xref: /openbmc/linux/kernel/cgroup/cgroup.c (revision 65a0d3c1)
1 /*
2  *  Generic process-grouping system.
3  *
4  *  Based originally on the cpuset system, extracted by Paul Menage
5  *  Copyright (C) 2006 Google, Inc
6  *
7  *  Notifications support
8  *  Copyright (C) 2009 Nokia Corporation
9  *  Author: Kirill A. Shutemov
10  *
11  *  Copyright notices from the original cpuset code:
12  *  --------------------------------------------------
13  *  Copyright (C) 2003 BULL SA.
14  *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
15  *
16  *  Portions derived from Patrick Mochel's sysfs code.
17  *  sysfs is Copyright (c) 2001-3 Patrick Mochel
18  *
19  *  2003-10-10 Written by Simon Derr.
20  *  2003-10-22 Updates by Stephen Hemminger.
21  *  2004 May-July Rework by Paul Jackson.
22  *  ---------------------------------------------------
23  *
24  *  This file is subject to the terms and conditions of the GNU General Public
25  *  License.  See the file COPYING in the main directory of the Linux
26  *  distribution for more details.
27  */
28 
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 
31 #include "cgroup-internal.h"
32 
33 #include <linux/cred.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/magic.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/sched/task.h>
45 #include <linux/slab.h>
46 #include <linux/spinlock.h>
47 #include <linux/percpu-rwsem.h>
48 #include <linux/string.h>
49 #include <linux/hashtable.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/atomic.h>
53 #include <linux/cpuset.h>
54 #include <linux/proc_ns.h>
55 #include <linux/nsproxy.h>
56 #include <linux/file.h>
57 #include <linux/fs_parser.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/psi.h>
60 #include <net/sock.h>
61 
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/cgroup.h>
64 
65 #define CGROUP_FILE_NAME_MAX		(MAX_CGROUP_TYPE_NAMELEN +	\
66 					 MAX_CFTYPE_NAME + 2)
67 /* let's not notify more than 100 times per second */
68 #define CGROUP_FILE_NOTIFY_MIN_INTV	DIV_ROUND_UP(HZ, 100)
69 
70 /*
71  * cgroup_mutex is the master lock.  Any modification to cgroup or its
72  * hierarchy must be performed while holding it.
73  *
74  * css_set_lock protects task->cgroups pointer, the list of css_set
75  * objects, and the chain of tasks off each css_set.
76  *
77  * These locks are exported if CONFIG_PROVE_RCU so that accessors in
78  * cgroup.h can use them for lockdep annotations.
79  */
80 DEFINE_MUTEX(cgroup_mutex);
81 DEFINE_SPINLOCK(css_set_lock);
82 
83 #ifdef CONFIG_PROVE_RCU
84 EXPORT_SYMBOL_GPL(cgroup_mutex);
85 EXPORT_SYMBOL_GPL(css_set_lock);
86 #endif
87 
88 DEFINE_SPINLOCK(trace_cgroup_path_lock);
89 char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
90 bool cgroup_debug __read_mostly;
91 
92 /*
93  * Protects cgroup_idr and css_idr so that IDs can be released without
94  * grabbing cgroup_mutex.
95  */
96 static DEFINE_SPINLOCK(cgroup_idr_lock);
97 
98 /*
99  * Protects cgroup_file->kn for !self csses.  It synchronizes notifications
100  * against file removal/re-creation across css hiding.
101  */
102 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
103 
104 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
105 
106 #define cgroup_assert_mutex_or_rcu_locked()				\
107 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&			\
108 			   !lockdep_is_held(&cgroup_mutex),		\
109 			   "cgroup_mutex or RCU read lock required");
110 
111 /*
112  * cgroup destruction makes heavy use of work items and there can be a lot
113  * of concurrent destructions.  Use a separate workqueue so that cgroup
114  * destruction work items don't end up filling up max_active of system_wq
115  * which may lead to deadlock.
116  */
117 static struct workqueue_struct *cgroup_destroy_wq;
118 
119 /* generate an array of cgroup subsystem pointers */
120 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
121 struct cgroup_subsys *cgroup_subsys[] = {
122 #include <linux/cgroup_subsys.h>
123 };
124 #undef SUBSYS
125 
126 /* array of cgroup subsystem names */
127 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
128 static const char *cgroup_subsys_name[] = {
129 #include <linux/cgroup_subsys.h>
130 };
131 #undef SUBSYS
132 
133 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
134 #define SUBSYS(_x)								\
135 	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key);			\
136 	DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key);			\
137 	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key);			\
138 	EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
139 #include <linux/cgroup_subsys.h>
140 #undef SUBSYS
141 
142 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
143 static struct static_key_true *cgroup_subsys_enabled_key[] = {
144 #include <linux/cgroup_subsys.h>
145 };
146 #undef SUBSYS
147 
148 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
149 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
150 #include <linux/cgroup_subsys.h>
151 };
152 #undef SUBSYS
153 
154 static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
155 
156 /* the default hierarchy */
157 struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
158 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
159 
160 /*
161  * The default hierarchy always exists but is hidden until mounted for the
162  * first time.  This is for backward compatibility.
163  */
164 static bool cgrp_dfl_visible;
165 
166 /* some controllers are not supported in the default hierarchy */
167 static u16 cgrp_dfl_inhibit_ss_mask;
168 
169 /* some controllers are implicitly enabled on the default hierarchy */
170 static u16 cgrp_dfl_implicit_ss_mask;
171 
172 /* some controllers can be threaded on the default hierarchy */
173 static u16 cgrp_dfl_threaded_ss_mask;
174 
175 /* The list of hierarchy roots */
176 LIST_HEAD(cgroup_roots);
177 static int cgroup_root_count;
178 
179 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
180 static DEFINE_IDR(cgroup_hierarchy_idr);
181 
182 /*
183  * Assign a monotonically increasing serial number to csses.  It guarantees
184  * cgroups with bigger numbers are newer than those with smaller numbers.
185  * Also, as csses are always appended to the parent's ->children list, it
186  * guarantees that sibling csses are always sorted in the ascending serial
187  * number order on the list.  Protected by cgroup_mutex.
188  */
189 static u64 css_serial_nr_next = 1;
190 
191 /*
192  * These bitmasks identify subsystems with specific features to avoid
193  * having to do iterative checks repeatedly.
194  */
195 static u16 have_fork_callback __read_mostly;
196 static u16 have_exit_callback __read_mostly;
197 static u16 have_release_callback __read_mostly;
198 static u16 have_canfork_callback __read_mostly;
199 
200 /* cgroup namespace for init task */
201 struct cgroup_namespace init_cgroup_ns = {
202 	.ns.count	= REFCOUNT_INIT(2),
203 	.user_ns	= &init_user_ns,
204 	.ns.ops		= &cgroupns_operations,
205 	.ns.inum	= PROC_CGROUP_INIT_INO,
206 	.root_cset	= &init_css_set,
207 };
208 
209 static struct file_system_type cgroup2_fs_type;
210 static struct cftype cgroup_base_files[];
211 
212 static int cgroup_apply_control(struct cgroup *cgrp);
213 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
214 static void css_task_iter_skip(struct css_task_iter *it,
215 			       struct task_struct *task);
216 static int cgroup_destroy_locked(struct cgroup *cgrp);
217 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
218 					      struct cgroup_subsys *ss);
219 static void css_release(struct percpu_ref *ref);
220 static void kill_css(struct cgroup_subsys_state *css);
221 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
222 			      struct cgroup *cgrp, struct cftype cfts[],
223 			      bool is_add);
224 
225 /**
226  * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
227  * @ssid: subsys ID of interest
228  *
229  * cgroup_subsys_enabled() can only be used with literal subsys names which
230  * is fine for individual subsystems but unsuitable for cgroup core.  This
231  * is slower static_key_enabled() based test indexed by @ssid.
232  */
233 bool cgroup_ssid_enabled(int ssid)
234 {
235 	if (CGROUP_SUBSYS_COUNT == 0)
236 		return false;
237 
238 	return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
239 }
240 
241 /**
242  * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
243  * @cgrp: the cgroup of interest
244  *
245  * The default hierarchy is the v2 interface of cgroup and this function
246  * can be used to test whether a cgroup is on the default hierarchy for
247  * cases where a subsystem should behave differently depending on the
248  * interface version.
249  *
250  * List of changed behaviors:
251  *
252  * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
253  *   and "name" are disallowed.
254  *
255  * - When mounting an existing superblock, mount options should match.
256  *
257  * - Remount is disallowed.
258  *
259  * - rename(2) is disallowed.
260  *
261  * - "tasks" is removed.  Everything should be at process granularity.  Use
262  *   "cgroup.procs" instead.
263  *
264  * - "cgroup.procs" is not sorted.  pids will be unique unless they got
265  *   recycled in-between reads.
266  *
267  * - "release_agent" and "notify_on_release" are removed.  Replacement
268  *   notification mechanism will be implemented.
269  *
270  * - "cgroup.clone_children" is removed.
271  *
272  * - "cgroup.subtree_populated" is available.  Its value is 0 if the cgroup
273  *   and its descendants contain no task; otherwise, 1.  The file also
274  *   generates kernfs notification which can be monitored through poll and
275  *   [di]notify when the value of the file changes.
276  *
277  * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
278  *   take masks of ancestors with non-empty cpus/mems, instead of being
279  *   moved to an ancestor.
280  *
281  * - cpuset: a task can be moved into an empty cpuset, and again it takes
282  *   masks of ancestors.
283  *
284  * - blkcg: blk-throttle becomes properly hierarchical.
285  *
286  * - debug: disallowed on the default hierarchy.
287  */
288 bool cgroup_on_dfl(const struct cgroup *cgrp)
289 {
290 	return cgrp->root == &cgrp_dfl_root;
291 }
292 
293 /* IDR wrappers which synchronize using cgroup_idr_lock */
294 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
295 			    gfp_t gfp_mask)
296 {
297 	int ret;
298 
299 	idr_preload(gfp_mask);
300 	spin_lock_bh(&cgroup_idr_lock);
301 	ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
302 	spin_unlock_bh(&cgroup_idr_lock);
303 	idr_preload_end();
304 	return ret;
305 }
306 
307 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
308 {
309 	void *ret;
310 
311 	spin_lock_bh(&cgroup_idr_lock);
312 	ret = idr_replace(idr, ptr, id);
313 	spin_unlock_bh(&cgroup_idr_lock);
314 	return ret;
315 }
316 
317 static void cgroup_idr_remove(struct idr *idr, int id)
318 {
319 	spin_lock_bh(&cgroup_idr_lock);
320 	idr_remove(idr, id);
321 	spin_unlock_bh(&cgroup_idr_lock);
322 }
323 
324 static bool cgroup_has_tasks(struct cgroup *cgrp)
325 {
326 	return cgrp->nr_populated_csets;
327 }
328 
329 bool cgroup_is_threaded(struct cgroup *cgrp)
330 {
331 	return cgrp->dom_cgrp != cgrp;
332 }
333 
334 /* can @cgrp host both domain and threaded children? */
335 static bool cgroup_is_mixable(struct cgroup *cgrp)
336 {
337 	/*
338 	 * Root isn't under domain level resource control exempting it from
339 	 * the no-internal-process constraint, so it can serve as a thread
340 	 * root and a parent of resource domains at the same time.
341 	 */
342 	return !cgroup_parent(cgrp);
343 }
344 
345 /* can @cgrp become a thread root? Should always be true for a thread root */
346 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
347 {
348 	/* mixables don't care */
349 	if (cgroup_is_mixable(cgrp))
350 		return true;
351 
352 	/* domain roots can't be nested under threaded */
353 	if (cgroup_is_threaded(cgrp))
354 		return false;
355 
356 	/* can only have either domain or threaded children */
357 	if (cgrp->nr_populated_domain_children)
358 		return false;
359 
360 	/* and no domain controllers can be enabled */
361 	if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
362 		return false;
363 
364 	return true;
365 }
366 
367 /* is @cgrp root of a threaded subtree? */
368 bool cgroup_is_thread_root(struct cgroup *cgrp)
369 {
370 	/* thread root should be a domain */
371 	if (cgroup_is_threaded(cgrp))
372 		return false;
373 
374 	/* a domain w/ threaded children is a thread root */
375 	if (cgrp->nr_threaded_children)
376 		return true;
377 
378 	/*
379 	 * A domain which has tasks and explicit threaded controllers
380 	 * enabled is a thread root.
381 	 */
382 	if (cgroup_has_tasks(cgrp) &&
383 	    (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
384 		return true;
385 
386 	return false;
387 }
388 
389 /* a domain which isn't connected to the root w/o brekage can't be used */
390 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
391 {
392 	/* the cgroup itself can be a thread root */
393 	if (cgroup_is_threaded(cgrp))
394 		return false;
395 
396 	/* but the ancestors can't be unless mixable */
397 	while ((cgrp = cgroup_parent(cgrp))) {
398 		if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
399 			return false;
400 		if (cgroup_is_threaded(cgrp))
401 			return false;
402 	}
403 
404 	return true;
405 }
406 
407 /* subsystems visibly enabled on a cgroup */
408 static u16 cgroup_control(struct cgroup *cgrp)
409 {
410 	struct cgroup *parent = cgroup_parent(cgrp);
411 	u16 root_ss_mask = cgrp->root->subsys_mask;
412 
413 	if (parent) {
414 		u16 ss_mask = parent->subtree_control;
415 
416 		/* threaded cgroups can only have threaded controllers */
417 		if (cgroup_is_threaded(cgrp))
418 			ss_mask &= cgrp_dfl_threaded_ss_mask;
419 		return ss_mask;
420 	}
421 
422 	if (cgroup_on_dfl(cgrp))
423 		root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
424 				  cgrp_dfl_implicit_ss_mask);
425 	return root_ss_mask;
426 }
427 
428 /* subsystems enabled on a cgroup */
429 static u16 cgroup_ss_mask(struct cgroup *cgrp)
430 {
431 	struct cgroup *parent = cgroup_parent(cgrp);
432 
433 	if (parent) {
434 		u16 ss_mask = parent->subtree_ss_mask;
435 
436 		/* threaded cgroups can only have threaded controllers */
437 		if (cgroup_is_threaded(cgrp))
438 			ss_mask &= cgrp_dfl_threaded_ss_mask;
439 		return ss_mask;
440 	}
441 
442 	return cgrp->root->subsys_mask;
443 }
444 
445 /**
446  * cgroup_css - obtain a cgroup's css for the specified subsystem
447  * @cgrp: the cgroup of interest
448  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
449  *
450  * Return @cgrp's css (cgroup_subsys_state) associated with @ss.  This
451  * function must be called either under cgroup_mutex or rcu_read_lock() and
452  * the caller is responsible for pinning the returned css if it wants to
453  * keep accessing it outside the said locks.  This function may return
454  * %NULL if @cgrp doesn't have @subsys_id enabled.
