374	{
3375	u16 enable = 0, disable = 0;
3376	struct cgroup cgrp, child;
3377	struct cgroup_subsys *ss;
3378	char *tok;
3379	int ssid, ret;
3380
3381	/*
3382	* Parse input - space separated list of subsystem names prefixed
3383	* with either + or -.
3384	*/
3385	buf = strstrip(buf);
3386	while ((tok = strsep(&buf, " "))) {
3387	if (tok[0] == '\0')
3388	continue;
3389	do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3390	if (!cgroup_ssid_enabled(ssid) \|\|
3391	strcmp(tok + 1, ss->name))
3392	continue;
3393
3394	if (*tok == '+') {
3395	enable \|= 1 << ssid;
3396	disable &= ~(1 << ssid);
3397	} else if (*tok == '-') {
3398	disable \|= 1 << ssid;
3399	enable &= ~(1 << ssid);
3400	} else {
3401	return -EINVAL;
3402	}
3403	break;
3404	} while_each_subsys_mask();
3405	if (ssid == CGROUP_SUBSYS_COUNT)
3406	return -EINVAL;
3407	}
3408
3409	cgrp = cgroup_kn_lock_live(of->kn, true);
3410	if (!cgrp)
3411	return -ENODEV;
3412
3413	for_each_subsys(ss, ssid) {
3414	if (enable & (1 << ssid)) {
3415	if (cgrp->subtree_control & (1 << ssid)) {
3416	enable &= ~(1 << ssid);
3417	continue;
3418	}
3419
3420	if (!(cgroup_control(cgrp) & (1 << ssid))) {
3421	ret = -ENOENT;
3422	goto out_unlock;
3423	}
3424	} else if (disable & (1 << ssid)) {
3425	if (!(cgrp->subtree_control & (1 << ssid))) {
3426	disable &= ~(1 << ssid);
3427	continue;
3428	}
3429
3430	/* a child has it enabled? */
3431	cgroup_for_each_live_child(child, cgrp) {
3432	if (child->subtree_control & (1 << ssid)) {
3433	ret = -EBUSY;
3434	goto out_unlock;
3435	}
3436	}
3437	}
3438	}
3439
3440	if (!enable && !disable) {
3441	ret = 0;
3442	goto out_unlock;
3443	}
3444
3445	/*
3446	* Except for the root, subtree_control must be zero for a cgroup
3447	* with tasks so that child cgroups don't compete against tasks.
3448	*/
3449	if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) {
3450	ret = -EBUSY;
3451	goto out_unlock;
3452	}
3453
3454	/* save and update control masks and prepare csses */
3455	cgroup_save_control(cgrp);
3456
3457	cgrp->subtree_control \|= enable;
3458	cgrp->subtree_control &= ~disable;
3459
3460	ret = cgroup_apply_control(cgrp);
3461
3462	cgroup_finalize_control(cgrp, ret);
3463
3464	kernfs_activate(cgrp->kn);
3465	ret = 0;
3466	out_unlock:
3467	cgroup_kn_unlock(of->kn);
3468	return ret ?: nbytes;
3469	}
3470
3471	static int cgroup_events_show(struct seq_file seq, void v)
3472	{
3473	seq_printf(seq, "populated %d\n",
3474	cgroup_is_populated(seq_css(seq)->cgroup));
3475	return 0;
3476	}
3477
3478	static ssize_t cgroup_file_write(struct kernfs_open_file of, char buf,
3479	size_t nbytes, loff_t off)
3480	{
3481	struct cgroup *cgrp = of->kn->parent->priv;
3482	struct cftype *cft = of->kn->priv;
3483	struct cgroup_subsys_state *css;
3484	int ret;
3485
3486	if (cft->write)
3487	return cft->write(of, buf, nbytes, off);
3488
3489	/*
3490	* kernfs guarantees that a file isn't deleted with operations in
3491	* flight, which means that the matching css is and stays alive and
3492	* doesn't need to be pinned. The RCU locking is not necessary
3493	* either. It's just for the convenience of using cgroup_css().
3494	*/
3495	rcu_read_lock();
3496	css = cgroup_css(cgrp, cft->ss);
3497	rcu_read_unlock();
3498
3499	if (cft->write_u64) {
3500	unsigned long long v;
3501	ret = kstrtoull(buf, 0, &v);
3502	if (!ret)
3503	ret = cft->write_u64(css, cft, v);
3504	} else if (cft->write_s64) {
3505	long long v;
3506	ret = kstrtoll(buf, 0, &v);
3507	if (!ret)
3508	ret = cft->write_s64(css, cft, v);
3509	} else {
3510	ret = -EINVAL;
3511	}
3512
3513	return ret ?: nbytes;
3514	}
3515
3516	static void cgroup_seqfile_start(struct seq_file seq, loff_t *ppos)
3517	{
3518	return seq_cft(seq)->seq_start(seq, ppos);
3519	}
3520
3521	static void cgroup_seqfile_next(struct seq_file seq, void v, loff_t ppos)
3522	{
3523	return seq_cft(seq)->seq_next(seq, v, ppos);
3524	}
3525
3526	static void cgroup_seqfile_stop(struct seq_file seq, void v)
3527	{
3528	seq_cft(seq)->seq_stop(seq, v);
3529	}
3530
3531	static int cgroup_seqfile_show(struct seq_file m, void arg)
3532	{
3533	struct cftype *cft = seq_cft(m);
3534	struct cgroup_subsys_state *css = seq_css(m);
3535
3536	if (cft->seq_show)
3537	return cft->seq_show(m, arg);
3538
3539	if (cft->read_u64)
3540	seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3541	else if (cft->read_s64)
3542	seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3543	else
3544	return -EINVAL;
3545	return 0;
3546	}
3547
3548	static struct kernfs_ops cgroup_kf_single_ops = {
3549	.atomic_write_len = PAGE_SIZE,
3550	.write = cgroup_file_write,
3551	.seq_show = cgroup_seqfile_show,
3552	};
3553
3554	static struct kernfs_ops cgroup_kf_ops = {
3555	.atomic_write_len = PAGE_SIZE,
3556	.write = cgroup_file_write,
3557	.seq_start = cgroup_seqfile_start,
3558	.seq_next = cgroup_seqfile_next,
3559	.seq_stop = cgroup_seqfile_stop,
3560	.seq_show = cgroup_seqfile_show,
3561	};
3562
3563	/*
3564	* cgroup_rename - Only allow simple rename of directories in place.
3565	*/
3566	static int cgroup_rename(struct kernfs_node kn, struct kernfs_node new_parent,
3567	const char *new_name_str)
3568	{
3569	struct cgroup *cgrp = kn->priv;
3570	int ret;
3571
3572	if (kernfs_type(kn) != KERNFS_DIR)
3573	return -ENOTDIR;
3574	if (kn->parent != new_parent)
3575	return -EIO;
3576
3577	/*
3578	* This isn't a proper migration and its usefulness is very
3579	* limited. Disallow on the default hierarchy.
3580	*/
3581	if (cgroup_on_dfl(cgrp))
3582	return -EPERM;
3583
3584	/*
3585	* We're gonna grab cgroup_mutex which nests outside kernfs
3586	* active_ref. kernfs_rename() doesn't require active_ref
3587	* protection. Break them before grabbing cgroup_mutex.
3588	*/
3589	kernfs_break_active_protection(new_parent);
3590	kernfs_break_active_protection(kn);
3591
3592	mutex_lock(&cgroup_mutex);
3593
3594	ret = kernfs_rename(kn, new_parent, new_name_str);
3595
3596	mutex_unlock(&cgroup_mutex);
3597
3598	kernfs_unbreak_active_protection(kn);
3599	kernfs_unbreak_active_protection(new_parent);
3600	return ret;
3601	}
3602
3603	/* set uid and gid of cgroup dirs and files to that of the creator */
3604	static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3605	{
3606	struct iattr iattr = { .ia_valid = ATTR_UID \| ATTR_GID,
3607	.ia_uid = current_fsuid(),
3608	.ia_gid = current_fsgid(), };
3609
3610	if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3611	gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3612	return 0;
3613
3614	return kernfs_setattr(kn, &iattr);
3615	}
3616
3617	static int cgroup_add_file(struct cgroup_subsys_state css, struct cgroup cgrp,
3618	struct cftype *cft)
3619	{
3620	char name[CGROUP_FILE_NAME_MAX];
3621	struct kernfs_node *kn;
3622	struct lock_class_key *key = NULL;
3623	int ret;
3624
3625	#ifdef CONFIG_DEBUG_LOCK_ALLOC
3626	key = &cft->lockdep_key;
3627	#endif
3628	kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3629	cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3630	NULL, key);
3631	if (IS_ERR(kn))
3632	return PTR_ERR(kn);
3633
3634	ret = cgroup_kn_set_ugid(kn);
3635	if (ret) {
3636	kernfs_remove(kn);
3637	return ret;
3638	}
3639
3640	if (cft->file_offset) {
3641	struct cgroup_file cfile = (void )css + cft->file_offset;
3642
3643	spin_lock_irq(&cgroup_file_kn_lock);
3644	cfile->kn = kn;
3645	spin_unlock_irq(&cgroup_file_kn_lock);
3646	}
3647
3648	return 0;
3649	}
3650
3651	/**
3652	* cgroup_addrm_files - add or remove files to a cgroup directory
3653	* @css: the target css
3654	* @cgrp: the target cgroup (usually css->cgroup)
3655	* @cfts: array of cftypes to be added
3656	* @is_add: whether to add or remove
3657	*
3658	* Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3659	* For removals, this function never fails.
3660	*/
3661	static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3662	struct cgroup *cgrp, struct cftype cfts[],
3663	bool is_add)
3664	{
3665	struct cftype cft, cft_end = NULL;
3666	int ret = 0;
3667
3668	lockdep_assert_held(&cgroup_mutex);
3669
3670	restart:
3671	for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3672	/* does cft->flags tell us to skip this file on @cgrp? */
3673	if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3674	continue;
3675	if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3676	continue;
3677	if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3678	continue;
3679	if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3680	continue;
3681
3682	if (is_add) {
3683	ret = cgroup_add_file(css, cgrp, cft);
3684	if (ret) {
3685	pr_warn("%s: failed to add %s, err=%d\n",
3686	__func__, cft->name, ret);
3687	cft_end = cft;
3688	is_add = false;
3689	goto restart;
3690	}
3691	} else {
3692	cgroup_rm_file(cgrp, cft);
3693	}
3694	}
3695	return ret;
3696	}
3697
3698	static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3699	{
3700	LIST_HEAD(pending);
3701	struct cgroup_subsys *ss = cfts[0].ss;
3702	struct cgroup *root = &ss->root->cgrp;
3703	struct cgroup_subsys_state *css;
3704	int ret = 0;
3705
3706	lockdep_assert_held(&cgroup_mutex);
3707
3708	/* add/rm files for all cgroups created before */
3709	css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3710	struct cgroup *cgrp = css->cgroup;
3711
3712	if (!(css->flags & CSS_VISIBLE))
3713	continue;
3714
3715	ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3716	if (ret)
3717	break;
3718	}
3719
3720	if (is_add && !ret)
3721	kernfs_activate(root->kn);
3722	return ret;
3723	}
3724
3725	static void cgroup_exit_cftypes(struct cftype *cfts)
3726	{
3727	struct cftype *cft;
3728
3729	for (cft = cfts; cft->name[0] != '\0'; cft++) {
3730	/* free copy for custom atomic_write_len, see init_cftypes() */
3731	if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3732	kfree(cft->kf_ops);
3733	cft->kf_ops = NULL;
3734	cft->ss = NULL;
3735
3736	/* revert flags set by cgroup core while adding @cfts */
3737	cft->flags &= ~(__CFTYPE_ONLY_ON_DFL \| __CFTYPE_NOT_ON_DFL);
3738	}
3739	}
3740
3741	static int cgroup_init_cftypes(struct cgroup_subsys ss, struct cftype cfts)
3742	{
3743	struct cftype *cft;
3744
3745	for (cft = cfts; cft->name[0] != '\0'; cft++) {
3746	struct kernfs_ops *kf_ops;
3747
3748	WARN_ON(cft->ss \|\| cft->kf_ops);
3749
3750	if (cft->seq_start)
3751	kf_ops = &cgroup_kf_ops;
3752	else
3753	kf_ops = &cgroup_kf_single_ops;
3754
3755	/*
3756	* Ugh... if @cft wants a custom max_write_len, we need to
3757	* make a copy of kf_ops to set its atomic_write_len.
3758	*/
3759	if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3760	kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3761	if (!kf_ops) {
3762	cgroup_exit_cftypes(cfts);
3763	return -ENOMEM;
3764	}
3765	kf_ops->atomic_write_len = cft->max_write_len;
3766	}
3767
3768	cft->kf_ops = kf_ops;
3769	cft->ss = ss;
3770	}
3771
3772	return 0;
3773	}
3774
3775	static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3776	{
3777	lockdep_assert_held(&cgroup_mutex);
3778
3779	if (!cfts \|\| !cfts[0].ss)
3780	return -ENOENT;
3781
3782	list_del(&cfts->node);
3783	cgroup_apply_cftypes(cfts, false);
3784	cgroup_exit_cftypes(cfts);
3785	return 0;
3786	}
3787
3788	/**
3789	* cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3790	* @cfts: zero-length name terminated array of cftypes
3791	*
3792	* Unregister @cfts. Files described by @cfts are removed from all
3793	* existing cgroups and all future cgroups won't have them either. This
3794	* function can be called anytime whether @cfts' subsys is attached or not.
3795	*
3796	* Returns 0 on successful unregistration, -ENOENT if @cfts is not
3797	* registered.
3798	*/
3799	int cgroup_rm_cftypes(struct cftype *cfts)
3800	{
3801	int ret;
3802
3803	mutex_lock(&cgroup_mutex);
3804	ret = cgroup_rm_cftypes_locked(cfts);
3805	mutex_unlock(&cgroup_mutex);
3806	return ret;
3807	}
3808
3809	/**
3810	* cgroup_add_cftypes - add an array of cftypes to a subsystem
3811	* @ss: target cgroup subsystem
3812	* @cfts: zero-length name terminated array of cftypes
3813	*
3814	* Register @cfts to @ss. Files described by @cfts are created for all
3815	* existing cgroups to which @ss is attached and all future cgroups will
3816	* have them too. This function can be called anytime whether @ss is
3817	* attached or not.
3818	*
3819	* Returns 0 on successful registration, -errno on failure. Note that this
3820	* function currently returns 0 as long as @cfts registration is successful
3821	* even if some file creation attempts on existing cgroups fail.
3822	*/
3823	static int cgroup_add_cftypes(struct cgroup_subsys ss, struct cftype cfts)
3824	{
3825	int ret;
3826
3827	if (!cgroup_ssid_enabled(ss->id))
3828	return 0;
3829
3830	if (!cfts \|\| cfts[0].name[0] == '\0')
3831	return 0;
3832
3833	ret = cgroup_init_cftypes(ss, cfts);
3834	if (ret)
3835	return ret;
3836
3837	mutex_lock(&cgroup_mutex);
3838
3839	list_add_tail(&cfts->node, &ss->cfts);
3840	ret = cgroup_apply_cftypes(cfts, true);
3841	if (ret)
3842	cgroup_rm_cftypes_locked(cfts);
3843
3844	mutex_unlock(&cgroup_mutex);
3845	return ret;
3846	}
3847
3848	/**
3849	* cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3850	* @ss: target cgroup subsystem
3851	* @cfts: zero-length name terminated array of cftypes
3852	*
3853	* Similar to cgroup_add_cftypes() but the added files are only used for
3854	* the default hierarchy.
3855	*/
3856	int cgroup_add_dfl_cftypes(struct cgroup_subsys ss, struct cftype cfts)
3857	{
3858	struct cftype *cft;
3859
3860	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3861	cft->flags \|= __CFTYPE_ONLY_ON_DFL;
3862	return cgroup_add_cftypes(ss, cfts);
3863	}
3864
3865	/**
3866	* cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3867	* @ss: target cgroup subsystem
3868	* @cfts: zero-length name terminated array of cftypes
3869	*
3870	* Similar to cgroup_add_cftypes() but the added files are only used for
3871	* the legacy hierarchies.
3872	*/
3873	int cgroup_add_legacy_cftypes(struct cgroup_subsys ss, struct cftype cfts)
3874	{
3875	struct cftype *cft;
3876
3877	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3878	cft->flags \|= __CFTYPE_NOT_ON_DFL;
3879	return cgroup_add_cftypes(ss, cfts);
3880	}
3881
3882	/**
3883	* cgroup_file_notify - generate a file modified event for a cgroup_file
3884	* @cfile: target cgroup_file
3885	*
3886	* @cfile must have been obtained by setting cftype->file_offset.
3887	*/
3888	void cgroup_file_notify(struct cgroup_file *cfile)
3889	{
3890	unsigned long flags;
3891
3892	spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3893	if (cfile->kn)
3894	kernfs_notify(cfile->kn);
3895	spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3896	}
3897
3898	/**
3899	* cgroup_task_count - count the number of tasks in a cgroup.
3900	* @cgrp: the cgroup in question
3901	*
3902	* Return the number of tasks in the cgroup.
3903	*/
3904	static int cgroup_task_count(const struct cgroup *cgrp)
3905	{
3906	int count = 0;
3907	struct cgrp_cset_link *link;
3908
3909	spin_lock_irq(&css_set_lock);
3910	list_for_each_entry(link, &cgrp->cset_links, cset_link)
3911	count += atomic_read(&link->cset->refcount);
3912	spin_unlock_irq(&css_set_lock);
3913	return count;
3914	}
3915
3916	/**
3917	* css_next_child - find the next child of a given css
3918	* @pos: the current position (%NULL to initiate traversal)
3919	* @parent: css whose children to walk
3920	*
3921	* This function returns the next child of @parent and should be called
3922	* under either cgroup_mutex or RCU read lock. The only requirement is
3923	* that @parent and @pos are accessible. The next sibling is guaranteed to
3924	* be returned regardless of their states.
3925	*
3926	* If a subsystem synchronizes ->css_online() and the start of iteration, a
3927	* css which finished ->css_online() is guaranteed to be visible in the
3928	* future iterations and will stay visible until the last reference is put.
3929	* A css which hasn't finished ->css_online() or already finished
3930	* ->css_offline() may show up during traversal. It's each subsystem's
3931	* responsibility to synchronize against on/offlining.
3932	*/
3933	struct cgroup_subsys_state css_next_child(struct cgroup_subsys_state pos,
3934	struct cgroup_subsys_state *parent)
3935	{
3936	struct cgroup_subsys_state *next;
3937
3938	cgroup_assert_mutex_or_rcu_locked();
3939
3940	/*
3941	* @pos could already have been unlinked from the sibling list.
3942	* Once a cgroup is removed, its ->sibling.next is no longer
3943	* updated when its next sibling changes. CSS_RELEASED is set when
3944	* @pos is taken off list, at which time its next pointer is valid,
3945	* and, as releases are serialized, the one pointed to by the next
3946	* pointer is guaranteed to not have started release yet. This
3947	* implies that if we observe !CSS_RELEASED on @pos in this RCU
3948	* critical section, the one pointed to by its next pointer is
3949	* guaranteed to not have finished its RCU grace period even if we
3950	* have dropped rcu_read_lock() inbetween iterations.
3951	*
3952	* If @pos has CSS_RELEASED set, its next pointer can't be
3953	* dereferenced; however, as each css is given a monotonically
3954	* increasing unique serial number and always appended to the
3955	* sibling list, the next one can be found by walking the parent's
3956	* children until the first css with higher serial number than
3957	* @pos's. While this path can be slower, it happens iff iteration
3958	* races against release and the race window is very small.
3959	*/
3960	if (!pos) {
3961	next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3962	} else if (likely(!(pos->flags & CSS_RELEASED))) {
3963	next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
3964	} else {
3965	list_for_each_entry_rcu(next, &parent->children, sibling)
3966	if (next->serial_nr > pos->serial_nr)
3967	break;
3968	}
3969
3970	/*
3971	* @next, if not pointing to the head, can be dereferenced and is
3972	* the next sibling.
3973	*/
3974	if (&next->sibling != &parent->children)
3975	return next;
3976	return NULL;
3977	}
3978
3979	/**
3980	* css_next_descendant_pre - find the next descendant for pre-order walk
3981	* @pos: the current position (%NULL to initiate traversal)
3982	* @root: css whose descendants to walk
3983	*
3984	* To be used by css_for_each_descendant_pre(). Find the next descendant
3985	* to visit for pre-order traversal of @root's descendants. @root is
3986	* included in the iteration and the first node to be visited.
3987	*
3988	* While this function requires cgroup_mutex or RCU read locking, it
3989	* doesn't require the whole traversal to be contained in a single critical
3990	* section. This function will return the correct next descendant as long
3991	* as both @pos and @root are accessible and @pos is a descendant of @root.
3992	*
3993	* If a subsystem synchronizes ->css_online() and the start of iteration, a
3994	* css which finished ->css_online() is guaranteed to be visible in the
3995	* future iterations and will stay visible until the last reference is put.
3996	* A css which hasn't finished ->css_online() or already finished
3997	* ->css_offline() may show up during traversal. It's each subsystem's
3998	* responsibility to synchronize against on/offlining.
3999	*/
4000	struct cgroup_subsys_state *
4001	css_next_descendant_pre(struct cgroup_subsys_state *pos,
4002	struct cgroup_subsys_state *root)
4003	{
4004	struct cgroup_subsys_state *next;
4005
4006	cgroup_assert_mutex_or_rcu_locked();
4007
4008	/* if first iteration, visit @root */
4009	if (!pos)
4010	return root;
4011
4012	/* visit the first child if exists */
4013	next = css_next_child(NULL, pos);
4014	if (next)
4015	return next;
4016
4017	/* no child, visit my or the closest ancestor's next sibling */
4018	while (pos != root) {
4019	next = css_next_child(pos, pos->parent);
4020	if (next)
4021	return next;
4022	pos = pos->parent;
4023	}
4024
4025	return NULL;
4026	}
4027
4028	/**
4029	* css_rightmost_descendant - return the rightmost descendant of a css
4030	* @pos: css of interest
4031	*
4032	* Return the rightmost descendant of @pos. If there's no descendant, @pos
4033	* is returned. This can be used during pre-order traversal to skip
4034	* subtree of @pos.
4035	*
4036	* While this function requires cgroup_mutex or RCU read locking, it
4037	* doesn't require the whole traversal to be contained in a single critical
4038	* section. This function will return the correct rightmost descendant as
4039	* long as @pos is accessible.
4040	*/
4041	struct cgroup_subsys_state *
4042	css_rightmost_descendant(struct cgroup_subsys_state *pos)
4043	{
4044	struct cgroup_subsys_state last, tmp;
4045
4046	cgroup_assert_mutex_or_rcu_locked();
4047
4048	do {
4049	last = pos;
4050	/* ->prev isn't RCU safe, walk ->next till the end */
4051	pos = NULL;
4052	css_for_each_child(tmp, last)
4053	pos = tmp;
4054	} while (pos);
4055
4056	return last;
4057	}
4058
4059	static struct cgroup_subsys_state *
4060	css_leftmost_descendant(struct cgroup_subsys_state *pos)
4061	{
4062	struct cgroup_subsys_state *last;
4063
4064	do {
4065	last = pos;
4066	pos = css_next_child(NULL, pos);
4067	} while (pos);
4068
4069	return last;
4070	}
4071
4072	/**
4073	* css_next_descendant_post - find the next descendant for post-order walk
4074	* @pos: the current position (%NULL to initiate traversal)
4075	* @root: css whose descendants to walk
4076	*
4077	* To be used by css_for_each_descendant_post(). Find the next descendant
4078	* to visit for post-order traversal of @root's descendants. @root is
4079	* included in the iteration and the last node to be visited.
4080	*
4081	* While this function requires cgroup_mutex or RCU read locking, it
4082	* doesn't require the whole traversal to be contained in a single critical
4083	* section. This function will return the correct next descendant as long
4084	* as both @pos and @cgroup are accessible and @pos is a descendant of
4085	* @cgroup.
4086	*
4087	* If a subsystem synchronizes ->css_online() and the start of iteration, a
4088	* css which finished ->css_online() is guaranteed to be visible in the
4089	* future iterations and will stay visible until the last reference is put.
4090	* A css which hasn't finished ->css_online() or already finished
4091	* ->css_offline() may show up during traversal. It's each subsystem's
4092	* responsibility to synchronize against on/offlining.
4093	*/
4094	struct cgroup_subsys_state *
4095	css_next_descendant_post(struct cgroup_subsys_state *pos,
4096	struct cgroup_subsys_state *root)
4097	{
4098	struct cgroup_subsys_state *next;
4099
4100	cgroup_assert_mutex_or_rcu_locked();
4101
4102	/* if first iteration, visit leftmost descendant which may be @root */
4103	if (!pos)
4104	return css_leftmost_descendant(root);
4105
4106	/* if we visited @root, we're done */
4107	if (pos == root)
4108	return NULL;
4109
4110	/* if there's an unvisited sibling, visit its leftmost descendant */
4111	next = css_next_child(pos, pos->parent);
4112	if (next)
4113	return css_leftmost_descendant(next);
4114
4115	/* no sibling left, visit parent */
4116	return pos->parent;
4117	}
4118
4119	/**
4120	* css_has_online_children - does a css have online children
4121	* @css: the target css
4122	*
4123	* Returns %true if @css has any online children; otherwise, %false. This
4124	* function can be called from any context but the caller is responsible
4125	* for synchronizing against on/offlining as necessary.
4126	*/
4127	bool css_has_online_children(struct cgroup_subsys_state *css)
4128	{
4129	struct cgroup_subsys_state *child;
4130	bool ret = false;
4131
4132	rcu_read_lock();
4133	css_for_each_child(child, css) {
4134	if (child->flags & CSS_ONLINE) {
4135	ret = true;
4136	break;
4137	}
4138	}
4139	rcu_read_unlock();
4140	return ret;
4141	}
4142
4143	/**
4144	* css_task_iter_advance_css_set - advance a task itererator to the next css_set
4145	* @it: the iterator to advance
4146	*
4147	* Advance @it to the next css_set to walk.