455  */
456 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
457 					      struct cgroup_subsys *ss)
458 {
459 	if (ss)
460 		return rcu_dereference_check(cgrp->subsys[ss->id],
461 					lockdep_is_held(&cgroup_mutex));
462 	else
463 		return &cgrp->self;
464 }
465 
466 /**
467  * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
468  * @cgrp: the cgroup of interest
469  * @ss: the subsystem of interest
470  *
471  * Find and get @cgrp's css associated with @ss.  If the css doesn't exist
472  * or is offline, %NULL is returned.
473  */
474 static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
475 						     struct cgroup_subsys *ss)
476 {
477 	struct cgroup_subsys_state *css;
478 
479 	rcu_read_lock();
480 	css = cgroup_css(cgrp, ss);
481 	if (css && !css_tryget_online(css))
482 		css = NULL;
483 	rcu_read_unlock();
484 
485 	return css;
486 }
487 
488 /**
489  * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
490  * @cgrp: the cgroup of interest
491  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
492  *
493  * Similar to cgroup_css() but returns the effective css, which is defined
494  * as the matching css of the nearest ancestor including self which has @ss
495  * enabled.  If @ss is associated with the hierarchy @cgrp is on, this
496  * function is guaranteed to return non-NULL css.
497  */
498 static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
499 							struct cgroup_subsys *ss)
500 {
501 	lockdep_assert_held(&cgroup_mutex);
502 
503 	if (!ss)
504 		return &cgrp->self;
505 
506 	/*
507 	 * This function is used while updating css associations and thus
508 	 * can't test the csses directly.  Test ss_mask.
509 	 */
510 	while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
511 		cgrp = cgroup_parent(cgrp);
512 		if (!cgrp)
513 			return NULL;
514 	}
515 
516 	return cgroup_css(cgrp, ss);
517 }
518 
519 /**
520  * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
521  * @cgrp: the cgroup of interest
522  * @ss: the subsystem of interest
523  *
524  * Find and get the effective css of @cgrp for @ss.  The effective css is
525  * defined as the matching css of the nearest ancestor including self which
526  * has @ss enabled.  If @ss is not mounted on the hierarchy @cgrp is on,
527  * the root css is returned, so this function always returns a valid css.
528  *
529  * The returned css is not guaranteed to be online, and therefore it is the
530  * callers responsibility to try get a reference for it.
531  */
532 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
533 					 struct cgroup_subsys *ss)
534 {
535 	struct cgroup_subsys_state *css;
536 
537 	do {
538 		css = cgroup_css(cgrp, ss);
539 
540 		if (css)
541 			return css;
542 		cgrp = cgroup_parent(cgrp);
543 	} while (cgrp);
544 
545 	return init_css_set.subsys[ss->id];
546 }
547 
548 /**
549  * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
550  * @cgrp: the cgroup of interest
551  * @ss: the subsystem of interest
552  *
553  * Find and get the effective css of @cgrp for @ss.  The effective css is
554  * defined as the matching css of the nearest ancestor including self which
555  * has @ss enabled.  If @ss is not mounted on the hierarchy @cgrp is on,
556  * the root css is returned, so this function always returns a valid css.
557  * The returned css must be put using css_put().
558  */
559 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
560 					     struct cgroup_subsys *ss)
561 {
562 	struct cgroup_subsys_state *css;
563 
564 	rcu_read_lock();
565 
566 	do {
567 		css = cgroup_css(cgrp, ss);
568 
569 		if (css && css_tryget_online(css))
570 			goto out_unlock;
571 		cgrp = cgroup_parent(cgrp);
572 	} while (cgrp);
573 
574 	css = init_css_set.subsys[ss->id];
575 	css_get(css);
576 out_unlock:
577 	rcu_read_unlock();
578 	return css;
579 }
580 EXPORT_SYMBOL_GPL(cgroup_get_e_css);
581 
582 static void cgroup_get_live(struct cgroup *cgrp)
583 {
584 	WARN_ON_ONCE(cgroup_is_dead(cgrp));
585 	css_get(&cgrp->self);
586 }
587 
588 /**
589  * __cgroup_task_count - count the number of tasks in a cgroup. The caller
590  * is responsible for taking the css_set_lock.
591  * @cgrp: the cgroup in question
592  */
593 int __cgroup_task_count(const struct cgroup *cgrp)
594 {
595 	int count = 0;
596 	struct cgrp_cset_link *link;
597 
598 	lockdep_assert_held(&css_set_lock);
599 
600 	list_for_each_entry(link, &cgrp->cset_links, cset_link)
601 		count += link->cset->nr_tasks;
602 
603 	return count;
604 }
605 
606 /**
607  * cgroup_task_count - count the number of tasks in a cgroup.
608  * @cgrp: the cgroup in question
609  */
610 int cgroup_task_count(const struct cgroup *cgrp)
611 {
612 	int count;
613 
614 	spin_lock_irq(&css_set_lock);
615 	count = __cgroup_task_count(cgrp);
616 	spin_unlock_irq(&css_set_lock);
617 
618 	return count;
619 }
620 
621 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
622 {
623 	struct cgroup *cgrp = of->kn->parent->priv;
624 	struct cftype *cft = of_cft(of);
625 
626 	/*
627 	 * This is open and unprotected implementation of cgroup_css().
628 	 * seq_css() is only called from a kernfs file operation which has
629 	 * an active reference on the file.  Because all the subsystem
630 	 * files are drained before a css is disassociated with a cgroup,
631 	 * the matching css from the cgroup's subsys table is guaranteed to
632 	 * be and stay valid until the enclosing operation is complete.
633 	 */
634 	if (cft->ss)
635 		return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
636 	else
637 		return &cgrp->self;
638 }
639 EXPORT_SYMBOL_GPL(of_css);
640 
641 /**
642  * for_each_css - iterate all css's of a cgroup
643  * @css: the iteration cursor
644  * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
645  * @cgrp: the target cgroup to iterate css's of
646  *
647  * Should be called under cgroup_[tree_]mutex.
648  */
649 #define for_each_css(css, ssid, cgrp)					\
650 	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++)	\
651 		if (!((css) = rcu_dereference_check(			\
652 				(cgrp)->subsys[(ssid)],			\
653 				lockdep_is_held(&cgroup_mutex)))) { }	\
654 		else
655 
656 /**
657  * for_each_e_css - iterate all effective css's of a cgroup
658  * @css: the iteration cursor
659  * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
660  * @cgrp: the target cgroup to iterate css's of
661  *
662  * Should be called under cgroup_[tree_]mutex.
663  */
664 #define for_each_e_css(css, ssid, cgrp)					    \
665 	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++)	    \
666 		if (!((css) = cgroup_e_css_by_mask(cgrp,		    \
667 						   cgroup_subsys[(ssid)]))) \
668 			;						    \
669 		else
670 
671 /**
672  * do_each_subsys_mask - filter for_each_subsys with a bitmask
673  * @ss: the iteration cursor
674  * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
675  * @ss_mask: the bitmask
676  *
677  * The block will only run for cases where the ssid-th bit (1 << ssid) of
678  * @ss_mask is set.
679  */
680 #define do_each_subsys_mask(ss, ssid, ss_mask) do {			\
681 	unsigned long __ss_mask = (ss_mask);				\
682 	if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */	\
683 		(ssid) = 0;						\
684 		break;							\
685 	}								\
686 	for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) {	\
687 		(ss) = cgroup_subsys[ssid];				\
688 		{
689 
690 #define while_each_subsys_mask()					\
691 		}							\
692 	}								\
693 } while (false)
694 
695 /* iterate over child cgrps, lock should be held throughout iteration */
696 #define cgroup_for_each_live_child(child, cgrp)				\
697 	list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
698 		if (({ lockdep_assert_held(&cgroup_mutex);		\
699 		       cgroup_is_dead(child); }))			\
700 			;						\
701 		else
702 
703 /* walk live descendants in pre order */
704 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)		\
705 	css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL))	\
706 		if (({ lockdep_assert_held(&cgroup_mutex);		\
707 		       (dsct) = (d_css)->cgroup;			\
708 		       cgroup_is_dead(dsct); }))			\
709 			;						\
710 		else
711 
712 /* walk live descendants in postorder */
713 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp)		\
714 	css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL))	\
715 		if (({ lockdep_assert_held(&cgroup_mutex);		\
716 		       (dsct) = (d_css)->cgroup;			\
717 		       cgroup_is_dead(dsct); }))			\
718 			;						\
719 		else
720 
721 /*
722  * The default css_set - used by init and its children prior to any
723  * hierarchies being mounted. It contains a pointer to the root state
724  * for each subsystem. Also used to anchor the list of css_sets. Not
725  * reference-counted, to improve performance when child cgroups
726  * haven't been created.
727  */
728 struct css_set init_css_set = {
729 	.refcount		= REFCOUNT_INIT(1),
730 	.dom_cset		= &init_css_set,
731 	.tasks			= LIST_HEAD_INIT(init_css_set.tasks),
732 	.mg_tasks		= LIST_HEAD_INIT(init_css_set.mg_tasks),
733 	.dying_tasks		= LIST_HEAD_INIT(init_css_set.dying_tasks),
734 	.task_iters		= LIST_HEAD_INIT(init_css_set.task_iters),
735 	.threaded_csets		= LIST_HEAD_INIT(init_css_set.threaded_csets),
736 	.cgrp_links		= LIST_HEAD_INIT(init_css_set.cgrp_links),
737 	.mg_preload_node	= LIST_HEAD_INIT(init_css_set.mg_preload_node),
738 	.mg_node		= LIST_HEAD_INIT(init_css_set.mg_node),
739 
740 	/*
741 	 * The following field is re-initialized when this cset gets linked
742 	 * in cgroup_init().  However, let's initialize the field
743 	 * statically too so that the default cgroup can be accessed safely
744 	 * early during boot.
745 	 */
746 	.dfl_cgrp		= &cgrp_dfl_root.cgrp,
747 };
748 
749 static int css_set_count	= 1;	/* 1 for init_css_set */
750 
751 static bool css_set_threaded(struct css_set *cset)
752 {
753 	return cset->dom_cset != cset;
754 }
755 
756 /**
757  * css_set_populated - does a css_set contain any tasks?
758  * @cset: target css_set
759  *
760  * css_set_populated() should be the same as !!cset->nr_tasks at steady
761  * state. However, css_set_populated() can be called while a task is being
762  * added to or removed from the linked list before the nr_tasks is
763  * properly updated. Hence, we can't just look at ->nr_tasks here.
764  */
765 static bool css_set_populated(struct css_set *cset)
766 {
767 	lockdep_assert_held(&css_set_lock);
768 
769 	return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
770 }
771 
772 /**
773  * cgroup_update_populated - update the populated count of a cgroup
774  * @cgrp: the target cgroup
775  * @populated: inc or dec populated count
776  *
777  * One of the css_sets associated with @cgrp is either getting its first
778  * task or losing the last.  Update @cgrp->nr_populated_* accordingly.  The
779  * count is propagated towards root so that a given cgroup's
780  * nr_populated_children is zero iff none of its descendants contain any
781  * tasks.
782  *
783  * @cgrp's interface file "cgroup.populated" is zero if both
784  * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
785  * 1 otherwise.  When the sum changes from or to zero, userland is notified
786  * that the content of the interface file has changed.  This can be used to
787  * detect when @cgrp and its descendants become populated or empty.
788  */
789 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
790 {
791 	struct cgroup *child = NULL;
792 	int adj = populated ? 1 : -1;
793 
794 	lockdep_assert_held(&css_set_lock);
795 
796 	do {
797 		bool was_populated = cgroup_is_populated(cgrp);
798 
799 		if (!child) {
800 			cgrp->nr_populated_csets += adj;
801 		} else {
802 			if (cgroup_is_threaded(child))
803 				cgrp->nr_populated_threaded_children += adj;
804 			else
805 				cgrp->nr_populated_domain_children += adj;
806 		}
807 
808 		if (was_populated == cgroup_is_populated(cgrp))
809 			break;
810 
811 		cgroup1_check_for_release(cgrp);
812 		TRACE_CGROUP_PATH(notify_populated, cgrp,
813 				  cgroup_is_populated(cgrp));
814 		cgroup_file_notify(&cgrp->events_file);
815 
816 		child = cgrp;
817 		cgrp = cgroup_parent(cgrp);
818 	} while (cgrp);
819 }
820 
821 /**
822  * css_set_update_populated - update populated state of a css_set
823  * @cset: target css_set
824  * @populated: whether @cset is populated or depopulated
825  *
826  * @cset is either getting the first task or losing the last.  Update the
827  * populated counters of all associated cgroups accordingly.
828  */
829 static void css_set_update_populated(struct css_set *cset, bool populated)
830 {
831 	struct cgrp_cset_link *link;
832 
833 	lockdep_assert_held(&css_set_lock);
834 
835 	list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
836 		cgroup_update_populated(link->cgrp, populated);
837 }
838 
839 /*
840  * @task is leaving, advance task iterators which are pointing to it so
841  * that they can resume at the next position.  Advancing an iterator might
842  * remove it from the list, use safe walk.  See css_task_iter_skip() for
843  * details.
844  */
845 static void css_set_skip_task_iters(struct css_set *cset,
846 				    struct task_struct *task)
847 {
848 	struct css_task_iter *it, *pos;
849 
850 	list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
851 		css_task_iter_skip(it, task);
852 }
853 
854 /**
855  * css_set_move_task - move a task from one css_set to another
856  * @task: task being moved
857  * @from_cset: css_set @task currently belongs to (may be NULL)
858  * @to_cset: new css_set @task is being moved to (may be NULL)
859  * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
860  *
861  * Move @task from @from_cset to @to_cset.  If @task didn't belong to any
862  * css_set, @from_cset can be NULL.  If @task is being disassociated
863  * instead of moved, @to_cset can be NULL.
864  *
865  * This function automatically handles populated counter updates and
866  * css_task_iter adjustments but the caller is responsible for managing
867  * @from_cset and @to_cset's reference counts.
868  */
869 static void css_set_move_task(struct task_struct *task,
870 			      struct css_set *from_cset, struct css_set *to_cset,
871 			      bool use_mg_tasks)
872 {
873 	lockdep_assert_held(&css_set_lock);
874 
875 	if (to_cset && !css_set_populated(to_cset))
876 		css_set_update_populated(to_cset, true);
877 
878 	if (from_cset) {
879 		WARN_ON_ONCE(list_empty(&task->cg_list));
880 
881 		css_set_skip_task_iters(from_cset, task);
882 		list_del_init(&task->cg_list);
883 		if (!css_set_populated(from_cset))
884 			css_set_update_populated(from_cset, false);
885 	} else {
886 		WARN_ON_ONCE(!list_empty(&task->cg_list));
887 	}
888 
889 	if (to_cset) {
890 		/*
891 		 * We are synchronized through cgroup_threadgroup_rwsem
892 		 * against PF_EXITING setting such that we can't race
893 		 * against cgroup_exit()/cgroup_free() dropping the css_set.
894 		 */
895 		WARN_ON_ONCE(task->flags & PF_EXITING);
896 
897 		cgroup_move_task(task, to_cset);
898 		list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
899 							     &to_cset->tasks);
900 	}
901 }
902 
903 /*
904  * hash table for cgroup groups. This improves the performance to find
905  * an existing css_set. This hash doesn't (currently) take into
906  * account cgroups in empty hierarchies.
907  */
908 #define CSS_SET_HASH_BITS	7
909 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
910 
911 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
912 {
913 	unsigned long key = 0UL;
914 	struct cgroup_subsys *ss;
915 	int i;
916 
917 	for_each_subsys(ss, i)
918 		key += (unsigned long)css[i];
919 	key = (key >> 16) ^ key;
920 
921 	return key;
922 }
923 
924 void put_css_set_locked(struct css_set *cset)
925 {
926 	struct cgrp_cset_link *link, *tmp_link;
927 	struct cgroup_subsys *ss;
928 	int ssid;
929 
930 	lockdep_assert_held(&css_set_lock);
931 
932 	if (!refcount_dec_and_test(&cset->refcount))
933 		return;
934 
935 	WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
936 
937 	/* This css_set is dead. Unlink it and release cgroup and css refs */
938 	for_each_subsys(ss, ssid) {
939 		list_del(&cset->e_cset_node[ssid]);
940 		css_put(cset->subsys[ssid]);
941 	}
942 	hash_del(&cset->hlist);
943 	css_set_count--;
944 
945 	list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
946 		list_del(&link->cset_link);
947 		list_del(&link->cgrp_link);
948 		if (cgroup_parent(link->cgrp))
949 			cgroup_put(link->cgrp);
950 		kfree(link);
951 	}
952 
953 	if (css_set_threaded(cset)) {
954 		list_del(&cset->threaded_csets_node);
955 		put_css_set_locked(cset->dom_cset);
956 	}
957 
958 	kfree_rcu(cset, rcu_head);
959 }
960 
961 /**
962  * compare_css_sets - helper function for find_existing_css_set().
963  * @cset: candidate css_set being tested
964  * @old_cset: existing css_set for a task
965  * @new_cgrp: cgroup that's being entered by the task
966  * @template: desired set of css pointers in css_set (pre-calculated)
967  *
968  * Returns true if "cset" matches "old_cset" except for the hierarchy
969  * which "new_cgrp" belongs to, for which it should match "new_cgrp".
970  */
971 static bool compare_css_sets(struct css_set *cset,
972 			     struct css_set *old_cset,
973 			     struct cgroup *new_cgrp,
974 			     struct cgroup_subsys_state *template[])
975 {
976 	struct cgroup *new_dfl_cgrp;
977 	struct list_head *l1, *l2;
978 
979 	/*
980 	 * On the default hierarchy, there can be csets which are
981 	 * associated with the same set of cgroups but different csses.
982 	 * Let's first ensure that csses match.
983 	 */
984 	if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
985 		return false;
986 
987 
988 	/* @cset's domain should match the default cgroup's */
989 	if (cgroup_on_dfl(new_cgrp))
990 		new_dfl_cgrp = new_cgrp;
991 	else
992 		new_dfl_cgrp = old_cset->dfl_cgrp;
993 
994 	if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
995 		return false;
996 
997 	/*
998 	 * Compare cgroup pointers in order to distinguish between
999 	 * different cgroups in hierarchies.  As different cgroups may
1000 	 * share the same effective css, this comparison is always
1001 	 * necessary.
1002 	 */
1003 	l1 = &cset->cgrp_links;
1004 	l2 = &old_cset->cgrp_links;
1005 	while (1) {
1006 		struct cgrp_cset_link *link1, *link2;
1007 		struct cgroup *cgrp1, *cgrp2;
1008 
1009 		l1 = l1->next;
1010 		l2 = l2->next;
1011 		/* See if we reached the end - both lists are equal length. */
1012 		if (l1 == &cset->cgrp_links) {
1013 			BUG_ON(l2 != &old_cset->cgrp_links);
1014 			break;
1015 		} else {
1016 			BUG_ON(l2 == &old_cset->cgrp_links);
1017 		}
1018 		/* Locate the cgroups associated with these links. */
1019 		link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1020 		link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1021 		cgrp1 = link1->cgrp;
1022 		cgrp2 = link2->cgrp;
1023 		/* Hierarchies should be linked in the same order. */
1024 		BUG_ON(cgrp1->root != cgrp2->root);
1025 
1026 		/*
1027 		 * If this hierarchy is the hierarchy of the cgroup
1028 		 * that's changing, then we need to check that this
1029 		 * css_set points to the new cgroup; if it's any other
1030 		 * hierarchy, then this css_set should point to the
1031 		 * same cgroup as the old css_set.
1032 		 */
1033 		if (cgrp1->root == new_cgrp->root) {
1034 			if (cgrp1 != new_cgrp)
1035 				return false;
1036 		} else {
1037 			if (cgrp1 != cgrp2)
1038 				return false;
1039 		}
1040 	}
1041 	return true;
1042 }
1043 
1044 /**
1045  * find_existing_css_set - init css array and find the matching css_set
1046  * @old_cset: the css_set that we're using before the cgroup transition
1047  * @cgrp: the cgroup that we're moving into
1048  * @template: out param for the new set of csses, should be clear on entry
1049  */
1050 static struct css_set *find_existing_css_set(struct css_set *old_cset,
1051 					struct cgroup *cgrp,
1052 					struct cgroup_subsys_state *template[])
1053 {
1054 	struct cgroup_root *root = cgrp->root;
1055 	struct cgroup_subsys *ss;
1056 	struct css_set *cset;
1057 	unsigned long key;
1058 	int i;
1059 
1060 	/*
1061 	 * Build the set of subsystem state objects that we want to see in the
1062 	 * new css_set. While subsystems can change globally, the entries here
1063 	 * won't change, so no need for locking.
1064 	 */
1065 	for_each_subsys(ss, i) {
1066 		if (root->subsys_mask & (1UL << i)) {
1067 			/*
1068 			 * @ss is in this hierarchy, so we want the
1069 			 * effective css from @cgrp.
1070 			 */
1071 			template[i] = cgroup_e_css_by_mask(cgrp, ss);
1072 		} else {
1073 			/*
1074 			 * @ss is not in this hierarchy, so we don't want
1075 			 * to change the css.
1076 			 */
1077 			template[i] = old_cset->subsys[i];
1078 		}
1079 	}
1080 
1081 	key = css_set_hash(template);
1082 	hash_for_each_possible(css_set_table, cset, hlist, key) {
1083 		if (!compare_css_sets(cset, old_cset, cgrp, template))
1084 			continue;
1085 
1086 		/* This css_set matches what we need */
1087 		return cset;
1088 	}
1089 
1090 	/* No existing cgroup group matched */
1091 	return NULL;
1092 }
1093 
1094 static void free_cgrp_cset_links(struct list_head *links_to_free)
1095 {
1096 	struct cgrp_cset_link *link, *tmp_link;
1097 
1098 	list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1099 		list_del(&link->cset_link);
1100 		kfree(link);
1101 	}
1102 }
1103 
1104 /**
1105  * allocate_cgrp_cset_links - allocate cgrp_cset_links
1106  * @count: the number of links to allocate
1107  * @tmp_links: list_head the allocated links are put on
1108  *
1109  * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1110  * through ->cset_link.  Returns 0 on success or -errno.
1111  */
1112 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1113 {
1114 	struct cgrp_cset_link *link;
1115 	int i;
1116 
1117 	INIT_LIST_HEAD(tmp_links);
1118 
1119 	for (i = 0; i < count; i++) {
1120 		link = kzalloc(sizeof(*link), GFP_KERNEL);
1121 		if (!link) {
1122 			free_cgrp_cset_links(tmp_links);
1123 			return -ENOMEM;
1124 		}
1125 		list_add(&link->cset_link, tmp_links);
1126 	}
1127 	return 0;
1128 }
1129 
1130 /**
1131  * link_css_set - a helper function to link a css_set to a cgroup
1132  * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1133  * @cset: the css_set to be linked
1134  * @cgrp: the destination cgroup
1135  */
1136 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1137 			 struct cgroup *cgrp)
1138 {
1139 	struct cgrp_cset_link *link;
1140 
1141 	BUG_ON(list_empty(tmp_links));
1142 
1143 	if (cgroup_on_dfl(cgrp))
1144 		cset->dfl_cgrp = cgrp;
1145 
1146 	link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1147 	link->cset = cset;
1148 	link->cgrp = cgrp;
1149 
1150 	/*
1151 	 * Always add links to the tail of the lists so that the lists are
1152 	 * in chronological order.
1153 	 */
1154 	list_move_tail(&link->cset_link, &cgrp->cset_links);
1155 	list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1156 
1157 	if (cgroup_parent(cgrp))
1158 		cgroup_get_live(cgrp);
1159 }
1160 
1161 /**
1162  * find_css_set - return a new css_set with one cgroup updated
1163  * @old_cset: the baseline css_set
1164  * @cgrp: the cgroup to be updated
1165  *
1166  * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1167  * substituted into the appropriate hierarchy.
1168  */
1169 static struct css_set *find_css_set(struct css_set *old_cset,
1170 				    struct cgroup *cgrp)
1171 {
1172 	struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1173 	struct css_set *cset;
1174 	struct list_head tmp_links;
1175 	struct cgrp_cset_link *link;
1176 	struct cgroup_subsys *ss;
1177 	unsigned long key;
1178 	int ssid;
1179 
1180 	lockdep_assert_held(&cgroup_mutex);
1181 
1182 	/* First see if we already have a cgroup group that matches
1183 	 * the desired set */
1184 	spin_lock_irq(&css_set_lock);
1185 	cset = find_existing_css_set(old_cset, cgrp, template);
1186 	if (cset)
1187 		get_css_set(cset);
1188 	spin_unlock_irq(&css_set_lock);
1189 
1190 	if (cset)
1191 		return cset;
1192 
1193 	cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1194 	if (!cset)
1195 		return NULL;
1196 
1197 	/* Allocate all the cgrp_cset_link objects that we'll need */
1198 	if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1199 		kfree(cset);
1200 		return NULL;
1201 	}
1202 
1203 	refcount_set(&cset->refcount, 1);
1204 	cset->dom_cset = cset;
1205 	INIT_LIST_HEAD(&cset->tasks);
1206 	INIT_LIST_HEAD(&cset->mg_tasks);
1207 	INIT_LIST_HEAD(&cset->dying_tasks);
1208 	INIT_LIST_HEAD(&cset->task_iters);
1209 	INIT_LIST_HEAD(&cset->threaded_csets);
1210 	INIT_HLIST_NODE(&cset->hlist);
1211 	INIT_LIST_HEAD(&cset->cgrp_links);
1212 	INIT_LIST_HEAD(&cset->mg_preload_node);
1213 	INIT_LIST_HEAD(&cset->mg_node);
1214 
1215 	/* Copy the set of subsystem state objects generated in
1216 	 * find_existing_css_set() */
1217 	memcpy(cset->subsys, template, sizeof(cset->subsys));
1218 
1219 	spin_lock_irq(&css_set_lock);
1220 	/* Add reference counts and links from the new css_set. */
1221 	list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1222 		struct cgroup *c = link->cgrp;
1223 
1224 		if (c->root == cgrp->root)
1225 			c = cgrp;
1226 		link_css_set(&tmp_links, cset, c);
1227 	}
1228 
1229 	BUG_ON(!list_empty(&tmp_links));
1230 
1231 	css_set_count++;
1232 
1233 	/* Add @cset to the hash table */
1234 	key = css_set_hash(cset->subsys);
1235 	hash_add(css_set_table, &cset->hlist, key);
1236 
1237 	for_each_subsys(ss, ssid) {
1238 		struct cgroup_subsys_state *css = cset->subsys[ssid];
1239 
1240 		list_add_tail(&cset->e_cset_node[ssid],
1241 			      &css->cgroup->e_csets[ssid]);
1242 		css_get(css);
1243 	}
1244 
1245 	spin_unlock_irq(&css_set_lock);
1246 
1247 	/*
1248 	 * If @cset should be threaded, look up the matching dom_cset and
1249 	 * link them up.  We first fully initialize @cset then look for the
1250 	 * dom_cset.  It's simpler this way and safe as @cset is guaranteed
1251 	 * to stay empty until we return.
1252 	 */
1253 	if (cgroup_is_threaded(cset->dfl_cgrp)) {
1254 		struct css_set *dcset;
1255 
1256 		dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1257 		if (!dcset) {
1258 			put_css_set(cset);
1259 			return NULL;
1260 		}
1261 
1262 		spin_lock_irq(&css_set_lock);
1263 		cset->dom_cset = dcset;
1264 		list_add_tail(&cset->threaded_csets_node,
1265 			      &dcset->threaded_csets);
1266 		spin_unlock_irq(&css_set_lock);
1267 	}
1268 
1269 	return cset;
1270 }
1271 
1272 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1273 {
1274 	struct cgroup *root_cgrp = kf_root->kn->priv;
1275 
1276 	return root_cgrp->root;
1277 }
1278 
1279 static int cgroup_init_root_id(struct cgroup_root *root)
1280 {
1281 	int id;
1282 
1283 	lockdep_assert_held(&cgroup_mutex);
1284 
1285 	id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1286 	if (id < 0)
1287 		return id;
1288 
1289 	root->hierarchy_id = id;
1290 	return 0;
1291 }
1292 
1293 static void cgroup_exit_root_id(struct cgroup_root *root)
1294 {
1295 	lockdep_assert_held(&cgroup_mutex);
1296 
1297 	idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1298 }
1299 
1300 void cgroup_free_root(struct cgroup_root *root)
1301 {
1302 	kfree(root);
1303 }
1304 
1305 static void cgroup_destroy_root(struct cgroup_root *root)
1306 {
1307 	struct cgroup *cgrp = &root->cgrp;
1308 	struct cgrp_cset_link *link, *tmp_link;
1309 
1310 	trace_cgroup_destroy_root(root);
1311 
1312 	cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1313 
1314 	BUG_ON(atomic_read(&root->nr_cgrps));
1315 	BUG_ON(!list_empty(&cgrp->self.children));
1316 
1317 	/* Rebind all subsystems back to the default hierarchy */
1318 	WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1319 
1320 	/*
1321 	 * Release all the links from cset_links to this hierarchy's
1322 	 * root cgroup
1323 	 */
1324 	spin_lock_irq(&css_set_lock);
1325 
1326 	list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1327 		list_del(&link->cset_link);
1328 		list_del(&link->cgrp_link);
1329 		kfree(link);
1330 	}
1331 
1332 	spin_unlock_irq(&css_set_lock);
1333 
1334 	if (!list_empty(&root->root_list)) {
1335 		list_del(&root->root_list);
1336 		cgroup_root_count--;
1337 	}
1338 
1339 	cgroup_exit_root_id(root);
1340 
1341 	mutex_unlock(&cgroup_mutex);
1342 
1343 	cgroup_rstat_exit(cgrp);
1344 	kernfs_destroy_root(root->kf_root);
1345 	cgroup_free_root(root);
1346 }
1347 
1348 /*
1349  * look up cgroup associated with current task's cgroup namespace on the
1350  * specified hierarchy
1351  */
1352 static struct cgroup *
1353 current_cgns_cgroup_from_root(struct cgroup_root *root)
1354 {
1355 	struct cgroup *res = NULL;
1356 	struct css_set *cset;
1357 
1358 	lockdep_assert_held(&css_set_lock);
1359 
1360 	rcu_read_lock();
1361 
1362 	cset = current->nsproxy->cgroup_ns->root_cset;
1363 	if (cset == &init_css_set) {
1364 		res = &root->cgrp;
1365 	} else if (root == &cgrp_dfl_root) {
1366 		res = cset->dfl_cgrp;
1367 	} else {
1368 		struct cgrp_cset_link *link;
1369 
1370 		list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1371 			struct cgroup *c = link->cgrp;
1372 
1373 			if (c->root == root) {
1374 				res = c;
1375 				break;
1376 			}
1377 		}
1378 	}
1379 	rcu_read_unlock();
1380 
1381 	BUG_ON(!res);
1382 	return res;
1383 }
1384 
1385 /* look up cgroup associated with given css_set on the specified hierarchy */
1386 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1387 					    struct cgroup_root *root)
1388 {
1389 	struct cgroup *res = NULL;
1390 
1391 	lockdep_assert_held(&cgroup_mutex);
1392 	lockdep_assert_held(&css_set_lock);
1393 
1394 	if (cset == &init_css_set) {
1395 		res = &root->cgrp;
1396 	} else if (root == &cgrp_dfl_root) {
1397 		res = cset->dfl_cgrp;
1398 	} else {
1399 		struct cgrp_cset_link *link;
1400 
1401 		list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1402 			struct cgroup *c = link->cgrp;
1403 
1404 			if (c->root == root) {
1405 				res = c;
1406 				break;
1407 			}
1408 		}
1409 	}
1410 
1411 	BUG_ON(!res);
1412 	return res;
1413 }
1414 
1415 /*
1416  * Return the cgroup for "task" from the given hierarchy. Must be
1417  * called with cgroup_mutex and css_set_lock held.