4148	*/
4149	static void css_task_iter_advance_css_set(struct css_task_iter *it)
4150	{
4151	struct list_head *l = it->cset_pos;
4152	struct cgrp_cset_link *link;
4153	struct css_set *cset;
4154
4155	lockdep_assert_held(&css_set_lock);
4156
4157	/* Advance to the next non-empty css_set */
4158	do {
4159	l = l->next;
4160	if (l == it->cset_head) {
4161	it->cset_pos = NULL;
4162	it->task_pos = NULL;
4163	return;
4164	}
4165
4166	if (it->ss) {
4167	cset = container_of(l, struct css_set,
4168	e_cset_node[it->ss->id]);
4169	} else {
4170	link = list_entry(l, struct cgrp_cset_link, cset_link);
4171	cset = link->cset;
4172	}
4173	} while (!css_set_populated(cset));
4174
4175	it->cset_pos = l;
4176
4177	if (!list_empty(&cset->tasks))
4178	it->task_pos = cset->tasks.next;
4179	else
4180	it->task_pos = cset->mg_tasks.next;
4181
4182	it->tasks_head = &cset->tasks;
4183	it->mg_tasks_head = &cset->mg_tasks;
4184
4185	/*
4186	* We don't keep css_sets locked across iteration steps and thus
4187	* need to take steps to ensure that iteration can be resumed after
4188	* the lock is re-acquired. Iteration is performed at two levels -
4189	* css_sets and tasks in them.
4190	*
4191	* Once created, a css_set never leaves its cgroup lists, so a
4192	* pinned css_set is guaranteed to stay put and we can resume
4193	* iteration afterwards.
4194	*
4195	* Tasks may leave @cset across iteration steps. This is resolved
4196	* by registering each iterator with the css_set currently being
4197	* walked and making css_set_move_task() advance iterators whose
4198	* next task is leaving.
4199	*/
4200	if (it->cur_cset) {
4201	list_del(&it->iters_node);
4202	put_css_set_locked(it->cur_cset);
4203	}
4204	get_css_set(cset);
4205	it->cur_cset = cset;
4206	list_add(&it->iters_node, &cset->task_iters);
4207	}
4208
4209	static void css_task_iter_advance(struct css_task_iter *it)
4210	{
4211	struct list_head *l = it->task_pos;
4212
4213	lockdep_assert_held(&css_set_lock);
4214	WARN_ON_ONCE(!l);
4215
4216	/*
4217	* Advance iterator to find next entry. cset->tasks is consumed
4218	* first and then ->mg_tasks. After ->mg_tasks, we move onto the
4219	* next cset.
4220	*/
4221	l = l->next;
4222
4223	if (l == it->tasks_head)
4224	l = it->mg_tasks_head->next;
4225
4226	if (l == it->mg_tasks_head)
4227	css_task_iter_advance_css_set(it);
4228	else
4229	it->task_pos = l;
4230	}
4231
4232	/**
4233	* css_task_iter_start - initiate task iteration
4234	* @css: the css to walk tasks of
4235	* @it: the task iterator to use
4236	*
4237	* Initiate iteration through the tasks of @css. The caller can call
4238	* css_task_iter_next() to walk through the tasks until the function
4239	* returns NULL. On completion of iteration, css_task_iter_end() must be
4240	* called.
4241	*/
4242	void css_task_iter_start(struct cgroup_subsys_state *css,
4243	struct css_task_iter *it)
4244	{
4245	/* no one should try to iterate before mounting cgroups */
4246	WARN_ON_ONCE(!use_task_css_set_links);
4247
4248	memset(it, 0, sizeof(*it));
4249
4250	spin_lock_irq(&css_set_lock);
4251
4252	it->ss = css->ss;
4253
4254	if (it->ss)
4255	it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4256	else
4257	it->cset_pos = &css->cgroup->cset_links;
4258
4259	it->cset_head = it->cset_pos;
4260
4261	css_task_iter_advance_css_set(it);
4262
4263	spin_unlock_irq(&css_set_lock);
4264	}
4265
4266	/**
4267	* css_task_iter_next - return the next task for the iterator
4268	* @it: the task iterator being iterated
4269	*
4270	* The "next" function for task iteration. @it should have been
4271	* initialized via css_task_iter_start(). Returns NULL when the iteration
4272	* reaches the end.
4273	*/
4274	struct task_struct css_task_iter_next(struct css_task_iter it)
4275	{
4276	if (it->cur_task) {
4277	put_task_struct(it->cur_task);
4278	it->cur_task = NULL;
4279	}
4280
4281	spin_lock_irq(&css_set_lock);
4282
4283	if (it->task_pos) {
4284	it->cur_task = list_entry(it->task_pos, struct task_struct,
4285	cg_list);
4286	get_task_struct(it->cur_task);
4287	css_task_iter_advance(it);
4288	}
4289
4290	spin_unlock_irq(&css_set_lock);
4291
4292	return it->cur_task;
4293	}
4294
4295	/**
4296	* css_task_iter_end - finish task iteration
4297	* @it: the task iterator to finish
4298	*
4299	* Finish task iteration started by css_task_iter_start().
4300	*/
4301	void css_task_iter_end(struct css_task_iter *it)
4302	{
4303	if (it->cur_cset) {
4304	spin_lock_irq(&css_set_lock);
4305	list_del(&it->iters_node);
4306	put_css_set_locked(it->cur_cset);
4307	spin_unlock_irq(&css_set_lock);
4308	}
4309
4310	if (it->cur_task)
4311	put_task_struct(it->cur_task);
4312	}
4313
4314	/**
4315	* cgroup_trasnsfer_tasks - move tasks from one cgroup to another
4316	* @to: cgroup to which the tasks will be moved
4317	* @from: cgroup in which the tasks currently reside
4318	*
4319	* Locking rules between cgroup_post_fork() and the migration path
4320	* guarantee that, if a task is forking while being migrated, the new child
4321	* is guaranteed to be either visible in the source cgroup after the
4322	* parent's migration is complete or put into the target cgroup. No task
4323	* can slip out of migration through forking.
4324	*/
4325	int cgroup_transfer_tasks(struct cgroup to, struct cgroup from)
4326	{
4327	LIST_HEAD(preloaded_csets);
4328	struct cgrp_cset_link *link;
4329	struct css_task_iter it;
4330	struct task_struct *task;
4331	int ret;
4332
4333	if (!cgroup_may_migrate_to(to))
4334	return -EBUSY;
4335
4336	mutex_lock(&cgroup_mutex);
4337
4338	percpu_down_write(&cgroup_threadgroup_rwsem);
4339
4340	/* all tasks in @from are being moved, all csets are source */
4341	spin_lock_irq(&css_set_lock);
4342	list_for_each_entry(link, &from->cset_links, cset_link)
4343	cgroup_migrate_add_src(link->cset, to, &preloaded_csets);
4344	spin_unlock_irq(&css_set_lock);
4345
4346	ret = cgroup_migrate_prepare_dst(&preloaded_csets);
4347	if (ret)
4348	goto out_err;
4349
4350	/*
4351	* Migrate tasks one-by-one until @from is empty. This fails iff
4352	* ->can_attach() fails.
4353	*/
4354	do {
4355	css_task_iter_start(&from->self, &it);
4356	task = css_task_iter_next(&it);
4357	if (task)
4358	get_task_struct(task);
4359	css_task_iter_end(&it);
4360
4361	if (task) {
4362	ret = cgroup_migrate(task, false, to->root);
4363	put_task_struct(task);
4364	}
4365	} while (task && !ret);
4366	out_err:
4367	cgroup_migrate_finish(&preloaded_csets);
4368	percpu_up_write(&cgroup_threadgroup_rwsem);
4369	mutex_unlock(&cgroup_mutex);
4370	return ret;
4371	}
4372
4373	/*
4374	* Stuff for reading the 'tasks'/'procs' files.
4375	*
4376	* Reading this file can return large amounts of data if a cgroup has
4377	* lots of attached tasks. So it may need several calls to read(),
4378	* but we cannot guarantee that the information we produce is correct
4379	* unless we produce it entirely atomically.
4380	*
4381	*/
4382
4383	/* which pidlist file are we talking about? */
4384	enum cgroup_filetype {
4385	CGROUP_FILE_PROCS,
4386	CGROUP_FILE_TASKS,
4387	};
4388
4389	/*
4390	* A pidlist is a list of pids that virtually represents the contents of one
4391	* of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4392	* a pair (one each for procs, tasks) for each pid namespace that's relevant
4393	* to the cgroup.
4394	*/
4395	struct cgroup_pidlist {
4396	/*
4397	* used to find which pidlist is wanted. doesn't change as long as
4398	* this particular list stays in the list.
4399	*/
4400	struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
4401	/* array of xids */
4402	pid_t *list;
4403	/* how many elements the above list has */
4404	int length;
4405	/* each of these stored in a list by its cgroup */
4406	struct list_head links;
4407	/* pointer to the cgroup we belong to, for list removal purposes */
4408	struct cgroup *owner;
4409	/* for delayed destruction */
4410	struct delayed_work destroy_dwork;
4411	};
4412
4413	/*
4414	* The following two functions "fix" the issue where there are more pids
4415	* than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4416	* TODO: replace with a kernel-wide solution to this problem
4417	*/
4418	#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4419	static void *pidlist_allocate(int count)
4420	{
4421	if (PIDLIST_TOO_LARGE(count))
4422	return vmalloc(count * sizeof(pid_t));
4423	else
4424	return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
4425	}
4426
4427	static void pidlist_free(void *p)
4428	{
4429	kvfree(p);
4430	}
4431
4432	/*
4433	* Used to destroy all pidlists lingering waiting for destroy timer. None
4434	* should be left afterwards.
4435	*/
4436	static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
4437	{
4438	struct cgroup_pidlist l, tmp_l;
4439
4440	mutex_lock(&cgrp->pidlist_mutex);
4441	list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
4442	mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
4443	mutex_unlock(&cgrp->pidlist_mutex);
4444
4445	flush_workqueue(cgroup_pidlist_destroy_wq);
4446	BUG_ON(!list_empty(&cgrp->pidlists));
4447	}
4448
4449	static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
4450	{
4451	struct delayed_work *dwork = to_delayed_work(work);
4452	struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
4453	destroy_dwork);
4454	struct cgroup_pidlist *tofree = NULL;
4455
4456	mutex_lock(&l->owner->pidlist_mutex);
4457
4458	/*
4459	* Destroy iff we didn't get queued again. The state won't change
4460	* as destroy_dwork can only be queued while locked.
4461	*/
4462	if (!delayed_work_pending(dwork)) {
4463	list_del(&l->links);
4464	pidlist_free(l->list);
4465	put_pid_ns(l->key.ns);
4466	tofree = l;
4467	}
4468
4469	mutex_unlock(&l->owner->pidlist_mutex);
4470	kfree(tofree);
4471	}
4472
4473	/*
4474	* pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
4475	* Returns the number of unique elements.
4476	*/
4477	static int pidlist_uniq(pid_t *list, int length)
4478	{
4479	int src, dest = 1;
4480
4481	/*
4482	* we presume the 0th element is unique, so i starts at 1. trivial
4483	* edge cases first; no work needs to be done for either
4484	*/
4485	if (length == 0 \|\| length == 1)
4486	return length;
4487	/* src and dest walk down the list; dest counts unique elements */
4488	for (src = 1; src < length; src++) {
4489	/* find next unique element */
4490	while (list[src] == list[src-1]) {
4491	src++;
4492	if (src == length)
4493	goto after;
4494	}
4495	/* dest always points to where the next unique element goes */
4496	list[dest] = list[src];
4497	dest++;
4498	}
4499	after:
4500	return dest;
4501	}
4502
4503	/*
4504	* The two pid files - task and cgroup.procs - guaranteed that the result
4505	* is sorted, which forced this whole pidlist fiasco. As pid order is
4506	* different per namespace, each namespace needs differently sorted list,
4507	* making it impossible to use, for example, single rbtree of member tasks
4508	* sorted by task pointer. As pidlists can be fairly large, allocating one
4509	* per open file is dangerous, so cgroup had to implement shared pool of
4510	* pidlists keyed by cgroup and namespace.
4511	*
4512	* All this extra complexity was caused by the original implementation
4513	* committing to an entirely unnecessary property. In the long term, we
4514	* want to do away with it. Explicitly scramble sort order if on the
4515	* default hierarchy so that no such expectation exists in the new
4516	* interface.
4517	*
4518	* Scrambling is done by swapping every two consecutive bits, which is
4519	* non-identity one-to-one mapping which disturbs sort order sufficiently.
4520	*/
4521	static pid_t pid_fry(pid_t pid)
4522	{
4523	unsigned a = pid & 0x55555555;
4524	unsigned b = pid & 0xAAAAAAAA;
4525
4526	return (a << 1) \| (b >> 1);
4527	}
4528
4529	static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
4530	{
4531	if (cgroup_on_dfl(cgrp))
4532	return pid_fry(pid);
4533	else
4534	return pid;
4535	}
4536
4537	static int cmppid(const void a, const void b)
4538	{
4539	return (pid_t )a - (pid_t )b;
4540	}
4541
4542	static int fried_cmppid(const void a, const void b)
4543	{
4544	return pid_fry((pid_t )a) - pid_fry((pid_t )b);
4545	}
4546
4547	static struct cgroup_pidlist cgroup_pidlist_find(struct cgroup cgrp,
4548	enum cgroup_filetype type)
4549	{
4550	struct cgroup_pidlist *l;
4551	/* don't need task_nsproxy() if we're looking at ourself */
4552	struct pid_namespace *ns = task_active_pid_ns(current);
4553
4554	lockdep_assert_held(&cgrp->pidlist_mutex);
4555
4556	list_for_each_entry(l, &cgrp->pidlists, links)
4557	if (l->key.type == type && l->key.ns == ns)
4558	return l;
4559	return NULL;
4560	}
4561
4562	/*
4563	* find the appropriate pidlist for our purpose (given procs vs tasks)
4564	* returns with the lock on that pidlist already held, and takes care
4565	* of the use count, or returns NULL with no locks held if we're out of
4566	* memory.
4567	*/
4568	static struct cgroup_pidlist cgroup_pidlist_find_create(struct cgroup cgrp,
4569	enum cgroup_filetype type)
4570	{
4571	struct cgroup_pidlist *l;
4572
4573	lockdep_assert_held(&cgrp->pidlist_mutex);
4574
4575	l = cgroup_pidlist_find(cgrp, type);
4576	if (l)
4577	return l;
4578
4579	/* entry not found; create a new one */
4580	l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
4581	if (!l)
4582	return l;
4583
4584	INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
4585	l->key.type = type;
4586	/* don't need task_nsproxy() if we're looking at ourself */
4587	l->key.ns = get_pid_ns(task_active_pid_ns(current));
4588	l->owner = cgrp;
4589	list_add(&l->links, &cgrp->pidlists);
4590	return l;
4591	}
4592
4593	/*
4594	* Load a cgroup's pidarray with either procs' tgids or tasks' pids
4595	*/
4596	static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
4597	struct cgroup_pidlist **lp)
4598	{
4599	pid_t *array;
4600	int length;
4601	int pid, n = 0; /* used for populating the array */
4602	struct css_task_iter it;
4603	struct task_struct *tsk;
4604	struct cgroup_pidlist *l;
4605
4606	lockdep_assert_held(&cgrp->pidlist_mutex);
4607
4608	/*
4609	* If cgroup gets more users after we read count, we won't have
4610	* enough space - tough. This race is indistinguishable to the
4611	* caller from the case that the additional cgroup users didn't
4612	* show up until sometime later on.
4613	*/
4614	length = cgroup_task_count(cgrp);
4615	array = pidlist_allocate(length);
4616	if (!array)
4617	return -ENOMEM;
4618	/* now, populate the array */
4619	css_task_iter_start(&cgrp->self, &it);
4620	while ((tsk = css_task_iter_next(&it))) {
4621	if (unlikely(n == length))
4622	break;
4623	/* get tgid or pid for procs or tasks file respectively */
4624	if (type == CGROUP_FILE_PROCS)
4625	pid = task_tgid_vnr(tsk);
4626	else
4627	pid = task_pid_vnr(tsk);
4628	if (pid > 0) /* make sure to only use valid results */
4629	array[n++] = pid;
4630	}
4631	css_task_iter_end(&it);
4632	length = n;
4633	/* now sort & (if procs) strip out duplicates */
4634	if (cgroup_on_dfl(cgrp))
4635	sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
4636	else
4637	sort(array, length, sizeof(pid_t), cmppid, NULL);
4638	if (type == CGROUP_FILE_PROCS)
4639	length = pidlist_uniq(array, length);
4640
4641	l = cgroup_pidlist_find_create(cgrp, type);
4642	if (!l) {
4643	pidlist_free(array);
4644	return -ENOMEM;
4645	}
4646
4647	/* store array, freeing old if necessary */
4648	pidlist_free(l->list);
4649	l->list = array;
4650	l->length = length;
4651	*lp = l;
4652	return 0;
4653	}
4654
4655	/**
4656	* cgroupstats_build - build and fill cgroupstats
4657	* @stats: cgroupstats to fill information into
4658	* @dentry: A dentry entry belonging to the cgroup for which stats have
4659	* been requested.
4660	*
4661	* Build and fill cgroupstats so that taskstats can export it to user
4662	* space.
4663	*/
4664	int cgroupstats_build(struct cgroupstats stats, struct dentry dentry)
4665	{
4666	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
4667	struct cgroup *cgrp;
4668	struct css_task_iter it;
4669	struct task_struct *tsk;
4670
4671	/* it should be kernfs_node belonging to cgroupfs and is a directory */
4672	if (dentry->d_sb->s_type != &cgroup_fs_type \|\| !kn \|\|
4673	kernfs_type(kn) != KERNFS_DIR)
4674	return -EINVAL;
4675
4676	mutex_lock(&cgroup_mutex);
4677
4678	/*
4679	* We aren't being called from kernfs and there's no guarantee on
4680	* @kn->priv's validity. For this and css_tryget_online_from_dir(),
4681	* @kn->priv is RCU safe. Let's do the RCU dancing.
4682	*/
4683	rcu_read_lock();
4684	cgrp = rcu_dereference(kn->priv);
4685	if (!cgrp \|\| cgroup_is_dead(cgrp)) {
4686	rcu_read_unlock();
4687	mutex_unlock(&cgroup_mutex);
4688	return -ENOENT;
4689	}
4690	rcu_read_unlock();
4691
4692	css_task_iter_start(&cgrp->self, &it);
4693	while ((tsk = css_task_iter_next(&it))) {
4694	switch (tsk->state) {
4695	case TASK_RUNNING:
4696	stats->nr_running++;
4697	break;
4698	case TASK_INTERRUPTIBLE:
4699	stats->nr_sleeping++;
4700	break;
4701	case TASK_UNINTERRUPTIBLE:
4702	stats->nr_uninterruptible++;
4703	break;
4704	case TASK_STOPPED:
4705	stats->nr_stopped++;
4706	break;
4707	default:
4708	if (delayacct_is_task_waiting_on_io(tsk))
4709	stats->nr_io_wait++;
4710	break;
4711	}
4712	}
4713	css_task_iter_end(&it);
4714
4715	mutex_unlock(&cgroup_mutex);
4716	return 0;
4717	}
4718
4719
4720	/*
4721	* seq_file methods for the tasks/procs files. The seq_file position is the
4722	* next pid to display; the seq_file iterator is a pointer to the pid
4723	* in the cgroup->l->list array.
4724	*/
4725
4726	static void cgroup_pidlist_start(struct seq_file s, loff_t *pos)
4727	{
4728	/*
4729	* Initially we receive a position value that corresponds to
4730	* one more than the last pid shown (or 0 on the first call or
4731	* after a seek to the start). Use a binary-search to find the
4732	* next pid to display, if any
4733	*/
4734	struct kernfs_open_file *of = s->private;
4735	struct cgroup *cgrp = seq_css(s)->cgroup;
4736	struct cgroup_pidlist *l;
4737	enum cgroup_filetype type = seq_cft(s)->private;
4738	int index = 0, pid = *pos;
4739	int *iter, ret;
4740
4741	mutex_lock(&cgrp->pidlist_mutex);
4742
4743	/*
4744	* !NULL @of->priv indicates that this isn't the first start()
4745	* after open. If the matching pidlist is around, we can use that.
4746	* Look for it. Note that @of->priv can't be used directly. It
4747	* could already have been destroyed.
4748	*/
4749	if (of->priv)
4750	of->priv = cgroup_pidlist_find(cgrp, type);
4751
4752	/*
4753	* Either this is the first start() after open or the matching
4754	* pidlist has been destroyed inbetween. Create a new one.
4755	*/
4756	if (!of->priv) {
4757	ret = pidlist_array_load(cgrp, type,
4758	(struct cgroup_pidlist **)&of->priv);
4759	if (ret)
4760	return ERR_PTR(ret);
4761	}
4762	l = of->priv;
4763
4764	if (pid) {
4765	int end = l->length;
4766
4767	while (index < end) {
4768	int mid = (index + end) / 2;
4769	if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
4770	index = mid;
4771	break;
4772	} else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
4773	index = mid + 1;
4774	else
4775	end = mid;
4776	}
4777	}
4778	/* If we're off the end of the array, we're done */
4779	if (index >= l->length)
4780	return NULL;
4781	/* Update the abstract position to be the actual pid that we found */
4782	iter = l->list + index;
4783	pos = cgroup_pid_fry(cgrp, iter);
4784	return iter;
4785	}
4786
4787	static void cgroup_pidlist_stop(struct seq_file s, void v)
4788	{
4789	struct kernfs_open_file *of = s->private;
4790	struct cgroup_pidlist *l = of->priv;
4791
4792	if (l)
4793	mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
4794	CGROUP_PIDLIST_DESTROY_DELAY);
4795	mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
4796	}
4797
4798	static void cgroup_pidlist_next(struct seq_file s, void v, loff_t pos)
4799	{
4800	struct kernfs_open_file *of = s->private;
4801	struct cgroup_pidlist *l = of->priv;
4802	pid_t *p = v;
4803	pid_t *end = l->list + l->length;
4804	/*
4805	* Advance to the next pid in the array. If this goes off the
4806	* end, we're done
4807	*/
4808	p++;
4809	if (p >= end) {
4810	return NULL;
4811	} else {
4812	pos = cgroup_pid_fry(seq_css(s)->cgroup, p);
4813	return p;
4814	}
4815	}
4816
4817	static int cgroup_pidlist_show(struct seq_file s, void v)
4818	{
4819	seq_printf(s, "%d\n", (int )v);
4820
4821	return 0;
4822	}
4823
4824	static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
4825	struct cftype *cft)
4826	{
4827	return notify_on_release(css->cgroup);
4828	}
4829
4830	static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
4831	struct cftype *cft, u64 val)
4832	{
4833	if (val)
4834	set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
4835	else
4836	clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
4837	return 0;
4838	}
4839
4840	static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
4841	struct cftype *cft)
4842	{
4843	return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4844	}
4845
4846	static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
4847	struct cftype *cft, u64 val)
4848	{
4849	if (val)
4850	set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4851	else
4852	clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4853	return 0;
4854	}
4855
4856	/* cgroup core interface files for the default hierarchy */
4857	static struct cftype cgroup_dfl_base_files[] = {
4858	{
4859	.name = "cgroup.procs",
4860	.file_offset = offsetof(struct cgroup, procs_file),
4861	.seq_start = cgroup_pidlist_start,
4862	.seq_next = cgroup_pidlist_next,
4863	.seq_stop = cgroup_pidlist_stop,
4864	.seq_show = cgroup_pidlist_show,
4865	.private = CGROUP_FILE_PROCS,
4866	.write = cgroup_procs_write,
4867	},
4868	{
4869	.name = "cgroup.controllers",
4870	.seq_show = cgroup_controllers_show,
4871	},
4872	{
4873	.name = "cgroup.subtree_control",
4874	.seq_show = cgroup_subtree_control_show,
4875	.write = cgroup_subtree_control_write,
4876	},
4877	{
4878	.name = "cgroup.events",
4879	.flags = CFTYPE_NOT_ON_ROOT,
4880	.file_offset = offsetof(struct cgroup, events_file),
4881	.seq_show = cgroup_events_show,
4882	},
4883	{ } /* terminate */
4884	};
4885
4886	/* cgroup core interface files for the legacy hierarchies */
4887	static struct cftype cgroup_legacy_base_files[] = {
4888	{
4889	.name = "cgroup.procs",
4890	.seq_start = cgroup_pidlist_start,
4891	.seq_next = cgroup_pidlist_next,
4892	.seq_stop = cgroup_pidlist_stop,
4893	.seq_show = cgroup_pidlist_show,
4894	.private = CGROUP_FILE_PROCS,
4895	.write = cgroup_procs_write,
4896	},
4897	{
4898	.name = "cgroup.clone_children",
4899	.read_u64 = cgroup_clone_children_read,
4900	.write_u64 = cgroup_clone_children_write,
4901	},
4902	{
4903	.name = "cgroup.sane_behavior",
4904	.flags = CFTYPE_ONLY_ON_ROOT,
4905	.seq_show = cgroup_sane_behavior_show,
4906	},
4907	{
4908	.name = "tasks",
4909	.seq_start = cgroup_pidlist_start,
4910	.seq_next = cgroup_pidlist_next,
4911	.seq_stop = cgroup_pidlist_stop,
4912	.seq_show = cgroup_pidlist_show,
4913	.private = CGROUP_FILE_TASKS,
4914	.write = cgroup_tasks_write,
4915	},
4916	{
4917	.name = "notify_on_release",
4918	.read_u64 = cgroup_read_notify_on_release,
4919	.write_u64 = cgroup_write_notify_on_release,
4920	},
4921	{
4922	.name = "release_agent",
4923	.flags = CFTYPE_ONLY_ON_ROOT,
4924	.seq_show = cgroup_release_agent_show,
4925	.write = cgroup_release_agent_write,
4926	.max_write_len = PATH_MAX - 1,
4927	},
4928	{ } /* terminate */
4929	};
4930
4931	/*
4932	* css destruction is four-stage process.
4933	*
4934	* 1. Destruction starts. Killing of the percpu_ref is initiated.
4935	* Implemented in kill_css().
4936	*
4937	* 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4938	* and thus css_tryget_online() is guaranteed to fail, the css can be
4939	* offlined by invoking offline_css(). After offlining, the base ref is
4940	* put. Implemented in css_killed_work_fn().
4941	*
4942	* 3. When the percpu_ref reaches zero, the only possible remaining
4943	* accessors are inside RCU read sections. css_release() schedules the
4944	* RCU callback.
4945	*
4946	* 4. After the grace period, the css can be freed. Implemented in
4947	* css_free_work_fn().