1418  */
1419 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1420 				     struct cgroup_root *root)
1421 {
1422 	/*
1423 	 * No need to lock the task - since we hold css_set_lock the
1424 	 * task can't change groups.
1425 	 */
1426 	return cset_cgroup_from_root(task_css_set(task), root);
1427 }
1428 
1429 /*
1430  * A task must hold cgroup_mutex to modify cgroups.
1431  *
1432  * Any task can increment and decrement the count field without lock.
1433  * So in general, code holding cgroup_mutex can't rely on the count
1434  * field not changing.  However, if the count goes to zero, then only
1435  * cgroup_attach_task() can increment it again.  Because a count of zero
1436  * means that no tasks are currently attached, therefore there is no
1437  * way a task attached to that cgroup can fork (the other way to
1438  * increment the count).  So code holding cgroup_mutex can safely
1439  * assume that if the count is zero, it will stay zero. Similarly, if
1440  * a task holds cgroup_mutex on a cgroup with zero count, it
1441  * knows that the cgroup won't be removed, as cgroup_rmdir()
1442  * needs that mutex.
1443  *
1444  * A cgroup can only be deleted if both its 'count' of using tasks
1445  * is zero, and its list of 'children' cgroups is empty.  Since all
1446  * tasks in the system use _some_ cgroup, and since there is always at
1447  * least one task in the system (init, pid == 1), therefore, root cgroup
1448  * always has either children cgroups and/or using tasks.  So we don't
1449  * need a special hack to ensure that root cgroup cannot be deleted.
1450  *
1451  * P.S.  One more locking exception.  RCU is used to guard the
1452  * update of a tasks cgroup pointer by cgroup_attach_task()
1453  */
1454 
1455 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1456 
1457 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1458 			      char *buf)
1459 {
1460 	struct cgroup_subsys *ss = cft->ss;
1461 
1462 	if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1463 	    !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1464 		const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1465 
1466 		snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1467 			 dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1468 			 cft->name);
1469 	} else {
1470 		strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1471 	}
1472 	return buf;
1473 }
1474 
1475 /**
1476  * cgroup_file_mode - deduce file mode of a control file
1477  * @cft: the control file in question
1478  *
1479  * S_IRUGO for read, S_IWUSR for write.
1480  */
1481 static umode_t cgroup_file_mode(const struct cftype *cft)
1482 {
1483 	umode_t mode = 0;
1484 
1485 	if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1486 		mode |= S_IRUGO;
1487 
1488 	if (cft->write_u64 || cft->write_s64 || cft->write) {
1489 		if (cft->flags & CFTYPE_WORLD_WRITABLE)
1490 			mode |= S_IWUGO;
1491 		else
1492 			mode |= S_IWUSR;
1493 	}
1494 
1495 	return mode;
1496 }
1497 
1498 /**
1499  * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1500  * @subtree_control: the new subtree_control mask to consider
1501  * @this_ss_mask: available subsystems
1502  *
1503  * On the default hierarchy, a subsystem may request other subsystems to be
1504  * enabled together through its ->depends_on mask.  In such cases, more
1505  * subsystems than specified in "cgroup.subtree_control" may be enabled.
1506  *
1507  * This function calculates which subsystems need to be enabled if
1508  * @subtree_control is to be applied while restricted to @this_ss_mask.
1509  */
1510 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1511 {
1512 	u16 cur_ss_mask = subtree_control;
1513 	struct cgroup_subsys *ss;
1514 	int ssid;
1515 
1516 	lockdep_assert_held(&cgroup_mutex);
1517 
1518 	cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1519 
1520 	while (true) {
1521 		u16 new_ss_mask = cur_ss_mask;
1522 
1523 		do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1524 			new_ss_mask |= ss->depends_on;
1525 		} while_each_subsys_mask();
1526 
1527 		/*
1528 		 * Mask out subsystems which aren't available.  This can
1529 		 * happen only if some depended-upon subsystems were bound
1530 		 * to non-default hierarchies.
1531 		 */
1532 		new_ss_mask &= this_ss_mask;
1533 
1534 		if (new_ss_mask == cur_ss_mask)
1535 			break;
1536 		cur_ss_mask = new_ss_mask;
1537 	}
1538 
1539 	return cur_ss_mask;
1540 }
1541 
1542 /**
1543  * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1544  * @kn: the kernfs_node being serviced
1545  *
1546  * This helper undoes cgroup_kn_lock_live() and should be invoked before
1547  * the method finishes if locking succeeded.  Note that once this function
1548  * returns the cgroup returned by cgroup_kn_lock_live() may become
1549  * inaccessible any time.  If the caller intends to continue to access the
1550  * cgroup, it should pin it before invoking this function.
1551  */
1552 void cgroup_kn_unlock(struct kernfs_node *kn)
1553 {
1554 	struct cgroup *cgrp;
1555 
1556 	if (kernfs_type(kn) == KERNFS_DIR)
1557 		cgrp = kn->priv;
1558 	else
1559 		cgrp = kn->parent->priv;
1560 
1561 	mutex_unlock(&cgroup_mutex);
1562 
1563 	kernfs_unbreak_active_protection(kn);
1564 	cgroup_put(cgrp);
1565 }
1566 
1567 /**
1568  * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1569  * @kn: the kernfs_node being serviced
1570  * @drain_offline: perform offline draining on the cgroup
1571  *
1572  * This helper is to be used by a cgroup kernfs method currently servicing
1573  * @kn.  It breaks the active protection, performs cgroup locking and
1574  * verifies that the associated cgroup is alive.  Returns the cgroup if
1575  * alive; otherwise, %NULL.  A successful return should be undone by a
1576  * matching cgroup_kn_unlock() invocation.  If @drain_offline is %true, the
1577  * cgroup is drained of offlining csses before return.
1578  *
1579  * Any cgroup kernfs method implementation which requires locking the
1580  * associated cgroup should use this helper.  It avoids nesting cgroup
1581  * locking under kernfs active protection and allows all kernfs operations
1582  * including self-removal.
1583  */
1584 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1585 {
1586 	struct cgroup *cgrp;
1587 
1588 	if (kernfs_type(kn) == KERNFS_DIR)
1589 		cgrp = kn->priv;
1590 	else
1591 		cgrp = kn->parent->priv;
1592 
1593 	/*
1594 	 * We're gonna grab cgroup_mutex which nests outside kernfs
1595 	 * active_ref.  cgroup liveliness check alone provides enough
1596 	 * protection against removal.  Ensure @cgrp stays accessible and
1597 	 * break the active_ref protection.
1598 	 */
1599 	if (!cgroup_tryget(cgrp))
1600 		return NULL;
1601 	kernfs_break_active_protection(kn);
1602 
1603 	if (drain_offline)
1604 		cgroup_lock_and_drain_offline(cgrp);
1605 	else
1606 		mutex_lock(&cgroup_mutex);
1607 
1608 	if (!cgroup_is_dead(cgrp))
1609 		return cgrp;
1610 
1611 	cgroup_kn_unlock(kn);
1612 	return NULL;
1613 }
1614 
1615 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1616 {
1617 	char name[CGROUP_FILE_NAME_MAX];
1618 
1619 	lockdep_assert_held(&cgroup_mutex);
1620 
1621 	if (cft->file_offset) {
1622 		struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1623 		struct cgroup_file *cfile = (void *)css + cft->file_offset;
1624 
1625 		spin_lock_irq(&cgroup_file_kn_lock);
1626 		cfile->kn = NULL;
1627 		spin_unlock_irq(&cgroup_file_kn_lock);
1628 
1629 		del_timer_sync(&cfile->notify_timer);
1630 	}
1631 
1632 	kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1633 }
1634 
1635 /**
1636  * css_clear_dir - remove subsys files in a cgroup directory
1637  * @css: target css
1638  */
1639 static void css_clear_dir(struct cgroup_subsys_state *css)
1640 {
1641 	struct cgroup *cgrp = css->cgroup;
1642 	struct cftype *cfts;
1643 
1644 	if (!(css->flags & CSS_VISIBLE))
1645 		return;
1646 
1647 	css->flags &= ~CSS_VISIBLE;
1648 
1649 	if (!css->ss) {
1650 		if (cgroup_on_dfl(cgrp))
1651 			cfts = cgroup_base_files;
1652 		else
1653 			cfts = cgroup1_base_files;
1654 
1655 		cgroup_addrm_files(css, cgrp, cfts, false);
1656 	} else {
1657 		list_for_each_entry(cfts, &css->ss->cfts, node)
1658 			cgroup_addrm_files(css, cgrp, cfts, false);
1659 	}
1660 }
1661 
1662 /**
1663  * css_populate_dir - create subsys files in a cgroup directory
1664  * @css: target css
1665  *
1666  * On failure, no file is added.
1667  */
1668 static int css_populate_dir(struct cgroup_subsys_state *css)
1669 {
1670 	struct cgroup *cgrp = css->cgroup;
1671 	struct cftype *cfts, *failed_cfts;
1672 	int ret;
1673 
1674 	if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1675 		return 0;
1676 
1677 	if (!css->ss) {
1678 		if (cgroup_on_dfl(cgrp))
1679 			cfts = cgroup_base_files;
1680 		else
1681 			cfts = cgroup1_base_files;
1682 
1683 		ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1684 		if (ret < 0)
1685 			return ret;
1686 	} else {
1687 		list_for_each_entry(cfts, &css->ss->cfts, node) {
1688 			ret = cgroup_addrm_files(css, cgrp, cfts, true);
1689 			if (ret < 0) {
1690 				failed_cfts = cfts;
1691 				goto err;
1692 			}
1693 		}
1694 	}
1695 
1696 	css->flags |= CSS_VISIBLE;
1697 
1698 	return 0;
1699 err:
1700 	list_for_each_entry(cfts, &css->ss->cfts, node) {
1701 		if (cfts == failed_cfts)
1702 			break;
1703 		cgroup_addrm_files(css, cgrp, cfts, false);
1704 	}
1705 	return ret;
1706 }
1707 
1708 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1709 {
1710 	struct cgroup *dcgrp = &dst_root->cgrp;
1711 	struct cgroup_subsys *ss;
1712 	int ssid, i, ret;
1713 
1714 	lockdep_assert_held(&cgroup_mutex);
1715 
1716 	do_each_subsys_mask(ss, ssid, ss_mask) {
1717 		/*
1718 		 * If @ss has non-root csses attached to it, can't move.
1719 		 * If @ss is an implicit controller, it is exempt from this
1720 		 * rule and can be stolen.