4948	*
4949	* It is actually hairier because both step 2 and 4 require process context
4950	* and thus involve punting to css->destroy_work adding two additional
4951	* steps to the already complex sequence.
4952	*/
4953	static void css_free_work_fn(struct work_struct *work)
4954	{
4955	struct cgroup_subsys_state *css =
4956	container_of(work, struct cgroup_subsys_state, destroy_work);
4957	struct cgroup_subsys *ss = css->ss;
4958	struct cgroup *cgrp = css->cgroup;
4959
4960	percpu_ref_exit(&css->refcnt);
4961
4962	if (ss) {
4963	/* css free path */
4964	struct cgroup_subsys_state *parent = css->parent;
4965	int id = css->id;
4966
4967	ss->css_free(css);
4968	cgroup_idr_remove(&ss->css_idr, id);
4969	cgroup_put(cgrp);
4970
4971	if (parent)
4972	css_put(parent);
4973	} else {
4974	/* cgroup free path */
4975	atomic_dec(&cgrp->root->nr_cgrps);
4976	cgroup_pidlist_destroy_all(cgrp);
4977	cancel_work_sync(&cgrp->release_agent_work);
4978
4979	if (cgroup_parent(cgrp)) {
4980	/*
4981	* We get a ref to the parent, and put the ref when
4982	* this cgroup is being freed, so it's guaranteed
4983	* that the parent won't be destroyed before its
4984	* children.
4985	*/
4986	cgroup_put(cgroup_parent(cgrp));
4987	kernfs_put(cgrp->kn);
4988	kfree(cgrp);
4989	} else {
4990	/*
4991	* This is root cgroup's refcnt reaching zero,
4992	* which indicates that the root should be
4993	* released.
4994	*/
4995	cgroup_destroy_root(cgrp->root);
4996	}
4997	}
4998	}
4999
5000	static void css_free_rcu_fn(struct rcu_head *rcu_head)
5001	{
5002	struct cgroup_subsys_state *css =
5003	container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
5004
5005	INIT_WORK(&css->destroy_work, css_free_work_fn);
5006	queue_work(cgroup_destroy_wq, &css->destroy_work);
5007	}
5008
5009	static void css_release_work_fn(struct work_struct *work)
5010	{
5011	struct cgroup_subsys_state *css =
5012	container_of(work, struct cgroup_subsys_state, destroy_work);
5013	struct cgroup_subsys *ss = css->ss;
5014	struct cgroup *cgrp = css->cgroup;
5015
5016	mutex_lock(&cgroup_mutex);
5017
5018	css->flags \|= CSS_RELEASED;
5019	list_del_rcu(&css->sibling);
5020
5021	if (ss) {
5022	/* css release path */
5023	cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5024	if (ss->css_released)
5025	ss->css_released(css);
5026	} else {
5027	/* cgroup release path */
5028	cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
5029	cgrp->id = -1;
5030
5031	/*
5032	* There are two control paths which try to determine
5033	* cgroup from dentry without going through kernfs -
5034	* cgroupstats_build() and css_tryget_online_from_dir().
5035	* Those are supported by RCU protecting clearing of
5036	* cgrp->kn->priv backpointer.
5037	*/
5038	if (cgrp->kn)
5039	RCU_INIT_POINTER((void __rcu __force *)&cgrp->kn->priv,
5040	NULL);
5041	}
5042
5043	mutex_unlock(&cgroup_mutex);
5044
5045	call_rcu(&css->rcu_head, css_free_rcu_fn);
5046	}
5047
5048	static void css_release(struct percpu_ref *ref)
5049	{
5050	struct cgroup_subsys_state *css =
5051	container_of(ref, struct cgroup_subsys_state, refcnt);
5052
5053	INIT_WORK(&css->destroy_work, css_release_work_fn);
5054	queue_work(cgroup_destroy_wq, &css->destroy_work);
5055	}
5056
5057	static void init_and_link_css(struct cgroup_subsys_state *css,
5058	struct cgroup_subsys ss, struct cgroup cgrp)
5059	{
5060	lockdep_assert_held(&cgroup_mutex);
5061
5062	cgroup_get(cgrp);
5063
5064	memset(css, 0, sizeof(*css));
5065	css->cgroup = cgrp;
5066	css->ss = ss;
5067	css->id = -1;
5068	INIT_LIST_HEAD(&css->sibling);
5069	INIT_LIST_HEAD(&css->children);
5070	css->serial_nr = css_serial_nr_next++;
5071	atomic_set(&css->online_cnt, 0);
5072
5073	if (cgroup_parent(cgrp)) {
5074	css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5075	css_get(css->parent);
5076	}
5077
5078	BUG_ON(cgroup_css(cgrp, ss));
5079	}
5080
5081	/* invoke ->css_online() on a new CSS and mark it online if successful */
5082	static int online_css(struct cgroup_subsys_state *css)
5083	{
5084	struct cgroup_subsys *ss = css->ss;
5085	int ret = 0;
5086
5087	lockdep_assert_held(&cgroup_mutex);
5088
5089	if (ss->css_online)
5090	ret = ss->css_online(css);
5091	if (!ret) {
5092	css->flags \|= CSS_ONLINE;
5093	rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5094
5095	atomic_inc(&css->online_cnt);
5096	if (css->parent)
5097	atomic_inc(&css->parent->online_cnt);
5098	}
5099	return ret;
5100	}
5101
5102	/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5103	static void offline_css(struct cgroup_subsys_state *css)
5104	{
5105	struct cgroup_subsys *ss = css->ss;
5106
5107	lockdep_assert_held(&cgroup_mutex);
5108
5109	if (!(css->flags & CSS_ONLINE))
5110	return;
5111
5112	if (ss->css_reset)
5113	ss->css_reset(css);
5114
5115	if (ss->css_offline)
5116	ss->css_offline(css);
5117
5118	css->flags &= ~CSS_ONLINE;
5119	RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5120
5121	wake_up_all(&css->cgroup->offline_waitq);
5122	}
5123
5124	/**
5125	* css_create - create a cgroup_subsys_state
5126	* @cgrp: the cgroup new css will be associated with
5127	* @ss: the subsys of new css
5128	*
5129	* Create a new css associated with @cgrp - @ss pair. On success, the new
5130	* css is online and installed in @cgrp. This function doesn't create the
5131	* interface files. Returns 0 on success, -errno on failure.
5132	*/
5133	static struct cgroup_subsys_state css_create(struct cgroup cgrp,
5134	struct cgroup_subsys *ss)
5135	{
5136	struct cgroup *parent = cgroup_parent(cgrp);
5137	struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5138	struct cgroup_subsys_state *css;
5139	int err;
5140
5141	lockdep_assert_held(&cgroup_mutex);
5142
5143	css = ss->css_alloc(parent_css);
5144	if (!css)
5145	css = ERR_PTR(-ENOMEM);
5146	if (IS_ERR(css))
5147	return css;
5148
5149	init_and_link_css(css, ss, cgrp);
5150
5151	err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5152	if (err)
5153	goto err_free_css;
5154
5155	err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5156	if (err < 0)
5157	goto err_free_css;
5158	css->id = err;
5159
5160	/* @css is ready to be brought online now, make it visible */
5161	list_add_tail_rcu(&css->sibling, &parent_css->children);
5162	cgroup_idr_replace(&ss->css_idr, css, css->id);
5163
5164	err = online_css(css);
5165	if (err)
5166	goto err_list_del;
5167
5168	if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
5169	cgroup_parent(parent)) {
5170	pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5171	current->comm, current->pid, ss->name);
5172	if (!strcmp(ss->name, "memory"))
5173	pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5174	ss->warned_broken_hierarchy = true;
5175	}
5176
5177	return css;
5178
5179	err_list_del:
5180	list_del_rcu(&css->sibling);
5181	err_free_css:
5182	call_rcu(&css->rcu_head, css_free_rcu_fn);
5183	return ERR_PTR(err);
5184	}
5185
5186	static struct cgroup cgroup_create(struct cgroup parent)
5187	{
5188	struct cgroup_root *root = parent->root;
5189	struct cgroup cgrp, tcgrp;
5190	int level = parent->level + 1;
5191	int ret;
5192
5193	/* allocate the cgroup and its ID, 0 is reserved for the root */
5194	cgrp = kzalloc(sizeof(*cgrp) +
5195	sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
5196	if (!cgrp)
5197	return ERR_PTR(-ENOMEM);
5198
5199	ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5200	if (ret)
5201	goto out_free_cgrp;
5202
5203	/*
5204	* Temporarily set the pointer to NULL, so idr_find() won't return
5205	* a half-baked cgroup.
5206	*/
5207	cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
5208	if (cgrp->id < 0) {
5209	ret = -ENOMEM;
5210	goto out_cancel_ref;
5211	}
5212
5213	init_cgroup_housekeeping(cgrp);
5214
5215	cgrp->self.parent = &parent->self;
5216	cgrp->root = root;
5217	cgrp->level = level;
5218
5219	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp))
5220	cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
5221
5222	if (notify_on_release(parent))
5223	set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5224
5225	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5226	set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5227
5228	cgrp->self.serial_nr = css_serial_nr_next++;
5229
5230	/* allocation complete, commit to creation */
5231	list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5232	atomic_inc(&root->nr_cgrps);
5233	cgroup_get(parent);
5234
5235	/*
5236	* @cgrp is now fully operational. If something fails after this
5237	* point, it'll be released via the normal destruction path.
5238	*/
5239	cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
5240
5241	/*
5242	* On the default hierarchy, a child doesn't automatically inherit
5243	* subtree_control from the parent. Each is configured manually.
5244	*/
5245	if (!cgroup_on_dfl(cgrp))
5246	cgrp->subtree_control = cgroup_control(cgrp);
5247
5248	cgroup_propagate_control(cgrp);
5249
5250	/* @cgrp doesn't have dir yet so the following will only create csses */
5251	ret = cgroup_apply_control_enable(cgrp);
5252	if (ret)
5253	goto out_destroy;
5254
5255	return cgrp;
5256
5257	out_cancel_ref:
5258	percpu_ref_exit(&cgrp->self.refcnt);
5259	out_free_cgrp:
5260	kfree(cgrp);
5261	return ERR_PTR(ret);
5262	out_destroy:
5263	cgroup_destroy_locked(cgrp);
5264	return ERR_PTR(ret);
5265	}
5266
5267	static int cgroup_mkdir(struct kernfs_node parent_kn, const char name,
5268	umode_t mode)
5269	{
5270	struct cgroup parent, cgrp;
5271	struct kernfs_node *kn;
5272	int ret;
5273
5274	/* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5275	if (strchr(name, '\n'))
5276	return -EINVAL;
5277
5278	parent = cgroup_kn_lock_live(parent_kn, false);
5279	if (!parent)
5280	return -ENODEV;
5281
5282	cgrp = cgroup_create(parent);
5283	if (IS_ERR(cgrp)) {
5284	ret = PTR_ERR(cgrp);
5285	goto out_unlock;
5286	}
5287
5288	/* create the directory */
5289	kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5290	if (IS_ERR(kn)) {
5291	ret = PTR_ERR(kn);
5292	goto out_destroy;
5293	}
5294	cgrp->kn = kn;
5295
5296	/*
5297	* This extra ref will be put in cgroup_free_fn() and guarantees
5298	* that @cgrp->kn is always accessible.
5299	*/
5300	kernfs_get(kn);
5301
5302	ret = cgroup_kn_set_ugid(kn);
5303	if (ret)
5304	goto out_destroy;
5305
5306	ret = css_populate_dir(&cgrp->self);
5307	if (ret)
5308	goto out_destroy;
5309
5310	ret = cgroup_apply_control_enable(cgrp);
5311	if (ret)
5312	goto out_destroy;
5313
5314	/* let's create and online css's */
5315	kernfs_activate(kn);
5316
5317	ret = 0;
5318	goto out_unlock;
5319
5320	out_destroy:
5321	cgroup_destroy_locked(cgrp);
5322	out_unlock:
5323	cgroup_kn_unlock(parent_kn);
5324	return ret;
5325	}
5326
5327	/*
5328	* This is called when the refcnt of a css is confirmed to be killed.
5329	* css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5330	* initate destruction and put the css ref from kill_css().
5331	*/
5332	static void css_killed_work_fn(struct work_struct *work)
5333	{
5334	struct cgroup_subsys_state *css =
5335	container_of(work, struct cgroup_subsys_state, destroy_work);
5336
5337	mutex_lock(&cgroup_mutex);
5338
5339	do {
5340	offline_css(css);
5341	css_put(css);
5342	/* @css can't go away while we're holding cgroup_mutex */
5343	css = css->parent;
5344	} while (css && atomic_dec_and_test(&css->online_cnt));
5345
5346	mutex_unlock(&cgroup_mutex);
5347	}
5348
5349	/* css kill confirmation processing requires process context, bounce */
5350	static void css_killed_ref_fn(struct percpu_ref *ref)
5351	{
5352	struct cgroup_subsys_state *css =
5353	container_of(ref, struct cgroup_subsys_state, refcnt);
5354
5355	if (atomic_dec_and_test(&css->online_cnt)) {
5356	INIT_WORK(&css->destroy_work, css_killed_work_fn);
5357	queue_work(cgroup_destroy_wq, &css->destroy_work);
5358	}
5359	}
5360
5361	/**
5362	* kill_css - destroy a css
5363	* @css: css to destroy
5364	*
5365	* This function initiates destruction of @css by removing cgroup interface
5366	* files and putting its base reference. ->css_offline() will be invoked
5367	* asynchronously once css_tryget_online() is guaranteed to fail and when
5368	* the reference count reaches zero, @css will be released.
5369	*/
5370	static void kill_css(struct cgroup_subsys_state *css)
5371	{
5372	lockdep_assert_held(&cgroup_mutex);
5373
5374	/*
5375	* This must happen before css is disassociated with its cgroup.
5376	* See seq_css() for details.
5377	*/
5378	css_clear_dir(css);
5379
5380	/*
5381	* Killing would put the base ref, but we need to keep it alive
5382	* until after ->css_offline().
5383	*/
5384	css_get(css);
5385
5386	/*
5387	* cgroup core guarantees that, by the time ->css_offline() is
5388	* invoked, no new css reference will be given out via
5389	* css_tryget_online(). We can't simply call percpu_ref_kill() and
5390	* proceed to offlining css's because percpu_ref_kill() doesn't
5391	* guarantee that the ref is seen as killed on all CPUs on return.
5392	*
5393	* Use percpu_ref_kill_and_confirm() to get notifications as each
5394	* css is confirmed to be seen as killed on all CPUs.
5395	*/
5396	percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5397	}
5398
5399	/**
5400	* cgroup_destroy_locked - the first stage of cgroup destruction
5401	* @cgrp: cgroup to be destroyed
5402	*
5403	* css's make use of percpu refcnts whose killing latency shouldn't be
5404	* exposed to userland and are RCU protected. Also, cgroup core needs to
5405	* guarantee that css_tryget_online() won't succeed by the time
5406	* ->css_offline() is invoked. To satisfy all the requirements,
5407	* destruction is implemented in the following two steps.
5408	*
5409	* s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5410	* userland visible parts and start killing the percpu refcnts of
5411	* css's. Set up so that the next stage will be kicked off once all
5412	* the percpu refcnts are confirmed to be killed.
5413	*
5414	* s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5415	* rest of destruction. Once all cgroup references are gone, the
5416	* cgroup is RCU-freed.
5417	*
5418	* This function implements s1. After this step, @cgrp is gone as far as
5419	* the userland is concerned and a new cgroup with the same name may be
5420	* created. As cgroup doesn't care about the names internally, this
5421	* doesn't cause any problem.
5422	*/
5423	static int cgroup_destroy_locked(struct cgroup *cgrp)
5424	__releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5425	{
5426	struct cgroup_subsys_state *css;
5427	struct cgrp_cset_link *link;
5428	int ssid;
5429
5430	lockdep_assert_held(&cgroup_mutex);
5431
5432	/*
5433	* Only migration can raise populated from zero and we're already
5434	* holding cgroup_mutex.
5435	*/
5436	if (cgroup_is_populated(cgrp))
5437	return -EBUSY;
5438
5439	/*
5440	* Make sure there's no live children. We can't test emptiness of
5441	* ->self.children as dead children linger on it while being
5442	* drained; otherwise, "rmdir parent/child parent" may fail.
5443	*/
5444	if (css_has_online_children(&cgrp->self))
5445	return -EBUSY;
5446
5447	/*
5448	* Mark @cgrp and the associated csets dead. The former prevents
5449	* further task migration and child creation by disabling
5450	* cgroup_lock_live_group(). The latter makes the csets ignored by
5451	* the migration path.
5452	*/
5453	cgrp->self.flags &= ~CSS_ONLINE;
5454
5455	spin_lock_irq(&css_set_lock);
5456	list_for_each_entry(link, &cgrp->cset_links, cset_link)
5457	link->cset->dead = true;
5458	spin_unlock_irq(&css_set_lock);
5459
5460	/* initiate massacre of all css's */
5461	for_each_css(css, ssid, cgrp)
5462	kill_css(css);
5463
5464	/*
5465	* Remove @cgrp directory along with the base files. @cgrp has an
5466	* extra ref on its kn.
5467	*/
5468	kernfs_remove(cgrp->kn);
5469
5470	check_for_release(cgroup_parent(cgrp));
5471
5472	/* put the base reference */
5473	percpu_ref_kill(&cgrp->self.refcnt);
5474
5475	return 0;
5476	};
5477
5478	static int cgroup_rmdir(struct kernfs_node *kn)
5479	{
5480	struct cgroup *cgrp;
5481	int ret = 0;
5482
5483	cgrp = cgroup_kn_lock_live(kn, false);
5484	if (!cgrp)
5485	return 0;
5486
5487	ret = cgroup_destroy_locked(cgrp);
5488
5489	cgroup_kn_unlock(kn);
5490	return ret;
5491	}
5492
5493	static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5494	.remount_fs = cgroup_remount,
5495	.show_options = cgroup_show_options,
5496	.mkdir = cgroup_mkdir,
5497	.rmdir = cgroup_rmdir,
5498	.rename = cgroup_rename,
5499	.show_path = cgroup_show_path,
5500	};
5501
5502	static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5503	{
5504	struct cgroup_subsys_state *css;
5505
5506	pr_debug("Initializing cgroup subsys %s\n", ss->name);
5507
5508	mutex_lock(&cgroup_mutex);
5509
5510	idr_init(&ss->css_idr);
5511	INIT_LIST_HEAD(&ss->cfts);
5512
5513	/* Create the root cgroup state for this subsystem */
5514	ss->root = &cgrp_dfl_root;
5515	css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5516	/* We don't handle early failures gracefully */
5517	BUG_ON(IS_ERR(css));
5518	init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5519
5520	/*
5521	* Root csses are never destroyed and we can't initialize
5522	* percpu_ref during early init. Disable refcnting.
5523	*/
5524	css->flags \|= CSS_NO_REF;
5525
5526	if (early) {
5527	/* allocation can't be done safely during early init */
5528	css->id = 1;
5529	} else {
5530	css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5531	BUG_ON(css->id < 0);
5532	}
5533
5534	/* Update the init_css_set to contain a subsys
5535	* pointer to this state - since the subsystem is
5536	* newly registered, all tasks and hence the
5537	* init_css_set is in the subsystem's root cgroup. */
5538	init_css_set.subsys[ss->id] = css;
5539
5540	have_fork_callback \|= (bool)ss->fork << ss->id;
5541	have_exit_callback \|= (bool)ss->exit << ss->id;
5542	have_free_callback \|= (bool)ss->free << ss->id;
5543	have_canfork_callback \|= (bool)ss->can_fork << ss->id;
5544
5545	/* At system boot, before all subsystems have been
5546	* registered, no tasks have been forked, so we don't
5547	* need to invoke fork callbacks here. */
5548	BUG_ON(!list_empty(&init_task.tasks));
5549
5550	BUG_ON(online_css(css));
5551
5552	mutex_unlock(&cgroup_mutex);
5553	}
5554
5555	/**
5556	* cgroup_init_early - cgroup initialization at system boot
5557	*
5558	* Initialize cgroups at system boot, and initialize any
5559	* subsystems that request early init.
5560	*/
5561	int __init cgroup_init_early(void)
5562	{
5563	static struct cgroup_sb_opts __initdata opts;
5564	struct cgroup_subsys *ss;
5565	int i;
5566
5567	init_cgroup_root(&cgrp_dfl_root, &opts);
5568	cgrp_dfl_root.cgrp.self.flags \|= CSS_NO_REF;
5569
5570	RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5571
5572	for_each_subsys(ss, i) {
5573	WARN(!ss->css_alloc \|\| !ss->css_free \|\| ss->name \|\| ss->id,
5574	"invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5575	i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5576	ss->id, ss->name);
5577	WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5578	"cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5579
5580	ss->id = i;
5581	ss->name = cgroup_subsys_name[i];
5582	if (!ss->legacy_name)
5583	ss->legacy_name = cgroup_subsys_name[i];
5584
5585	if (ss->early_init)
5586	cgroup_init_subsys(ss, true);
5587	}
5588	return 0;
5589	}
5590
5591	static u16 cgroup_disable_mask __initdata;
5592
5593	/**
5594	* cgroup_init - cgroup initialization
5595	*
5596	* Register cgroup filesystem and /proc file, and initialize
5597	* any subsystems that didn't request early init.
5598	*/
5599	int __init cgroup_init(void)
5600	{
5601	struct cgroup_subsys *ss;
5602	int ssid;
5603
5604	BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5605	BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
5606	BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files));
5607	BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files));
5608
5609	get_user_ns(init_cgroup_ns.user_ns);
5610
5611	mutex_lock(&cgroup_mutex);
5612
5613	/*
5614	* Add init_css_set to the hash table so that dfl_root can link to
5615	* it during init.
5616	*/
5617	hash_add(css_set_table, &init_css_set.hlist,
5618	css_set_hash(init_css_set.subsys));
5619
5620	BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5621
5622	mutex_unlock(&cgroup_mutex);
5623
5624	for_each_subsys(ss, ssid) {
5625	if (ss->early_init) {
5626	struct cgroup_subsys_state *css =
5627	init_css_set.subsys[ss->id];
5628
5629	css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5630	GFP_KERNEL);
5631	BUG_ON(css->id < 0);
5632	} else {
5633	cgroup_init_subsys(ss, false);
5634	}
5635
5636	list_add_tail(&init_css_set.e_cset_node[ssid],
5637	&cgrp_dfl_root.cgrp.e_csets[ssid]);
5638
5639	/*
5640	* Setting dfl_root subsys_mask needs to consider the
5641	* disabled flag and cftype registration needs kmalloc,
5642	* both of which aren't available during early_init.
5643	*/
5644	if (cgroup_disable_mask & (1 << ssid)) {
5645	static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5646	printk(KERN_INFO "Disabling %s control group subsystem\n",
5647	ss->name);
5648	continue;
5649	}
5650
5651	if (cgroup_ssid_no_v1(ssid))
5652	printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5653	ss->name);
5654
5655	cgrp_dfl_root.subsys_mask \|= 1 << ss->id;
5656
5657	if (ss->implicit_on_dfl)
5658	cgrp_dfl_implicit_ss_mask \|= 1 << ss->id;
5659	else if (!ss->dfl_cftypes)
5660	cgrp_dfl_inhibit_ss_mask \|= 1 << ss->id;
5661
5662	if (ss->dfl_cftypes == ss->legacy_cftypes) {
5663	WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5664	} else {
5665	WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5666	WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5667	}
5668
5669	if (ss->bind)
5670	ss->bind(init_css_set.subsys[ssid]);
5671	}
5672
5673	/* init_css_set.subsys[] has been updated, re-hash */
5674	hash_del(&init_css_set.hlist);
5675	hash_add(css_set_table, &init_css_set.hlist,
5676	css_set_hash(init_css_set.subsys));
5677
5678	WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5679	WARN_ON(register_filesystem(&cgroup_fs_type));
5680	WARN_ON(register_filesystem(&cgroup2_fs_type));
5681	WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
5682
5683	return 0;
5684	}
5685
5686	static int __init cgroup_wq_init(void)
5687	{
5688	/*
5689	* There isn't much point in executing destruction path in
5690	* parallel. Good chunk is serialized with cgroup_mutex anyway.
5691	* Use 1 for @max_active.
5692	*
5693	* We would prefer to do this in cgroup_init() above, but that
5694	* is called before init_workqueues(): so leave this until after.
5695	*/
5696	cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5697	BUG_ON(!cgroup_destroy_wq);
5698
5699	/*
5700	* Used to destroy pidlists and separate to serve as flush domain.
5701	* Cap @max_active to 1 too.
5702	*/
5703	cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
5704	0, 1);
5705	BUG_ON(!cgroup_pidlist_destroy_wq);
5706
5707	return 0;
5708	}
5709	core_initcall(cgroup_wq_init);
5710
5711	/*
5712	* proc_cgroup_show()
5713	* - Print task's cgroup paths into seq_file, one line for each hierarchy
5714	* - Used for /proc/<pid>/cgroup.
5715	*/
5716	int proc_cgroup_show(struct seq_file m, struct pid_namespace ns,
5717	struct pid pid, struct task_struct tsk)
5718	{
5719	char buf, path;
5720	int retval;
5721	struct cgroup_root *root;
5722
5723	retval = -ENOMEM;
5724	buf = kmalloc(PATH_MAX, GFP_KERNEL);
5725	if (!buf)
5726	goto out;
5727
5728	mutex_lock(&cgroup_mutex);
5729	spin_lock_irq(&css_set_lock);
5730
5731	for_each_root(root) {
5732	struct cgroup_subsys *ss;
5733	struct cgroup *cgrp;
5734	int ssid, count = 0;
5735
5736	if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5737	continue;
5738
5739	seq_printf(m, "%d:", root->hierarchy_id);
5740	if (root != &cgrp_dfl_root)
5741	for_each_subsys(ss, ssid)
5742	if (root->subsys_mask & (1 << ssid))
5743	seq_printf(m, "%s%s", count++ ? "," : "",
5744	ss->legacy_name);
5745	if (strlen(root->name))
5746	seq_printf(m, "%sname=%s", count ? "," : "",
5747	root->name);
5748	seq_putc(m, ':');
5749
5750	cgrp = task_cgroup_from_root(tsk, root);
5751
5752	/*
5753	* On traditional hierarchies, all zombie tasks show up as
5754	* belonging to the root cgroup. On the default hierarchy,
5755	* while a zombie doesn't show up in "cgroup.procs" and
5756	* thus can't be migrated, its /proc/PID/cgroup keeps
5757	* reporting the cgroup it belonged to before exiting. If
5758	* the cgroup is removed before the zombie is reaped,
5759	* " (deleted)" is appended to the cgroup path.