1721 		 */
1722 		if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1723 		    !ss->implicit_on_dfl)
1724 			return -EBUSY;
1725 
1726 		/* can't move between two non-dummy roots either */
1727 		if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1728 			return -EBUSY;
1729 	} while_each_subsys_mask();
1730 
1731 	do_each_subsys_mask(ss, ssid, ss_mask) {
1732 		struct cgroup_root *src_root = ss->root;
1733 		struct cgroup *scgrp = &src_root->cgrp;
1734 		struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1735 		struct css_set *cset;
1736 
1737 		WARN_ON(!css || cgroup_css(dcgrp, ss));
1738 
1739 		/* disable from the source */
1740 		src_root->subsys_mask &= ~(1 << ssid);
1741 		WARN_ON(cgroup_apply_control(scgrp));
1742 		cgroup_finalize_control(scgrp, 0);
1743 
1744 		/* rebind */
1745 		RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1746 		rcu_assign_pointer(dcgrp->subsys[ssid], css);
1747 		ss->root = dst_root;
1748 		css->cgroup = dcgrp;
1749 
1750 		spin_lock_irq(&css_set_lock);
1751 		hash_for_each(css_set_table, i, cset, hlist)
1752 			list_move_tail(&cset->e_cset_node[ss->id],
1753 				       &dcgrp->e_csets[ss->id]);
1754 		spin_unlock_irq(&css_set_lock);
1755 
1756 		if (ss->css_rstat_flush) {
1757 			list_del_rcu(&css->rstat_css_node);
1758 			list_add_rcu(&css->rstat_css_node,
1759 				     &dcgrp->rstat_css_list);
1760 		}
1761 
1762 		/* default hierarchy doesn't enable controllers by default */
1763 		dst_root->subsys_mask |= 1 << ssid;
1764 		if (dst_root == &cgrp_dfl_root) {
1765 			static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1766 		} else {
1767 			dcgrp->subtree_control |= 1 << ssid;
1768 			static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1769 		}
1770 
1771 		ret = cgroup_apply_control(dcgrp);
1772 		if (ret)
1773 			pr_warn("partial failure to rebind %s controller (err=%d)\n",
1774 				ss->name, ret);
1775 
1776 		if (ss->bind)
1777 			ss->bind(css);
1778 	} while_each_subsys_mask();
1779 
1780 	kernfs_activate(dcgrp->kn);
1781 	return 0;
1782 }
1783 
1784 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1785 		     struct kernfs_root *kf_root)
1786 {
1787 	int len = 0;
1788 	char *buf = NULL;
1789 	struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1790 	struct cgroup *ns_cgroup;
1791 
1792 	buf = kmalloc(PATH_MAX, GFP_KERNEL);
1793 	if (!buf)
1794 		return -ENOMEM;
1795 
1796 	spin_lock_irq(&css_set_lock);
1797 	ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1798 	len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1799 	spin_unlock_irq(&css_set_lock);
1800 
1801 	if (len >= PATH_MAX)
1802 		len = -ERANGE;
1803 	else if (len > 0) {
1804 		seq_escape(sf, buf, " \t\n\\");
1805 		len = 0;
1806 	}
1807 	kfree(buf);
1808 	return len;
1809 }
1810 
1811 enum cgroup2_param {
1812 	Opt_nsdelegate,
1813 	Opt_memory_localevents,
1814 	Opt_memory_recursiveprot,
1815 	nr__cgroup2_params
1816 };
1817 
1818 static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1819 	fsparam_flag("nsdelegate",		Opt_nsdelegate),
1820 	fsparam_flag("memory_localevents",	Opt_memory_localevents),
1821 	fsparam_flag("memory_recursiveprot",	Opt_memory_recursiveprot),
1822 	{}
1823 };
1824 
1825 static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1826 {
1827 	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1828 	struct fs_parse_result result;
1829 	int opt;
1830 
1831 	opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
1832 	if (opt < 0)
1833 		return opt;
1834 
1835 	switch (opt) {
1836 	case Opt_nsdelegate:
1837 		ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1838 		return 0;
1839 	case Opt_memory_localevents:
1840 		ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1841 		return 0;
1842 	case Opt_memory_recursiveprot:
1843 		ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1844 		return 0;
1845 	}
1846 	return -EINVAL;
1847 }
1848 
1849 static void apply_cgroup_root_flags(unsigned int root_flags)
1850 {
1851 	if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1852 		if (root_flags & CGRP_ROOT_NS_DELEGATE)
1853 			cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1854 		else
1855 			cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1856 
1857 		if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1858 			cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1859 		else
1860 			cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1861 
1862 		if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1863 			cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1864 		else
1865 			cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1866 	}
1867 }
1868 
1869 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1870 {
1871 	if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1872 		seq_puts(seq, ",nsdelegate");
1873 	if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1874 		seq_puts(seq, ",memory_localevents");
1875 	if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1876 		seq_puts(seq, ",memory_recursiveprot");
1877 	return 0;
1878 }
1879 
1880 static int cgroup_reconfigure(struct fs_context *fc)
1881 {
1882 	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1883 
1884 	apply_cgroup_root_flags(ctx->flags);
1885 	return 0;
1886 }
1887 
1888 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1889 {
1890 	struct cgroup_subsys *ss;
1891 	int ssid;
1892 
1893 	INIT_LIST_HEAD(&cgrp->self.sibling);
1894 	INIT_LIST_HEAD(&cgrp->self.children);
1895 	INIT_LIST_HEAD(&cgrp->cset_links);
1896 	INIT_LIST_HEAD(&cgrp->pidlists);
1897 	mutex_init(&cgrp->pidlist_mutex);
1898 	cgrp->self.cgroup = cgrp;
1899 	cgrp->self.flags |= CSS_ONLINE;
1900 	cgrp->dom_cgrp = cgrp;
1901 	cgrp->max_descendants = INT_MAX;
1902 	cgrp->max_depth = INT_MAX;
1903 	INIT_LIST_HEAD(&cgrp->rstat_css_list);
1904 	prev_cputime_init(&cgrp->prev_cputime);
1905 
1906 	for_each_subsys(ss, ssid)
1907 		INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1908 
1909 	init_waitqueue_head(&cgrp->offline_waitq);
1910 	INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1911 }
1912 
1913 void init_cgroup_root(struct cgroup_fs_context *ctx)
1914 {
1915 	struct cgroup_root *root = ctx->root;
1916 	struct cgroup *cgrp = &root->cgrp;
1917 
1918 	INIT_LIST_HEAD(&root->root_list);
1919 	atomic_set(&root->nr_cgrps, 1);
1920 	cgrp->root = root;
1921 	init_cgroup_housekeeping(cgrp);
1922 
1923 	root->flags = ctx->flags;
1924 	if (ctx->release_agent)
1925 		strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1926 	if (ctx->name)
1927 		strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1928 	if (ctx->cpuset_clone_children)
1929 		set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1930 }
1931 
1932 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1933 {
1934 	LIST_HEAD(tmp_links);
1935 	struct cgroup *root_cgrp = &root->cgrp;
1936 	struct kernfs_syscall_ops *kf_sops;
1937 	struct css_set *cset;
1938 	int i, ret;
1939 
1940 	lockdep_assert_held(&cgroup_mutex);
1941 
1942 	ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1943 			      0, GFP_KERNEL);
1944 	if (ret)
1945 		goto out;
1946 
1947 	/*
1948 	 * We're accessing css_set_count without locking css_set_lock here,
1949 	 * but that's OK - it can only be increased by someone holding
1950 	 * cgroup_lock, and that's us.  Later rebinding may disable
1951 	 * controllers on the default hierarchy and thus create new csets,
1952 	 * which can't be more than the existing ones.  Allocate 2x.
1953 	 */
1954 	ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1955 	if (ret)
1956 		goto cancel_ref;
1957 
1958 	ret = cgroup_init_root_id(root);
1959 	if (ret)
1960 		goto cancel_ref;
1961 
1962 	kf_sops = root == &cgrp_dfl_root ?
1963 		&cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1964 
1965 	root->kf_root = kernfs_create_root(kf_sops,
1966 					   KERNFS_ROOT_CREATE_DEACTIVATED |
1967 					   KERNFS_ROOT_SUPPORT_EXPORTOP |
1968 					   KERNFS_ROOT_SUPPORT_USER_XATTR,
1969 					   root_cgrp);
1970 	if (IS_ERR(root->kf_root)) {
1971 		ret = PTR_ERR(root->kf_root);
1972 		goto exit_root_id;
1973 	}
1974 	root_cgrp->kn = root->kf_root->kn;
1975 	WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
1976 	root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
1977 
1978 	ret = css_populate_dir(&root_cgrp->self);
1979 	if (ret)
1980 		goto destroy_root;
1981 
1982 	ret = cgroup_rstat_init(root_cgrp);
1983 	if (ret)
1984 		goto destroy_root;
1985 
1986 	ret = rebind_subsystems(root, ss_mask);
1987 	if (ret)
1988 		goto exit_stats;
1989 
1990 	ret = cgroup_bpf_inherit(root_cgrp);
1991 	WARN_ON_ONCE(ret);
1992 
1993 	trace_cgroup_setup_root(root);
1994 
1995 	/*
1996 	 * There must be no failure case after here, since rebinding takes
1997 	 * care of subsystems' refcounts, which are explicitly dropped in
1998 	 * the failure exit path.
1999 	 */
2000 	list_add(&root->root_list, &cgroup_roots);
2001 	cgroup_root_count++;
2002 
2003 	/*
2004 	 * Link the root cgroup in this hierarchy into all the css_set
2005 	 * objects.
2006 	 */
2007 	spin_lock_irq(&css_set_lock);
2008 	hash_for_each(css_set_table, i, cset, hlist) {
2009 		link_css_set(&tmp_links, cset, root_cgrp);
2010 		if (css_set_populated(cset))
2011 			cgroup_update_populated(root_cgrp, true);
2012 	}
2013 	spin_unlock_irq(&css_set_lock);
2014 
2015 	BUG_ON(!list_empty(&root_cgrp->self.children));
2016 	BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2017 
2018 	ret = 0;
2019 	goto out;
2020 
2021 exit_stats:
2022 	cgroup_rstat_exit(root_cgrp);
2023 destroy_root:
2024 	kernfs_destroy_root(root->kf_root);
2025 	root->kf_root = NULL;
2026 exit_root_id:
2027 	cgroup_exit_root_id(root);
2028 cancel_ref:
2029 	percpu_ref_exit(&root_cgrp->self.refcnt);
2030 out:
2031 	free_cgrp_cset_links(&tmp_links);
2032 	return ret;
2033 }
2034 
2035 int cgroup_do_get_tree(struct fs_context *fc)
2036 {
2037 	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2038 	int ret;
2039 
2040 	ctx->kfc.root = ctx->root->kf_root;
2041 	if (fc->fs_type == &cgroup2_fs_type)
2042 		ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2043 	else
2044 		ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2045 	ret = kernfs_get_tree(fc);
2046 
2047 	/*
2048 	 * In non-init cgroup namespace, instead of root cgroup's dentry,
2049 	 * we return the dentry corresponding to the cgroupns->root_cgrp.
2050 	 */
2051 	if (!ret && ctx->ns != &init_cgroup_ns) {
2052 		struct dentry *nsdentry;
2053 		struct super_block *sb = fc->root->d_sb;
2054 		struct cgroup *cgrp;
2055 
2056 		mutex_lock(&cgroup_mutex);
2057 		spin_lock_irq(&css_set_lock);
2058 
2059 		cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2060 
2061 		spin_unlock_irq(&css_set_lock);
2062 		mutex_unlock(&cgroup_mutex);
2063 
2064 		nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2065 		dput(fc->root);
2066 		if (IS_ERR(nsdentry)) {
2067 			deactivate_locked_super(sb);
2068 			ret = PTR_ERR(nsdentry);
2069 			nsdentry = NULL;
2070 		}
2071 		fc->root = nsdentry;
2072 	}
2073 
2074 	if (!ctx->kfc.new_sb_created)
2075 		cgroup_put(&ctx->root->cgrp);
2076 
2077 	return ret;
2078 }
2079 
2080 /*
2081  * Destroy a cgroup filesystem context.
2082  */
2083 static void cgroup_fs_context_free(struct fs_context *fc)
2084 {
2085 	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2086 
2087 	kfree(ctx->name);
2088 	kfree(ctx->release_agent);
2089 	put_cgroup_ns(ctx->ns);
2090 	kernfs_free_fs_context(fc);
2091 	kfree(ctx);
2092 }
2093 
2094 static int cgroup_get_tree(struct fs_context *fc)
2095 {
2096 	struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2097 	int ret;
2098 
2099 	cgrp_dfl_visible = true;
2100 	cgroup_get_live(&cgrp_dfl_root.cgrp);
2101 	ctx->root = &cgrp_dfl_root;
2102 
2103 	ret = cgroup_do_get_tree(fc);
2104 	if (!ret)
2105 		apply_cgroup_root_flags(ctx->flags);
2106 	return ret;
2107 }
2108 
2109 static const struct fs_context_operations cgroup_fs_context_ops = {
2110 	.free		= cgroup_fs_context_free,
2111 	.parse_param	= cgroup2_parse_param,
2112 	.get_tree	= cgroup_get_tree,
2113 	.reconfigure	= cgroup_reconfigure,
2114 };
2115 
2116 static const struct fs_context_operations cgroup1_fs_context_ops = {
2117 	.free		= cgroup_fs_context_free,
2118 	.parse_param	= cgroup1_parse_param,
2119 	.get_tree	= cgroup1_get_tree,
2120 	.reconfigure	= cgroup1_reconfigure,
2121 };
2122 
2123 /*
2124  * Initialise the cgroup filesystem creation/reconfiguration context.  Notably,
2125  * we select the namespace we're going to use.
2126  */
2127 static int cgroup_init_fs_context(struct fs_context *fc)
2128 {
2129 	struct cgroup_fs_context *ctx;
2130 
2131 	ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2132 	if (!ctx)
2133 		return -ENOMEM;
2134 
2135 	ctx->ns = current->nsproxy->cgroup_ns;
2136 	get_cgroup_ns(ctx->ns);
2137 	fc->fs_private = &ctx->kfc;
2138 	if (fc->fs_type == &cgroup2_fs_type)
2139 		fc->ops = &cgroup_fs_context_ops;
2140 	else
2141 		fc->ops = &cgroup1_fs_context_ops;
2142 	put_user_ns(fc->user_ns);
2143 	fc->user_ns = get_user_ns(ctx->ns->user_ns);
2144 	fc->global = true;
2145 	return 0;
2146 }
2147 
2148 static void cgroup_kill_sb(struct super_block *sb)
2149 {
2150 	struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2151 	struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2152 
2153 	/*
2154 	 * If @root doesn't have any children, start killing it.
2155 	 * This prevents new mounts by disabling percpu_ref_tryget_live().
2156 	 * cgroup_mount() may wait for @root's release.
2157 	 *
2158 	 * And don't kill the default root.
2159 	 */
2160 	if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
2161 	    !percpu_ref_is_dying(&root->cgrp.self.refcnt))
2162 		percpu_ref_kill(&root->cgrp.self.refcnt);
2163 	cgroup_put(&root->cgrp);
2164 	kernfs_kill_sb(sb);
2165 }
2166 
2167 struct file_system_type cgroup_fs_type = {
2168 	.name			= "cgroup",
2169 	.init_fs_context	= cgroup_init_fs_context,
2170 	.parameters		= cgroup1_fs_parameters,
2171 	.kill_sb		= cgroup_kill_sb,
2172 	.fs_flags		= FS_USERNS_MOUNT,
2173 };
2174 
2175 static struct file_system_type cgroup2_fs_type = {
2176 	.name			= "cgroup2",
2177 	.init_fs_context	= cgroup_init_fs_context,
2178 	.parameters		= cgroup2_fs_parameters,
2179 	.kill_sb		= cgroup_kill_sb,
2180 	.fs_flags		= FS_USERNS_MOUNT,
2181 };
2182 
2183 #ifdef CONFIG_CPUSETS
2184 static const struct fs_context_operations cpuset_fs_context_ops = {
2185 	.get_tree	= cgroup1_get_tree,
2186 	.free		= cgroup_fs_context_free,
2187 };
2188 
2189 /*
2190  * This is ugly, but preserves the userspace API for existing cpuset
2191  * users. If someone tries to mount the "cpuset" filesystem, we
2192  * silently switch it to mount "cgroup" instead
2193  */
2194 static int cpuset_init_fs_context(struct fs_context *fc)
2195 {
2196 	char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2197 	struct cgroup_fs_context *ctx;
2198 	int err;
2199 
2200 	err = cgroup_init_fs_context(fc);
2201 	if (err) {
2202 		kfree(agent);
2203 		return err;
2204 	}
2205 
2206 	fc->ops = &cpuset_fs_context_ops;
2207 
2208 	ctx = cgroup_fc2context(fc);
2209 	ctx->subsys_mask = 1 << cpuset_cgrp_id;
2210 	ctx->flags |= CGRP_ROOT_NOPREFIX;
2211 	ctx->release_agent = agent;
2212 
2213 	get_filesystem(&cgroup_fs_type);
2214 	put_filesystem(fc->fs_type);
2215 	fc->fs_type = &cgroup_fs_type;
2216 
2217 	return 0;
2218 }
2219 
2220 static struct file_system_type cpuset_fs_type = {
2221 	.name			= "cpuset",
2222 	.init_fs_context	= cpuset_init_fs_context,
2223 	.fs_flags		= FS_USERNS_MOUNT,
2224 };
2225 #endif
2226 
2227 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2228 			  struct cgroup_namespace *ns)
2229 {
2230 	struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2231 
2232 	return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2233 }
2234 
2235 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2236 		   struct cgroup_namespace *ns)
2237 {
2238 	int ret;
2239 
2240 	mutex_lock(&cgroup_mutex);
2241 	spin_lock_irq(&css_set_lock);
2242 
2243 	ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2244 
2245 	spin_unlock_irq(&css_set_lock);
2246 	mutex_unlock(&cgroup_mutex);
2247 
2248 	return ret;
2249 }
2250 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2251 
2252 /**
2253  * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2254  * @task: target task
2255  * @buf: the buffer to write the path into
2256  * @buflen: the length of the buffer
2257  *
2258  * Determine @task's cgroup on the first (the one with the lowest non-zero
2259  * hierarchy_id) cgroup hierarchy and copy its path into @buf.  This
2260  * function grabs cgroup_mutex and shouldn't be used inside locks used by
2261  * cgroup controller callbacks.