5760	*/
5761	if (cgroup_on_dfl(cgrp) \|\| !(tsk->flags & PF_EXITING)) {
5762	path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5763	current->nsproxy->cgroup_ns);
5764	if (!path) {
5765	retval = -ENAMETOOLONG;
5766	goto out_unlock;
5767	}
5768	} else {
5769	path = "/";
5770	}
5771
5772	seq_puts(m, path);
5773
5774	if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5775	seq_puts(m, " (deleted)\n");
5776	else
5777	seq_putc(m, '\n');
5778	}
5779
5780	retval = 0;
5781	out_unlock:
5782	spin_unlock_irq(&css_set_lock);
5783	mutex_unlock(&cgroup_mutex);
5784	kfree(buf);
5785	out:
5786	return retval;
5787	}
5788
5789	/* Display information about each subsystem and each hierarchy */
5790	static int proc_cgroupstats_show(struct seq_file m, void v)
5791	{
5792	struct cgroup_subsys *ss;
5793	int i;
5794
5795	seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5796	/*
5797	* ideally we don't want subsystems moving around while we do this.
5798	* cgroup_mutex is also necessary to guarantee an atomic snapshot of
5799	* subsys/hierarchy state.
5800	*/
5801	mutex_lock(&cgroup_mutex);
5802
5803	for_each_subsys(ss, i)
5804	seq_printf(m, "%s\t%d\t%d\t%d\n",
5805	ss->legacy_name, ss->root->hierarchy_id,
5806	atomic_read(&ss->root->nr_cgrps),
5807	cgroup_ssid_enabled(i));
5808
5809	mutex_unlock(&cgroup_mutex);
5810	return 0;
5811	}
5812
5813	static int cgroupstats_open(struct inode inode, struct file file)
5814	{
5815	return single_open(file, proc_cgroupstats_show, NULL);
5816	}
5817
5818	static const struct file_operations proc_cgroupstats_operations = {
5819	.open = cgroupstats_open,
5820	.read = seq_read,
5821	.llseek = seq_lseek,
5822	.release = single_release,
5823	};
5824
5825	/**
5826	* cgroup_fork - initialize cgroup related fields during copy_process()
5827	* @child: pointer to task_struct of forking parent process.
5828	*
5829	* A task is associated with the init_css_set until cgroup_post_fork()
5830	* attaches it to the parent's css_set. Empty cg_list indicates that
5831	* @child isn't holding reference to its css_set.
5832	*/
5833	void cgroup_fork(struct task_struct *child)
5834	{
5835	RCU_INIT_POINTER(child->cgroups, &init_css_set);
5836	INIT_LIST_HEAD(&child->cg_list);
5837	}
5838
5839	/**
5840	* cgroup_can_fork - called on a new task before the process is exposed
5841	* @child: the task in question.
5842	*
5843	* This calls the subsystem can_fork() callbacks. If the can_fork() callback
5844	* returns an error, the fork aborts with that error code. This allows for
5845	* a cgroup subsystem to conditionally allow or deny new forks.
5846	*/
5847	int cgroup_can_fork(struct task_struct *child)
5848	{
5849	struct cgroup_subsys *ss;
5850	int i, j, ret;
5851
5852	do_each_subsys_mask(ss, i, have_canfork_callback) {
5853	ret = ss->can_fork(child);
5854	if (ret)
5855	goto out_revert;
5856	} while_each_subsys_mask();
5857
5858	return 0;
5859
5860	out_revert:
5861	for_each_subsys(ss, j) {
5862	if (j >= i)
5863	break;
5864	if (ss->cancel_fork)
5865	ss->cancel_fork(child);
5866	}
5867
5868	return ret;
5869	}
5870
5871	/**
5872	* cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5873	* @child: the task in question
5874	*
5875	* This calls the cancel_fork() callbacks if a fork failed after
5876	* cgroup_can_fork() succeded.
5877	*/
5878	void cgroup_cancel_fork(struct task_struct *child)
5879	{
5880	struct cgroup_subsys *ss;
5881	int i;
5882
5883	for_each_subsys(ss, i)
5884	if (ss->cancel_fork)
5885	ss->cancel_fork(child);
5886	}
5887
5888	/**
5889	* cgroup_post_fork - called on a new task after adding it to the task list
5890	* @child: the task in question
5891	*
5892	* Adds the task to the list running through its css_set if necessary and
5893	* call the subsystem fork() callbacks. Has to be after the task is
5894	* visible on the task list in case we race with the first call to
5895	* cgroup_task_iter_start() - to guarantee that the new task ends up on its
5896	* list.
5897	*/
5898	void cgroup_post_fork(struct task_struct *child)
5899	{
5900	struct cgroup_subsys *ss;
5901	int i;
5902
5903	/*
5904	* This may race against cgroup_enable_task_cg_lists(). As that
5905	* function sets use_task_css_set_links before grabbing
5906	* tasklist_lock and we just went through tasklist_lock to add
5907	* @child, it's guaranteed that either we see the set
5908	* use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5909	* @child during its iteration.
5910	*
5911	* If we won the race, @child is associated with %current's
5912	* css_set. Grabbing css_set_lock guarantees both that the
5913	* association is stable, and, on completion of the parent's
5914	* migration, @child is visible in the source of migration or
5915	* already in the destination cgroup. This guarantee is necessary
5916	* when implementing operations which need to migrate all tasks of
5917	* a cgroup to another.
5918	*
5919	* Note that if we lose to cgroup_enable_task_cg_lists(), @child
5920	* will remain in init_css_set. This is safe because all tasks are
5921	* in the init_css_set before cg_links is enabled and there's no
5922	* operation which transfers all tasks out of init_css_set.
5923	*/
5924	if (use_task_css_set_links) {
5925	struct css_set *cset;
5926
5927	spin_lock_irq(&css_set_lock);
5928	cset = task_css_set(current);
5929	if (list_empty(&child->cg_list)) {
5930	get_css_set(cset);
5931	css_set_move_task(child, NULL, cset, false);
5932	}
5933	spin_unlock_irq(&css_set_lock);
5934	}
5935
5936	/*
5937	* Call ss->fork(). This must happen after @child is linked on
5938	* css_set; otherwise, @child might change state between ->fork()
5939	* and addition to css_set.
5940	*/
5941	do_each_subsys_mask(ss, i, have_fork_callback) {
5942	ss->fork(child);
5943	} while_each_subsys_mask();
5944	}
5945
5946	/**
5947	* cgroup_exit - detach cgroup from exiting task
5948	* @tsk: pointer to task_struct of exiting process
5949	*
5950	* Description: Detach cgroup from @tsk and release it.
5951	*
5952	* Note that cgroups marked notify_on_release force every task in
5953	* them to take the global cgroup_mutex mutex when exiting.
5954	* This could impact scaling on very large systems. Be reluctant to
5955	* use notify_on_release cgroups where very high task exit scaling
5956	* is required on large systems.
5957	*
5958	* We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5959	* call cgroup_exit() while the task is still competent to handle
5960	* notify_on_release(), then leave the task attached to the root cgroup in
5961	* each hierarchy for the remainder of its exit. No need to bother with
5962	* init_css_set refcnting. init_css_set never goes away and we can't race
5963	* with migration path - PF_EXITING is visible to migration path.
5964	*/
5965	void cgroup_exit(struct task_struct *tsk)
5966	{
5967	struct cgroup_subsys *ss;
5968	struct css_set *cset;
5969	int i;
5970
5971	/*
5972	* Unlink from @tsk from its css_set. As migration path can't race
5973	* with us, we can check css_set and cg_list without synchronization.
5974	*/
5975	cset = task_css_set(tsk);
5976
5977	if (!list_empty(&tsk->cg_list)) {
5978	spin_lock_irq(&css_set_lock);
5979	css_set_move_task(tsk, cset, NULL, false);
5980	spin_unlock_irq(&css_set_lock);
5981	} else {
5982	get_css_set(cset);
5983	}
5984
5985	/* see cgroup_post_fork() for details */
5986	do_each_subsys_mask(ss, i, have_exit_callback) {
5987	ss->exit(tsk);
5988	} while_each_subsys_mask();
5989	}
5990
5991	void cgroup_free(struct task_struct *task)
5992	{
5993	struct css_set *cset = task_css_set(task);
5994	struct cgroup_subsys *ss;
5995	int ssid;
5996
5997	do_each_subsys_mask(ss, ssid, have_free_callback) {
5998	ss->free(task);
5999	} while_each_subsys_mask();
6000
6001	put_css_set(cset);
6002	}
6003
6004	static void check_for_release(struct cgroup *cgrp)
6005	{
6006	if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
6007	!css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
6008	schedule_work(&cgrp->release_agent_work);
6009	}
6010
6011	/*
6012	* Notify userspace when a cgroup is released, by running the
6013	* configured release agent with the name of the cgroup (path
6014	* relative to the root of cgroup file system) as the argument.
6015	*
6016	* Most likely, this user command will try to rmdir this cgroup.
6017	*
6018	* This races with the possibility that some other task will be
6019	* attached to this cgroup before it is removed, or that some other
6020	* user task will 'mkdir' a child cgroup of this cgroup. That's ok.
6021	* The presumed 'rmdir' will fail quietly if this cgroup is no longer
6022	* unused, and this cgroup will be reprieved from its death sentence,
6023	* to continue to serve a useful existence. Next time it's released,
6024	* we will get notified again, if it still has 'notify_on_release' set.
6025	*
6026	* The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
6027	* means only wait until the task is successfully execve()'d. The
6028	* separate release agent task is forked by call_usermodehelper(),
6029	* then control in this thread returns here, without waiting for the
6030	* release agent task. We don't bother to wait because the caller of
6031	* this routine has no use for the exit status of the release agent
6032	* task, so no sense holding our caller up for that.
6033	*/
6034	static void cgroup_release_agent(struct work_struct *work)
6035	{
6036	struct cgroup *cgrp =
6037	container_of(work, struct cgroup, release_agent_work);
6038	char pathbuf = NULL, agentbuf = NULL, *path;
6039	char argv[3], envp[3];
6040
6041	mutex_lock(&cgroup_mutex);
6042
6043	pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
6044	agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
6045	if (!pathbuf \|\| !agentbuf)
6046	goto out;
6047
6048	spin_lock_irq(&css_set_lock);
6049	path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
6050	spin_unlock_irq(&css_set_lock);
6051	if (!path)
6052	goto out;
6053
6054	argv[0] = agentbuf;
6055	argv[1] = path;
6056	argv[2] = NULL;
6057
6058	/* minimal command environment */
6059	envp[0] = "HOME=/";
6060	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
6061	envp[2] = NULL;
6062
6063	mutex_unlock(&cgroup_mutex);
6064	call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
6065	goto out_free;
6066	out:
6067	mutex_unlock(&cgroup_mutex);
6068	out_free:
6069	kfree(agentbuf);
6070	kfree(pathbuf);
6071	}
6072
6073	static int __init cgroup_disable(char *str)
6074	{
6075	struct cgroup_subsys *ss;
6076	char *token;
6077	int i;
6078
6079	while ((token = strsep(&str, ",")) != NULL) {
6080	if (!*token)
6081	continue;
6082
6083	for_each_subsys(ss, i) {
6084	if (strcmp(token, ss->name) &&
6085	strcmp(token, ss->legacy_name))
6086	continue;
6087	cgroup_disable_mask \|= 1 << i;
6088	}
6089	}
6090	return 1;
6091	}
6092	__setup("cgroup_disable=", cgroup_disable);
6093
6094	static int __init cgroup_no_v1(char *str)
6095	{
6096	struct cgroup_subsys *ss;
6097	char *token;
6098	int i;
6099
6100	while ((token = strsep(&str, ",")) != NULL) {
6101	if (!*token)
6102	continue;
6103
6104	if (!strcmp(token, "all")) {
6105	cgroup_no_v1_mask = U16_MAX;
6106	break;
6107	}
6108
6109	for_each_subsys(ss, i) {
6110	if (strcmp(token, ss->name) &&
6111	strcmp(token, ss->legacy_name))
6112	continue;
6113
6114	cgroup_no_v1_mask \|= 1 << i;
6115	}
6116	}
6117	return 1;
6118	}
6119	__setup("cgroup_no_v1=", cgroup_no_v1);
6120
6121	/**
6122	* css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6123	* @dentry: directory dentry of interest
6124	* @ss: subsystem of interest
6125	*
6126	* If @dentry is a directory for a cgroup which has @ss enabled on it, try
6127	* to get the corresponding css and return it. If such css doesn't exist
6128	* or can't be pinned, an ERR_PTR value is returned.
6129	*/
6130	struct cgroup_subsys_state css_tryget_online_from_dir(struct dentry dentry,
6131	struct cgroup_subsys *ss)
6132	{
6133	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6134	struct file_system_type *s_type = dentry->d_sb->s_type;
6135	struct cgroup_subsys_state *css = NULL;
6136	struct cgroup *cgrp;
6137
6138	/* is @dentry a cgroup dir? */
6139	if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) \|\|
6140	!kn \|\| kernfs_type(kn) != KERNFS_DIR)
6141	return ERR_PTR(-EBADF);
6142
6143	rcu_read_lock();
6144
6145	/*
6146	* This path doesn't originate from kernfs and @kn could already
6147	* have been or be removed at any point. @kn->priv is RCU
6148	* protected for this access. See css_release_work_fn() for details.
6149	*/
6150	cgrp = rcu_dereference(kn->priv);
6151	if (cgrp)
6152	css = cgroup_css(cgrp, ss);
6153
6154	if (!css \|\| !css_tryget_online(css))
6155	css = ERR_PTR(-ENOENT);
6156
6157	rcu_read_unlock();
6158	return css;
6159	}
6160
6161	/**
6162	* css_from_id - lookup css by id
6163	* @id: the cgroup id
6164	* @ss: cgroup subsys to be looked into
6165	*
6166	* Returns the css if there's valid one with @id, otherwise returns NULL.
6167	* Should be called under rcu_read_lock().
6168	*/
6169	struct cgroup_subsys_state css_from_id(int id, struct cgroup_subsys ss)
6170	{
6171	WARN_ON_ONCE(!rcu_read_lock_held());
6172	return idr_find(&ss->css_idr, id);
6173	}
6174
6175	/**
6176	* cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6177	* @path: path on the default hierarchy
6178	*
6179	* Find the cgroup at @path on the default hierarchy, increment its
6180	* reference count and return it. Returns pointer to the found cgroup on
6181	* success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6182	* if @path points to a non-directory.
6183	*/
6184	struct cgroup cgroup_get_from_path(const char path)
6185	{
6186	struct kernfs_node *kn;
6187	struct cgroup *cgrp;
6188
6189	mutex_lock(&cgroup_mutex);
6190
6191	kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6192	if (kn) {
6193	if (kernfs_type(kn) == KERNFS_DIR) {
6194	cgrp = kn->priv;
6195	cgroup_get(cgrp);
6196	} else {
6197	cgrp = ERR_PTR(-ENOTDIR);
6198	}
6199	kernfs_put(kn);
6200	} else {
6201	cgrp = ERR_PTR(-ENOENT);
6202	}
6203
6204	mutex_unlock(&cgroup_mutex);
6205	return cgrp;
6206	}
6207	EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6208
6209	/**
6210	* cgroup_get_from_fd - get a cgroup pointer from a fd
6211	* @fd: fd obtained by open(cgroup2_dir)
6212	*
6213	* Find the cgroup from a fd which should be obtained
6214	* by opening a cgroup directory. Returns a pointer to the
6215	* cgroup on success. ERR_PTR is returned if the cgroup
6216	* cannot be found.
6217	*/
6218	struct cgroup *cgroup_get_from_fd(int fd)
6219	{
6220	struct cgroup_subsys_state *css;
6221	struct cgroup *cgrp;
6222	struct file *f;
6223
6224	f = fget_raw(fd);
6225	if (!f)
6226	return ERR_PTR(-EBADF);
6227
6228	css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6229	fput(f);
6230	if (IS_ERR(css))
6231	return ERR_CAST(css);
6232
6233	cgrp = css->cgroup;
6234	if (!cgroup_on_dfl(cgrp)) {
6235	cgroup_put(cgrp);
6236	return ERR_PTR(-EBADF);
6237	}
6238
6239	return cgrp;
6240	}
6241	EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6242
6243	/*
6244	* sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6245	* definition in cgroup-defs.h.
6246	*/
6247	#ifdef CONFIG_SOCK_CGROUP_DATA
6248
6249	#if defined(CONFIG_CGROUP_NET_PRIO) \|\| defined(CONFIG_CGROUP_NET_CLASSID)
6250
6251	DEFINE_SPINLOCK(cgroup_sk_update_lock);
6252	static bool cgroup_sk_alloc_disabled __read_mostly;
6253
6254	void cgroup_sk_alloc_disable(void)
6255	{
6256	if (cgroup_sk_alloc_disabled)
6257	return;
6258	pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6259	cgroup_sk_alloc_disabled = true;
6260	}
6261
6262	#else
6263
6264	#define cgroup_sk_alloc_disabled false
6265
6266	#endif
6267
6268	void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6269	{
6270	if (cgroup_sk_alloc_disabled)
6271	return;
6272
6273	rcu_read_lock();
6274
6275	while (true) {
6276	struct css_set *cset;
6277
6278	cset = task_css_set(current);
6279	if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6280	skcd->val = (unsigned long)cset->dfl_cgrp;
6281	break;
6282	}
6283	cpu_relax();
6284	}
6285
6286	rcu_read_unlock();
6287	}
6288
6289	void cgroup_sk_free(struct sock_cgroup_data *skcd)
6290	{
6291	cgroup_put(sock_cgroup_ptr(skcd));
6292	}
6293
6294	#endif /* CONFIG_SOCK_CGROUP_DATA */
6295
6296	/* cgroup namespaces */
6297
6298	static struct cgroup_namespace *alloc_cgroup_ns(void)
6299	{
6300	struct cgroup_namespace *new_ns;
6301	int ret;
6302
6303	new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL);
6304	if (!new_ns)
6305	return ERR_PTR(-ENOMEM);
6306	ret = ns_alloc_inum(&new_ns->ns);
6307	if (ret) {
6308	kfree(new_ns);
6309	return ERR_PTR(ret);
6310	}
6311	atomic_set(&new_ns->count, 1);
6312	new_ns->ns.ops = &cgroupns_operations;
6313	return new_ns;
6314	}
6315
6316	void free_cgroup_ns(struct cgroup_namespace *ns)
6317	{
6318	put_css_set(ns->root_cset);
6319	put_user_ns(ns->user_ns);
6320	ns_free_inum(&ns->ns);
6321	kfree(ns);
6322	}
6323	EXPORT_SYMBOL(free_cgroup_ns);
6324
6325	struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
6326	struct user_namespace *user_ns,
6327	struct cgroup_namespace *old_ns)
6328	{
6329	struct cgroup_namespace *new_ns;
6330	struct css_set *cset;
6331
6332	BUG_ON(!old_ns);
6333
6334	if (!(flags & CLONE_NEWCGROUP)) {
6335	get_cgroup_ns(old_ns);
6336	return old_ns;
6337	}
6338
6339	/* Allow only sysadmin to create cgroup namespace. */
6340	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
6341	return ERR_PTR(-EPERM);
6342
6343	/* It is not safe to take cgroup_mutex here */
6344	spin_lock_irq(&css_set_lock);
6345	cset = task_css_set(current);
6346	get_css_set(cset);
6347	spin_unlock_irq(&css_set_lock);
6348
6349	new_ns = alloc_cgroup_ns();
6350	if (IS_ERR(new_ns)) {
6351	put_css_set(cset);
6352	return new_ns;
6353	}
6354
6355	new_ns->user_ns = get_user_ns(user_ns);
6356	new_ns->root_cset = cset;
6357
6358	return new_ns;
6359	}
6360
6361	static inline struct cgroup_namespace to_cg_ns(struct ns_common ns)
6362	{
6363	return container_of(ns, struct cgroup_namespace, ns);
6364	}
6365
6366	static int cgroupns_install(struct nsproxy nsproxy, struct ns_common ns)
6367	{
6368	struct cgroup_namespace *cgroup_ns = to_cg_ns(ns);
6369
6370	if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN) \|\|
6371	!ns_capable(cgroup_ns->user_ns, CAP_SYS_ADMIN))
6372	return -EPERM;
6373
6374	/* Don't need to do anything if we are attaching to our own cgroupns. */
6375	if (cgroup_ns == nsproxy->cgroup_ns)
6376	return 0;
6377
6378	get_cgroup_ns(cgroup_ns);
6379	put_cgroup_ns(nsproxy->cgroup_ns);
6380	nsproxy->cgroup_ns = cgroup_ns;
6381
6382	return 0;
6383	}
6384
6385	static struct ns_common cgroupns_get(struct task_struct task)
6386	{
6387	struct cgroup_namespace *ns = NULL;
6388	struct nsproxy *nsproxy;
6389
6390	task_lock(task);
6391	nsproxy = task->nsproxy;
6392	if (nsproxy) {
6393	ns = nsproxy->cgroup_ns;
6394	get_cgroup_ns(ns);
6395	}
6396	task_unlock(task);
6397
6398	return ns ? &ns->ns : NULL;
6399	}
6400
6401	static void cgroupns_put(struct ns_common *ns)
6402	{
6403	put_cgroup_ns(to_cg_ns(ns));
6404	}
6405
6406	const struct proc_ns_operations cgroupns_operations = {
6407	.name = "cgroup",
6408	.type = CLONE_NEWCGROUP,
6409	.get = cgroupns_get,
6410	.put = cgroupns_put,
6411	.install = cgroupns_install,
6412	};
6413
6414	static __init int cgroup_namespaces_init(void)
6415	{
6416	return 0;
6417	}
6418	subsys_initcall(cgroup_namespaces_init);
6419
6420	#ifdef CONFIG_CGROUP_DEBUG
6421	static struct cgroup_subsys_state *
6422	debug_css_alloc(struct cgroup_subsys_state *parent_css)
6423	{
6424	struct cgroup_subsys_state css = kzalloc(sizeof(css), GFP_KERNEL);
6425
6426	if (!css)
6427	return ERR_PTR(-ENOMEM);
6428
6429	return css;
6430	}
6431
6432	static void debug_css_free(struct cgroup_subsys_state *css)
6433	{
6434	kfree(css);
6435	}
6436
6437	static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
6438	struct cftype *cft)
6439	{
6440	return cgroup_task_count(css->cgroup);
6441	}
6442
6443	static u64 current_css_set_read(struct cgroup_subsys_state *css,
6444	struct cftype *cft)
6445	{
6446	return (u64)(unsigned long)current->cgroups;
6447	}
6448
6449	static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
6450	struct cftype *cft)
6451	{
6452	u64 count;
6453
6454	rcu_read_lock();
6455	count = atomic_read(&task_css_set(current)->refcount);
6456	rcu_read_unlock();
6457	return count;
6458	}
6459
6460	static int current_css_set_cg_links_read(struct seq_file seq, void v)
6461	{
6462	struct cgrp_cset_link *link;
6463	struct css_set *cset;
6464	char *name_buf;
6465
6466	name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
6467	if (!name_buf)
6468	return -ENOMEM;
6469
6470	spin_lock_irq(&css_set_lock);
6471	rcu_read_lock();
6472	cset = rcu_dereference(current->cgroups);
6473	list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
6474	struct cgroup *c = link->cgrp;
6475
6476	cgroup_name(c, name_buf, NAME_MAX + 1);
6477	seq_printf(seq, "Root %d group %s\n",
6478	c->root->hierarchy_id, name_buf);
6479	}
6480	rcu_read_unlock();
6481	spin_unlock_irq(&css_set_lock);
6482	kfree(name_buf);
6483	return 0;
6484	}
6485
6486	#define MAX_TASKS_SHOWN_PER_CSS 25
6487	static int cgroup_css_links_read(struct seq_file seq, void v)
6488	{
6489	struct cgroup_subsys_state *css = seq_css(seq);
6490	struct cgrp_cset_link *link;
6491
6492	spin_lock_irq(&css_set_lock);
6493	list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
6494	struct css_set *cset = link->cset;
6495	struct task_struct *task;
6496	int count = 0;
6497
6498	seq_printf(seq, "css_set %p\n", cset);
6499
6500	list_for_each_entry(task, &cset->tasks, cg_list) {
6501	if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6502	goto overflow;
6503	seq_printf(seq, " task %d\n", task_pid_vnr(task));
6504	}
6505
6506	list_for_each_entry(task, &cset->mg_tasks, cg_list) {
6507	if (count++ > MAX_TASKS_SHOWN_PER_CSS)
6508	goto overflow;
6509	seq_printf(seq, " task %d\n", task_pid_vnr(task));
6510	}
6511	continue;
6512	overflow:
6513	seq_puts(seq, " ...\n");
6514	}
6515	spin_unlock_irq(&css_set_lock);
6516	return 0;
6517	}
6518
6519	static u64 releasable_read(struct cgroup_subsys_state css, struct cftype cft)
6520	{
6521	return (!cgroup_is_populated(css->cgroup) &&
6522	!css_has_online_children(&css->cgroup->self));
6523	}
6524
6525	static struct cftype debug_files[] = {
6526	{
6527	.name = "taskcount",
6528	.read_u64 = debug_taskcount_read,
6529	},
6530
6531	{
6532	.name = "current_css_set",
6533	.read_u64 = current_css_set_read,
6534	},
6535
6536	{
6537	.name = "current_css_set_refcount",
6538	.read_u64 = current_css_set_refcount_read,
6539	},
6540
6541	{
6542	.name = "current_css_set_cg_links",
6543	.seq_show = current_css_set_cg_links_read,
6544	},
6545
6546	{
6547	.name = "cgroup_css_links",
6548	.seq_show = cgroup_css_links_read,
6549	},
6550
6551	{
6552	.name = "releasable",
6553	.read_u64 = releasable_read,
6554	},
6555
6556	{ } /* terminate */
6557	};
6558
6559	struct cgroup_subsys debug_cgrp_subsys = {
6560	.css_alloc = debug_css_alloc,
6561	.css_free = debug_css_free,
6562	.legacy_cftypes = debug_files,
6563	};
6564	#endif /* CONFIG_CGROUP_DEBUG */