2262  *
2263  * Return value is the same as kernfs_path().
2264  */
2265 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2266 {
2267 	struct cgroup_root *root;
2268 	struct cgroup *cgrp;
2269 	int hierarchy_id = 1;
2270 	int ret;
2271 
2272 	mutex_lock(&cgroup_mutex);
2273 	spin_lock_irq(&css_set_lock);
2274 
2275 	root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2276 
2277 	if (root) {
2278 		cgrp = task_cgroup_from_root(task, root);
2279 		ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2280 	} else {
2281 		/* if no hierarchy exists, everyone is in "/" */
2282 		ret = strlcpy(buf, "/", buflen);
2283 	}
2284 
2285 	spin_unlock_irq(&css_set_lock);
2286 	mutex_unlock(&cgroup_mutex);
2287 	return ret;
2288 }
2289 EXPORT_SYMBOL_GPL(task_cgroup_path);
2290 
2291 /**
2292  * cgroup_migrate_add_task - add a migration target task to a migration context
2293  * @task: target task
2294  * @mgctx: target migration context
2295  *
2296  * Add @task, which is a migration target, to @mgctx->tset.  This function
2297  * becomes noop if @task doesn't need to be migrated.  @task's css_set
2298  * should have been added as a migration source and @task->cg_list will be
2299  * moved from the css_set's tasks list to mg_tasks one.
2300  */
2301 static void cgroup_migrate_add_task(struct task_struct *task,
2302 				    struct cgroup_mgctx *mgctx)
2303 {
2304 	struct css_set *cset;
2305 
2306 	lockdep_assert_held(&css_set_lock);
2307 
2308 	/* @task either already exited or can't exit until the end */
2309 	if (task->flags & PF_EXITING)
2310 		return;
2311 
2312 	/* cgroup_threadgroup_rwsem protects racing against forks */
2313 	WARN_ON_ONCE(list_empty(&task->cg_list));
2314 
2315 	cset = task_css_set(task);
2316 	if (!cset->mg_src_cgrp)
2317 		return;
2318 
2319 	mgctx->tset.nr_tasks++;
2320 
2321 	list_move_tail(&task->cg_list, &cset->mg_tasks);
2322 	if (list_empty(&cset->mg_node))
2323 		list_add_tail(&cset->mg_node,
2324 			      &mgctx->tset.src_csets);
2325 	if (list_empty(&cset->mg_dst_cset->mg_node))
2326 		list_add_tail(&cset->mg_dst_cset->mg_node,
2327 			      &mgctx->tset.dst_csets);
2328 }
2329 
2330 /**
2331  * cgroup_taskset_first - reset taskset and return the first task
2332  * @tset: taskset of interest
2333  * @dst_cssp: output variable for the destination css
2334  *
2335  * @tset iteration is initialized and the first task is returned.
2336  */
2337 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2338 					 struct cgroup_subsys_state **dst_cssp)
2339 {
2340 	tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2341 	tset->cur_task = NULL;
2342 
2343 	return cgroup_taskset_next(tset, dst_cssp);
2344 }
2345 
2346 /**
2347  * cgroup_taskset_next - iterate to the next task in taskset
2348  * @tset: taskset of interest
2349  * @dst_cssp: output variable for the destination css
2350  *
2351  * Return the next task in @tset.  Iteration must have been initialized
2352  * with cgroup_taskset_first().
2353  */
2354 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2355 					struct cgroup_subsys_state **dst_cssp)
2356 {
2357 	struct css_set *cset = tset->cur_cset;
2358 	struct task_struct *task = tset->cur_task;
2359 
2360 	while (&cset->mg_node != tset->csets) {
2361 		if (!task)
2362 			task = list_first_entry(&cset->mg_tasks,
2363 						struct task_struct, cg_list);
2364 		else
2365 			task = list_next_entry(task, cg_list);
2366 
2367 		if (&task->cg_list != &cset->mg_tasks) {
2368 			tset->cur_cset = cset;
2369 			tset->cur_task = task;
2370 
2371 			/*
2372 			 * This function may be called both before and
2373 			 * after cgroup_taskset_migrate().  The two cases
2374 			 * can be distinguished by looking at whether @cset
2375 			 * has its ->mg_dst_cset set.
2376 			 */
2377 			if (cset->mg_dst_cset)
2378 				*dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2379 			else
2380 				*dst_cssp = cset->subsys[tset->ssid];
2381 
2382 			return task;
2383 		}
2384 
2385 		cset = list_next_entry(cset, mg_node);
2386 		task = NULL;
2387 	}
2388 
2389 	return NULL;
2390 }
2391 
2392 /**
2393  * cgroup_taskset_migrate - migrate a taskset
2394  * @mgctx: migration context
2395  *
2396  * Migrate tasks in @mgctx as setup by migration preparation functions.
2397  * This function fails iff one of the ->can_attach callbacks fails and
2398  * guarantees that either all or none of the tasks in @mgctx are migrated.
2399  * @mgctx is consumed regardless of success.
2400  */
2401 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2402 {
2403 	struct cgroup_taskset *tset = &mgctx->tset;
2404 	struct cgroup_subsys *ss;
2405 	struct task_struct *task, *tmp_task;
2406 	struct css_set *cset, *tmp_cset;
2407 	int ssid, failed_ssid, ret;
2408 
2409 	/* check that we can legitimately attach to the cgroup */
2410 	if (tset->nr_tasks) {
2411 		do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2412 			if (ss->can_attach) {
2413 				tset->ssid = ssid;
2414 				ret = ss->can_attach(tset);
2415 				if (ret) {
2416 					failed_ssid = ssid;
2417 					goto out_cancel_attach;
2418 				}
2419 			}
2420 		} while_each_subsys_mask();
2421 	}
2422 
2423 	/*
2424 	 * Now that we're guaranteed success, proceed to move all tasks to
2425 	 * the new cgroup.  There are no failure cases after here, so this
2426 	 * is the commit point.
2427 	 */
2428 	spin_lock_irq(&css_set_lock);
2429 	list_for_each_entry(cset, &tset->src_csets, mg_node) {
2430 		list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2431 			struct css_set *from_cset = task_css_set(task);
2432 			struct css_set *to_cset = cset->mg_dst_cset;
2433 
2434 			get_css_set(to_cset);
2435 			to_cset->nr_tasks++;
2436 			css_set_move_task(task, from_cset, to_cset, true);
2437 			from_cset->nr_tasks--;
2438 			/*
2439 			 * If the source or destination cgroup is frozen,
2440 			 * the task might require to change its state.
2441 			 */
2442 			cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2443 						    to_cset->dfl_cgrp);
2444 			put_css_set_locked(from_cset);
2445 
2446 		}
2447 	}
2448 	spin_unlock_irq(&css_set_lock);
2449 
2450 	/*
2451 	 * Migration is committed, all target tasks are now on dst_csets.
2452 	 * Nothing is sensitive to fork() after this point.  Notify
2453 	 * controllers that migration is complete.
2454 	 */
2455 	tset->csets = &tset->dst_csets;
2456 
2457 	if (tset->nr_tasks) {
2458 		do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2459 			if (ss->attach) {
2460 				tset->ssid = ssid;
2461 				ss->attach(tset);
2462 			}
2463 		} while_each_subsys_mask();
2464 	}
2465 
2466 	ret = 0;
2467 	goto out_release_tset;
2468 
2469 out_cancel_attach:
2470 	if (tset->nr_tasks) {
2471 		do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2472 			if (ssid == failed_ssid)
2473 				break;
2474 			if (ss->cancel_attach) {
2475 				tset->ssid = ssid;
2476 				ss->cancel_attach(tset);
2477 			}
2478 		} while_each_subsys_mask();
2479 	}
2480 out_release_tset:
2481 	spin_lock_irq(&css_set_lock);
2482 	list_splice_init(&tset->dst_csets, &tset->src_csets);
2483 	list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2484 		list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2485 		list_del_init(&cset->mg_node);
2486 	}
2487 	spin_unlock_irq(&css_set_lock);
2488 
2489 	/*
2490 	 * Re-initialize the cgroup_taskset structure in case it is reused
2491 	 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2492 	 * iteration.
2493 	 */
2494 	tset->nr_tasks = 0;
2495 	tset->csets    = &tset->src_csets;
2496 	return ret;
2497 }
2498 
2499 /**
2500  * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2501  * @dst_cgrp: destination cgroup to test
2502  *
2503  * On the default hierarchy, except for the mixable, (possible) thread root
2504  * and threaded cgroups, subtree_control must be zero for migration
2505  * destination cgroups with tasks so that child cgroups don't compete
2506  * against tasks.
2507  */
2508 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2509 {
2510 	/* v1 doesn't have any restriction */
2511 	if (!cgroup_on_dfl(dst_cgrp))
2512 		return 0;
2513 
2514 	/* verify @dst_cgrp can host resources */
2515 	if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2516 		return -EOPNOTSUPP;
2517 
2518 	/* mixables don't care */
2519 	if (cgroup_is_mixable(dst_cgrp))
2520 		return 0;
2521 
2522 	/*
2523 	 * If @dst_cgrp is already or can become a thread root or is
2524 	 * threaded, it doesn't matter.
2525 	 */
2526 	if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2527 		return 0;
2528 
2529 	/* apply no-internal-process constraint */
2530 	if (dst_cgrp->subtree_control)
2531 		return -EBUSY;
2532 
2533 	return 0;
2534 }
2535 
2536 /**
2537  * cgroup_migrate_finish - cleanup after attach
2538  * @mgctx: migration context
2539  *
2540  * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst().  See
2541  * those functions for details.
2542  */
2543 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2544 {
2545 	LIST_HEAD(preloaded);
2546 	struct css_set *cset, *tmp_cset;
2547 
2548 	lockdep_assert_held(&cgroup_mutex);
2549 
2550 	spin_lock_irq(&css_set_lock);
2551 
2552 	list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2553 	list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2554 
2555 	list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2556 		cset->mg_src_cgrp = NULL;
2557 		cset->mg_dst_cgrp = NULL;
2558 		cset->mg_dst_cset = NULL;
2559 		list_del_init(&cset->mg_preload_node);
2560 		put_css_set_locked(cset);
2561 	}
2562 
2563 	spin_unlock_irq(&css_set_lock);
2564 }
2565 
2566 /**
2567  * cgroup_migrate_add_src - add a migration source css_set
2568  * @src_cset: the source css_set to add
2569  * @dst_cgrp: the destination cgroup
2570  * @mgctx: migration context
2571  *
2572  * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp.  Pin
2573  * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2574  * up by cgroup_migrate_finish().
2575  *
2576  * This function may be called without holding cgroup_threadgroup_rwsem
2577  * even if the target is a process.  Threads may be created and destroyed
2578  * but as long as cgroup_mutex is not dropped, no new css_set can be put
2579  * into play and the preloaded css_sets are guaranteed to cover all
2580  * migrations.
2581  */
2582 void cgroup_migrate_add_src(struct css_set *src_cset,
2583 			    struct cgroup *dst_cgrp,
2584 			    struct cgroup_mgctx *mgctx)
2585 {
2586 	struct cgroup *src_cgrp;
2587 
2588 	lockdep_assert_held(&cgroup_mutex);
2589 	lockdep_assert_held(&css_set_lock);
2590 
2591 	/*
2592 	 * If ->dead, @src_set is associated with one or more dead cgroups
2593 	 * and doesn't contain any migratable tasks.  Ignore it early so
2594 	 * that the rest of migration path doesn't get confused by it.
2595 	 */
2596 	if (src_cset->dead)
2597 		return;
2598 
2599 	src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2600 
2601 	if (!list_empty(&src_cset->mg_preload_node))
2602 		return;
2603 
2604 	WARN_ON(src_cset->mg_src_cgrp);
2605 	WARN_ON(src_cset->mg_dst_cgrp);
2606 	WARN_ON(!list_empty(&src_cset->mg_tasks));
2607 	WARN_ON(!list_empty(&src_cset->mg_node));
2608 
2609 	src_cset->mg_src_cgrp = src_cgrp;
2610 	src_cset->mg_dst_cgrp = dst_cgrp;
2611 	get_css_set(src_cset);
2612 	list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2613 }
2614 
2615 /**
2616  * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2617  * @mgctx: migration context
2618  *
2619  * Tasks are about to be moved and all the source css_sets have been
2620  * preloaded to @mgctx->preloaded_src_csets.  This function looks up and
2621  * pins all destination css_sets, links each to its source, and append them
2622  * to @mgctx->preloaded_dst_csets.
2623  *
2624  * This function must be called after cgroup_migrate_add_src() has been
2625  * called on each migration source css_set.  After migration is performed
2626  * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2627  * @mgctx.
2628  */
2629 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2630 {
2631 	struct css_set *src_cset, *tmp_cset;
2632 
2633 	lockdep_assert_held(&cgroup_mutex);
2634 
2635 	/* look up the dst cset for each src cset and link it to src */
2636 	list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2637 				 mg_preload_node) {
2638 		struct css_set *dst_cset;
2639 		struct cgroup_subsys *ss;
2640 		int ssid;
2641 
2642 		dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2643 		if (!dst_cset)
2644 			return -ENOMEM;
2645 
2646 		WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2647 
2648 		/*
2649 		 * If src cset equals dst, it's noop.  Drop the src.
2650 		 * cgroup_migrate() will skip the cset too.  Note that we
2651 		 * can't handle src == dst as some nodes are used by both.
2652 		 */
2653 		if (src_cset == dst_cset) {
2654 			src_cset->mg_src_cgrp = NULL;
2655 			src_cset->mg_dst_cgrp = NULL;
2656 			list_del_init(&src_cset->mg_preload_node);
2657 			put_css_set(src_cset);
2658 			put_css_set(dst_cset);
2659 			continue;
2660 		}
2661 
2662 		src_cset->mg_dst_cset = dst_cset;
2663 
2664 		if (list_empty(&dst_cset->mg_preload_node))
2665 			list_add_tail(&dst_cset->mg_preload_node,
2666 				      &mgctx->preloaded_dst_csets);
2667 		else
2668 			put_css_set(dst_cset);
2669 
2670 		for_each_subsys(ss, ssid)
2671 			if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2672 				mgctx->ss_mask |= 1 << ssid;
2673 	}
2674 
2675 	return 0;
2676 }
2677 
2678 /**
2679  * cgroup_migrate - migrate a process or task to a cgroup
2680  * @leader: the leader of the process or the task to migrate
2681  * @threadgroup: whether @leader points to the whole process or a single task
2682  * @mgctx: migration context
2683  *
2684  * Migrate a process or task denoted by @leader.  If migrating a process,
2685  * the caller must be holding cgroup_threadgroup_rwsem.  The caller is also
2686  * responsible for invoking cgroup_migrate_add_src() and
2687  * cgroup_migrate_prepare_dst() on the targets before invoking this
2688  * function and following up with cgroup_migrate_finish().
2689  *
2690  * As long as a controller's ->can_attach() doesn't fail, this function is
2691  * guaranteed to succeed.  This means that, excluding ->can_attach()
2692  * failure, when migrating multiple targets, the success or failure can be
2693  * decided for all targets by invoking group_migrate_prepare_dst() before
2694  * actually starting migrating.
2695  */
2696 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2697 		   struct cgroup_mgctx *mgctx)
2698 {
2699 	struct task_struct *task;
2700 
2701 	/*
2702 	 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2703 	 * already PF_EXITING could be freed from underneath us unless we
2704 	 * take an rcu_read_lock.
2705 	 */
2706 	spin_lock_irq(&css_set_lock);
2707 	rcu_read_lock();
2708 	task = leader;
2709 	do {
2710 		cgroup_migrate_add_task(task, mgctx);
2711 		if (!threadgroup)
2712 			break;
2713 	} while_each_thread(leader, task);
2714 	rcu_read_unlock();
2715 	spin_unlock_irq(&css_set_lock);
2716 
2717 	return cgroup_migrate_execute(mgctx);
2718 }
2719 
2720 /**
2721  * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2722  * @dst_cgrp: the cgroup to attach to
2723  * @leader: the task or the leader of the threadgroup to be attached
2724  * @threadgroup: attach the whole threadgroup?
2725  *
2726  * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2727  */
2728 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2729 		       bool threadgroup)
2730 {
2731 	DEFINE_CGROUP_MGCTX(mgctx);
2732 	struct task_struct *task;
2733 	int ret = 0;
2734 
2735 	/* look up all src csets */
2736 	spin_lock_irq(&css_set_lock);
2737 	rcu_read_lock();
2738 	task = leader;
2739 	do {
2740 		cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2741 		if (!threadgroup)
2742 			break;
2743 	} while_each_thread(leader, task);
2744 	rcu_read_unlock();
2745 	spin_unlock_irq(&css_set_lock);
2746 
2747 	/* prepare dst csets and commit */
2748 	ret = cgroup_migrate_prepare_dst(&mgctx);
2749 	if (!ret)
2750 		ret = cgroup_migrate(leader, threadgroup, &mgctx);
2751 
2752 	cgroup_migrate_finish(&mgctx);
2753 
2754 	if (!ret)
2755 		TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2756 
2757 	return ret;
2758 }
2759 
2760 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
2761 					     bool *locked)
2762 	__acquires(&cgroup_threadgroup_rwsem)
2763 {
2764 	struct task_struct *tsk;
2765 	pid_t pid;
2766 
2767 	if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2768 		return ERR_PTR(-EINVAL);
2769 
2770 	/*
2771 	 * If we migrate a single thread, we don't care about threadgroup
2772 	 * stability. If the thread is `current`, it won't exit(2) under our
2773 	 * hands or change PID through exec(2). We exclude
2774 	 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2775 	 * callers by cgroup_mutex.
2776 	 * Therefore, we can skip the global lock.
2777 	 */
2778 	lockdep_assert_held(&cgroup_mutex);
2779 	if (pid || threadgroup) {
2780 		percpu_down_write(&cgroup_threadgroup_rwsem);
2781 		*locked = true;
2782 	} else {
2783 		*locked = false;
2784 	}
2785 
2786 	rcu_read_lock();
2787 	if (pid) {
2788 		tsk = find_task_by_vpid(pid);
2789 		if (!tsk) {
2790 			tsk = ERR_PTR(-ESRCH);
2791 			goto out_unlock_threadgroup;
2792 		}
2793 	} else {
2794 		tsk = current;
2795 	}
2796 
2797 	if (threadgroup)
2798 		tsk = tsk->group_leader;
2799 
2800 	/*
2801 	 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2802 	 * If userland migrates such a kthread to a non-root cgroup, it can
2803 	 * become trapped in a cpuset, or RT kthread may be born in a
2804 	 * cgroup with no rt_runtime allocated.  Just say no.
2805 	 */
2806 	if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2807 		tsk = ERR_PTR(-EINVAL);
2808 		goto out_unlock_threadgroup;
2809 	}
2810 
2811 	get_task_struct(tsk);
2812 	goto out_unlock_rcu;
2813 
2814 out_unlock_threadgroup:
2815 	if (*locked) {
2816 		percpu_up_write(&cgroup_threadgroup_rwsem);
2817 		*locked = false;
2818 	}
2819 out_unlock_rcu:
2820 	rcu_read_unlock();
2821 	return tsk;
2822 }
2823 
2824 void cgroup_procs_write_finish(struct task_struct *task, bool locked)
2825 	__releases(&cgroup_threadgroup_rwsem)
2826 {
2827 	struct cgroup_subsys *ss;
2828 	int ssid;
2829 
2830 	/* release reference from cgroup_procs_write_start() */
2831 	put_task_struct(task);
2832 
2833 	if (locked)
2834 		percpu_up_write(&cgroup_threadgroup_rwsem);
2835 	for_each_subsys(ss, ssid)
2836 		if (ss->post_attach)
2837 			ss->post_attach();
2838 }
2839 
2840 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2841 {
2842 	struct cgroup_subsys *ss;
2843 	bool printed = false;
2844 	int ssid;
2845 
2846 	do_each_subsys_mask(ss, ssid, ss_mask) {
2847 		if (printed)
2848 			seq_putc(seq, ' ');
2849 		seq_puts(seq, ss->name);
2850 		printed = true;
2851 	} while_each_subsys_mask();
2852 	if (printed)
2853 		seq_putc(seq, '\n');
2854 }
2855 
2856 /* show controllers which are enabled from the parent */
2857 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2858 {
2859 	struct cgroup *cgrp = seq_css(seq)->cgroup;
2860 
2861 	cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2862 	return 0;
2863 }
2864 
2865 /* show controllers which are enabled for a given cgroup's children */
2866 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2867 {
2868 	struct cgroup *cgrp = seq_css(seq)->cgroup;
2869 
2870 	cgroup_print_ss_mask(seq, cgrp->subtree_control);
2871 	return 0;
2872 }
2873 
2874 /**
2875  * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2876  * @cgrp: root of the subtree to update csses for
2877  *
2878  * @cgrp's control masks have changed and its subtree's css associations
2879  * need to be updated accordingly.  This function looks up all css_sets
2880  * which are attached to the subtree, creates the matching updated css_sets
2881  * and migrates the tasks to the new ones.
2882  */
2883 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2884 {
2885 	DEFINE_CGROUP_MGCTX(mgctx);
2886 	struct cgroup_subsys_state *d_css;
2887 	struct cgroup *dsct;
2888 	struct css_set *src_cset;
2889 	int ret;
2890 
2891 	lockdep_assert_held(&cgroup_mutex);
2892 
2893 	percpu_down_write(&cgroup_threadgroup_rwsem);
2894 
2895 	/* look up all csses currently attached to @cgrp's subtree */
2896 	spin_lock_irq(&css_set_lock);
2897 	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2898 		struct cgrp_cset_link *link;
2899 
2900 		list_for_each_entry(link, &dsct->cset_links, cset_link)
2901 			cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2902 	}
2903 	spin_unlock_irq(&css_set_lock);
2904 
2905 	/* NULL dst indicates self on default hierarchy */
2906 	ret = cgroup_migrate_prepare_dst(&mgctx);
2907 	if (ret)
2908 		goto out_finish;
2909 
2910 	spin_lock_irq(&css_set_lock);
2911 	list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2912 		struct task_struct *task, *ntask;
2913 
2914 		/* all tasks in src_csets need to be migrated */
2915 		list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2916 			cgroup_migrate_add_task(task, &mgctx);
2917 	}
2918 	spin_unlock_irq(&css_set_lock);
2919 
2920 	ret = cgroup_migrate_execute(&mgctx);
2921 out_finish:
2922 	cgroup_migrate_finish(&mgctx);
2923 	percpu_up_write(&cgroup_threadgroup_rwsem);
2924 	return ret;
2925 }
2926 
2927 /**
2928  * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2929  * @cgrp: root of the target subtree
2930  *
2931  * Because css offlining is asynchronous, userland may try to re-enable a
2932  * controller while the previous css is still around.  This function grabs
2933  * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2934  */
2935 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2936 	__acquires(&cgroup_mutex)
2937 {
2938 	struct cgroup *dsct;
2939 	struct cgroup_subsys_state *d_css;
2940 	struct cgroup_subsys *ss;
2941 	int ssid;
2942 
2943 restart:
2944 	mutex_lock(&cgroup_mutex);
2945 
2946 	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2947 		for_each_subsys(ss, ssid) {
2948 			struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2949 			DEFINE_WAIT(wait);
2950 
2951 			if (!css || !percpu_ref_is_dying(&css->refcnt))
2952 				continue;
2953 
2954 			cgroup_get_live(dsct);
2955 			prepare_to_wait(&dsct->offline_waitq, &wait,
2956 					TASK_UNINTERRUPTIBLE);
2957 
2958 			mutex_unlock(&cgroup_mutex);
2959 			schedule();
2960 			finish_wait(&dsct->offline_waitq, &wait);
2961 
2962 			cgroup_put(dsct);
2963 			goto restart;
2964 		}
2965 	}
2966 }
2967 
2968 /**
2969  * cgroup_save_control - save control masks and dom_cgrp of a subtree
2970  * @cgrp: root of the target subtree
2971  *
2972  * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
2973  * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2974  * itself.
2975  */
2976 static void cgroup_save_control(struct cgroup *cgrp)
2977 {
2978 	struct cgroup *dsct;
2979 	struct cgroup_subsys_state *d_css;
2980 
2981 	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2982 		dsct->old_subtree_control = dsct->subtree_control;
2983 		dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2984 		dsct->old_dom_cgrp = dsct->dom_cgrp;
2985 	}
2986 }
2987 
2988 /**
2989  * cgroup_propagate_control - refresh control masks of a subtree
2990  * @cgrp: root of the target subtree
2991  *
2992  * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2993  * ->subtree_control and propagate controller availability through the
2994  * subtree so that descendants don't have unavailable controllers enabled.
2995  */
2996 static void cgroup_propagate_control(struct cgroup *cgrp)
2997 {
2998 	struct cgroup *dsct;
2999 	struct cgroup_subsys_state *d_css;
3000 
3001 	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3002 		dsct->subtree_control &= cgroup_control(dsct);
3003 		dsct->subtree_ss_mask =
3004 			cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3005 						    cgroup_ss_mask(dsct));
3006 	}
3007 }
3008 
3009 /**
3010  * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3011  * @cgrp: root of the target subtree
3012  *
3013  * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3014  * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3015  * itself.
3016  */
3017 static void cgroup_restore_control(struct cgroup *cgrp)
3018 {
3019 	struct cgroup *dsct;
3020 	struct cgroup_subsys_state *d_css;
3021 
3022 	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3023 		dsct->subtree_control = dsct->old_subtree_control;
3024 		dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3025 		dsct->dom_cgrp = dsct->old_dom_cgrp;
3026 	}
3027 }
3028 
3029 static bool css_visible(struct cgroup_subsys_state *css)
3030 {
3031 	struct cgroup_subsys *ss = css->ss;
3032 	struct cgroup *cgrp = css->cgroup;
3033 
3034 	if (cgroup_control(cgrp) & (1 << ss->id))
3035 		return true;
3036 	if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3037 		return false;
3038 	return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3039 }
3040 
3041 /**
3042  * cgroup_apply_control_enable - enable or show csses according to control
3043  * @cgrp: root of the target subtree
3044  *
3045  * Walk @cgrp's subtree and create new csses or make the existing ones
3046  * visible.  A css is created invisible if it's being implicitly enabled
3047  * through dependency.  An invisible css is made visible when the userland
3048  * explicitly enables it.
3049  *
3050  * Returns 0 on success, -errno on failure.  On failure, csses which have
3051  * been processed already aren't cleaned up.  The caller is responsible for
3052  * cleaning up with cgroup_apply_control_disable().
3053  */
3054 static int cgroup_apply_control_enable(struct cgroup *cgrp)
3055 {
3056 	struct cgroup *dsct;
3057 	struct cgroup_subsys_state *d_css;
3058 	struct cgroup_subsys *ss;
3059 	int ssid, ret;
3060 
3061 	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3062 		for_each_subsys(ss, ssid) {
3063 			struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3064 
3065 			if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3066 				continue;
3067 
3068 			if (!css) {
3069 				css = css_create(dsct, ss);
3070 				if (IS_ERR(css))
3071 					return PTR_ERR(css);
3072 			}
3073 
3074 			WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3075 
3076 			if (css_visible(css)) {
3077 				ret = css_populate_dir(css);
3078 				if (ret)
3079 					return ret;
3080 			}
3081 		}
3082 	}
3083 
3084 	return 0;
3085 }
3086 
3087 /**
3088  * cgroup_apply_control_disable - kill or hide csses according to control
3089  * @cgrp: root of the target subtree
3090  *
3091  * Walk @cgrp's subtree and kill and hide csses so that they match
3092  * cgroup_ss_mask() and cgroup_visible_mask().
3093  *
3094  * A css is hidden when the userland requests it to be disabled while other
3095  * subsystems are still depending on it.  The css must not actively control
3096  * resources and be in the vanilla state if it's made visible again later.
3097  * Controllers which may be depended upon should provide ->css_reset() for
3098  * this purpose.
3099  */
3100 static void cgroup_apply_control_disable(struct cgroup *cgrp)
3101 {
3102 	struct cgroup *dsct;
3103 	struct cgroup_subsys_state *d_css;
3104 	struct cgroup_subsys *ss;
3105 	int ssid;
3106 
3107 	cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3108 		for_each_subsys(ss, ssid) {
3109 			struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3110 
3111 			if (!css)
3112 				continue;
3113 
3114 			WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3115 
3116 			if (css->parent &&
3117 			    !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3118 				kill_css(css);
3119 			} else if (!css_visible(css)) {
3120 				css_clear_dir(css);
3121 				if (ss->css_reset)
3122 					ss->css_reset(css);
3123 			}
3124 		}
3125 	}
3126 }
3127 
3128 /**
3129  * cgroup_apply_control - apply control mask updates to the subtree
3130  * @cgrp: root of the target subtree
3131  *
3132  * subsystems can be enabled and disabled in a subtree using the following
3133  * steps.
3134  *
3135  * 1. Call cgroup_save_control() to stash the current state.
3136  * 2. Update ->subtree_control masks in the subtree as desired.
3137  * 3. Call cgroup_apply_control() to apply the changes.
3138  * 4. Optionally perform other related operations.
3139  * 5. Call cgroup_finalize_control() to finish up.
3140  *
3141  * This function implements step 3 and propagates the mask changes
3142  * throughout @cgrp's subtree, updates csses accordingly and perform
3143  * process migrations.
3144  */
3145 static int cgroup_apply_control(struct cgroup *cgrp)
3146 {
3147 	int ret;
3148 
3149 	cgroup_propagate_control(cgrp);
3150 
3151 	ret = cgroup_apply_control_enable(cgrp);
3152 	if (ret)
3153 		return ret;
3154 
3155 	/*
3156 	 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3157 	 * making the following cgroup_update_dfl_csses() properly update
3158 	 * css associations of all tasks in the subtree.
3159 	 */
3160 	ret = cgroup_update_dfl_csses(cgrp);
3161 	if (ret)
3162 		return ret;
3163 
3164 	return 0;
3165 }
3166 
3167 /**
3168  * cgroup_finalize_control - finalize control mask update
3169  * @cgrp: root of the target subtree
3170  * @ret: the result of the update
3171  *
3172  * Finalize control mask update.  See cgroup_apply_control() for more info.
3173  */
3174 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3175 {
3176 	if (ret) {
3177 		cgroup_restore_control(cgrp);
3178 		cgroup_propagate_control(cgrp);
3179 	}
3180 
3181 	cgroup_apply_control_disable(cgrp);
3182 }
3183 
3184 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3185 {
3186 	u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3187 
3188 	/* if nothing is getting enabled, nothing to worry about */
3189 	if (!enable)
3190 		return 0;
3191 
3192 	/* can @cgrp host any resources? */
3193 	if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3194 		return -EOPNOTSUPP;
3195 
3196 	/* mixables don't care */
3197 	if (cgroup_is_mixable(cgrp))
3198 		return 0;
3199 
3200 	if (domain_enable) {
3201 		/* can't enable domain controllers inside a thread subtree */
3202 		if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3203 			return -EOPNOTSUPP;
3204 	} else {
3205 		/*
3206 		 * Threaded controllers can handle internal competitions
3207 		 * and are always allowed inside a (prospective) thread
3208 		 * subtree.
3209 		 */
3210 		if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3211 			return 0;
3212 	}
3213 
3214 	/*
3215 	 * Controllers can't be enabled for a cgroup with tasks to avoid
3216 	 * child cgroups competing against tasks.
3217 	 */
3218 	if (cgroup_has_tasks(cgrp))
3219 		return -EBUSY;
3220 
3221 	return 0;
3222 }
3223 
3224 /* change the enabled child controllers for a cgroup in the default hierarchy */
3225 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3226 					    char *buf, size_t nbytes,
3227 					    loff_t off)
3228 {
3229 	u16 enable = 0, disable = 0;
3230 	struct cgroup *cgrp, *child;
3231 	struct cgroup_subsys *ss;
3232 	char *tok;
3233 	int ssid, ret;
3234 
3235 	/*
3236 	 * Parse input - space separated list of subsystem names prefixed
3237 	 * with either + or -.
3238 	 */
3239 	buf = strstrip(buf);
3240 	while ((tok = strsep(&buf, " "))) {
3241 		if (tok[0] == '\0')
3242 			continue;
3243 		do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3244 			if (!cgroup_ssid_enabled(ssid) ||
3245 			    strcmp(tok + 1, ss->name))
3246 				continue;
3247 
3248 			if (*tok == '+') {
3249 				enable |= 1 << ssid;
3250 				disable &= ~(1 << ssid);
3251 			} else if (*tok == '-') {
3252 				disable |= 1 << ssid;
3253 				enable &= ~(1 << ssid);
3254 			} else {
3255 				return -EINVAL;
3256 			}
3257 			break;
3258 		} while_each_subsys_mask();
3259 		if (ssid == CGROUP_SUBSYS_COUNT)
3260 			return -EINVAL;
3261 	}
3262 
3263 	cgrp = cgroup_kn_lock_live(of->kn, true);
3264 	if (!cgrp)
3265 		return -ENODEV;
3266 
3267 	for_each_subsys(ss, ssid) {
3268 		if (enable & (1 << ssid)) {
3269 			if (cgrp->subtree_control & (1 << ssid)) {
3270 				enable &= ~(1 << ssid);
3271 				continue;
3272 			}
3273 
3274 			if (!(cgroup_control(cgrp) & (1 << ssid))) {
3275 				ret = -ENOENT;
3276 				goto out_unlock;
3277 			}
3278 		} else if (disable & (1 << ssid)) {
3279 			if (!(cgrp->subtree_control & (1 << ssid))) {
3280 				disable &= ~(1 << ssid);
3281 				continue;
3282 			}
3283 
3284 			/* a child has it enabled? */
3285 			cgroup_for_each_live_child(child, cgrp) {
3286 				if (child->subtree_control & (1 << ssid)) {
3287 					ret = -EBUSY;
3288 					goto out_unlock;
3289 				}
3290 			}
3291 		}
3292 	}
3293 
3294 	if (!enable && !disable) {
3295 		ret = 0;
3296 		goto out_unlock;
3297 	}
3298 
3299 	ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3300 	if (ret)
3301 		goto out_unlock;
3302 
3303 	/* save and update control masks and prepare csses */
3304 	cgroup_save_control(cgrp);
3305 
3306 	cgrp->subtree_control |= enable;
3307 	cgrp->subtree_control &= ~disable;
3308 
3309 	ret = cgroup_apply_control(cgrp);
3310 	cgroup_finalize_control(cgrp, ret);
3311 	if (ret)
3312 		goto out_unlock;
3313 
3314 	kernfs_activate(cgrp->kn);
3315 out_unlock:
3316 	cgroup_kn_unlock(of->kn);
3317 	return ret ?: nbytes;
3318 }
3319 
3320 /**
3321  * cgroup_enable_threaded - make @cgrp threaded
3322  * @cgrp: the target cgroup
3323  *
3324  * Called when "threaded" is written to the cgroup.type interface file and
3325  * tries to make @cgrp threaded and join the parent's resource domain.
3326  * This function is never called on the root cgroup as cgroup.type doesn't
3327  * exist on it.
3328  */
3329 static int cgroup_enable_threaded(struct cgroup *cgrp)
3330 {
3331 	struct cgroup *parent = cgroup_parent(cgrp);
3332 	struct cgroup *dom_cgrp = parent->dom_cgrp;
3333 	struct cgroup *dsct;
3334 	struct cgroup_subsys_state *d_css;
3335 	int ret;
3336 
3337 	lockdep_assert_held(&cgroup_mutex);
3338 
3339 	/* noop if already threaded */
3340 	if (cgroup_is_threaded(cgrp))
3341 		return 0;
3342 
3343 	/*
3344 	 * If @cgroup is populated or has domain controllers enabled, it
3345 	 * can't be switched.  While the below cgroup_can_be_thread_root()
3346 	 * test can catch the same conditions, that's only when @parent is
3347 	 * not mixable, so let's check it explicitly.
3348 	 */
3349 	if (cgroup_is_populated(cgrp) ||
3350 	    cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3351 		return -EOPNOTSUPP;
3352 
3353 	/* we're joining the parent's domain, ensure its validity */
3354 	if (!cgroup_is_valid_domain(dom_cgrp) ||
3355 	    !cgroup_can_be_thread_root(dom_cgrp))
3356 		return -EOPNOTSUPP;
3357 
3358 	/*
3359 	 * The following shouldn't cause actual migrations and should
3360 	 * always succeed.
3361 	 */
3362 	cgroup_save_control(cgrp);
3363 
3364 	cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3365 		if (dsct == cgrp || cgroup_is_threaded(dsct))
3366 			dsct->dom_cgrp = dom_cgrp;
3367 
3368 	ret = cgroup_apply_control(cgrp);
3369 	if (!ret)
3370 		parent->nr_threaded_children++;
3371 
3372 	cgroup_finalize_control(cgrp, ret);
3373 	return ret;
3374 }
3375 
3376 static int cgroup_type_show(struct seq_file *seq, void *v)
3377 {
3378 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3379 
3380 	if (cgroup_is_threaded(cgrp))
3381 		seq_puts(seq, "threaded\n");
3382 	else if (!cgroup_is_valid_domain(cgrp))
3383 		seq_puts(seq, "domain invalid\n");
3384 	else if (cgroup_is_thread_root(cgrp))
3385 		seq_puts(seq, "domain threaded\n");
3386 	else
3387 		seq_puts(seq, "domain\n");
3388 
3389 	return 0;
3390 }
3391 
3392 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3393 				 size_t nbytes, loff_t off)
3394 {
3395 	struct cgroup *cgrp;
3396 	int ret;
3397 
3398 	/* only switching to threaded mode is supported */
3399 	if (strcmp(strstrip(buf), "threaded"))
3400 		return -EINVAL;
3401 
3402 	/* drain dying csses before we re-apply (threaded) subtree control */
3403 	cgrp = cgroup_kn_lock_live(of->kn, true);
3404 	if (!cgrp)
3405 		return -ENOENT;
3406 
3407 	/* threaded can only be enabled */
3408 	ret = cgroup_enable_threaded(cgrp);
3409 
3410 	cgroup_kn_unlock(of->kn);
3411 	return ret ?: nbytes;
3412 }
3413 
3414 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3415 {
3416 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3417 	int descendants = READ_ONCE(cgrp->max_descendants);
3418 
3419 	if (descendants == INT_MAX)
3420 		seq_puts(seq, "max\n");
3421 	else
3422 		seq_printf(seq, "%d\n", descendants);
3423 
3424 	return 0;
3425 }
3426 
3427 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3428 					   char *buf, size_t nbytes, loff_t off)
3429 {
3430 	struct cgroup *cgrp;
3431 	int descendants;
3432 	ssize_t ret;
3433 
3434 	buf = strstrip(buf);
3435 	if (!strcmp(buf, "max")) {
3436 		descendants = INT_MAX;
3437 	} else {
3438 		ret = kstrtoint(buf, 0, &descendants);
3439 		if (ret)
3440 			return ret;
3441 	}
3442 
3443 	if (descendants < 0)
3444 		return -ERANGE;
3445 
3446 	cgrp = cgroup_kn_lock_live(of->kn, false);
3447 	if (!cgrp)
3448 		return -ENOENT;
3449 
3450 	cgrp->max_descendants = descendants;
3451 
3452 	cgroup_kn_unlock(of->kn);
3453 
3454 	return nbytes;
3455 }
3456 
3457 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3458 {
3459 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3460 	int depth = READ_ONCE(cgrp->max_depth);
3461 
3462 	if (depth == INT_MAX)
3463 		seq_puts(seq, "max\n");
3464 	else
3465 		seq_printf(seq, "%d\n", depth);
3466 
3467 	return 0;
3468 }
3469 
3470 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3471 				      char *buf, size_t nbytes, loff_t off)
3472 {
3473 	struct cgroup *cgrp;
3474 	ssize_t ret;
3475 	int depth;
3476 
3477 	buf = strstrip(buf);
3478 	if (!strcmp(buf, "max")) {
3479 		depth = INT_MAX;
3480 	} else {
3481 		ret = kstrtoint(buf, 0, &depth);
3482 		if (ret)
3483 			return ret;
3484 	}
3485 
3486 	if (depth < 0)
3487 		return -ERANGE;
3488 
3489 	cgrp = cgroup_kn_lock_live(of->kn, false);
3490 	if (!cgrp)
3491 		return -ENOENT;
3492 
3493 	cgrp->max_depth = depth;
3494 
3495 	cgroup_kn_unlock(of->kn);
3496 
3497 	return nbytes;
3498 }
3499 
3500 static int cgroup_events_show(struct seq_file *seq, void *v)
3501 {
3502 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3503 
3504 	seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3505 	seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3506 
3507 	return 0;
3508 }
3509 
3510 static int cgroup_stat_show(struct seq_file *seq, void *v)
3511 {
3512 	struct cgroup *cgroup = seq_css(seq)->cgroup;
3513 
3514 	seq_printf(seq, "nr_descendants %d\n",
3515 		   cgroup->nr_descendants);
3516 	seq_printf(seq, "nr_dying_descendants %d\n",
3517 		   cgroup->nr_dying_descendants);
3518 
3519 	return 0;
3520 }
3521 
3522 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3523 						 struct cgroup *cgrp, int ssid)
3524 {
3525 	struct cgroup_subsys *ss = cgroup_subsys[ssid];
3526 	struct cgroup_subsys_state *css;
3527 	int ret;
3528 
3529 	if (!ss->css_extra_stat_show)
3530 		return 0;
3531 
3532 	css = cgroup_tryget_css(cgrp, ss);
3533 	if (!css)
3534 		return 0;
3535 
3536 	ret = ss->css_extra_stat_show(seq, css);
3537 	css_put(css);
3538 	return ret;
3539 }
3540 
3541 static int cpu_stat_show(struct seq_file *seq, void *v)
3542 {
3543 	struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3544 	int ret = 0;
3545 
3546 	cgroup_base_stat_cputime_show(seq);
3547 #ifdef CONFIG_CGROUP_SCHED
3548 	ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3549 #endif
3550 	return ret;
3551 }
3552 
3553 #ifdef CONFIG_PSI
3554 static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3555 {
3556 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3557 	struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3558 
3559 	return psi_show(seq, psi, PSI_IO);
3560 }
3561 static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3562 {
3563 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3564 	struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3565 
3566 	return psi_show(seq, psi, PSI_MEM);
3567 }
3568 static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3569 {
3570 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3571 	struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3572 
3573 	return psi_show(seq, psi, PSI_CPU);
3574 }
3575 
3576 static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3577 					  size_t nbytes, enum psi_res res)
3578 {
3579 	struct psi_trigger *new;
3580 	struct cgroup *cgrp;
3581 	struct psi_group *psi;
3582 
3583 	cgrp = cgroup_kn_lock_live(of->kn, false);
3584 	if (!cgrp)
3585 		return -ENODEV;
3586 
3587 	cgroup_get(cgrp);
3588 	cgroup_kn_unlock(of->kn);
3589 
3590 	psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3591 	new = psi_trigger_create(psi, buf, nbytes, res);
3592 	if (IS_ERR(new)) {
3593 		cgroup_put(cgrp);
3594 		return PTR_ERR(new);
3595 	}
3596 
3597 	psi_trigger_replace(&of->priv, new);
3598 
3599 	cgroup_put(cgrp);
3600 
3601 	return nbytes;
3602 }
3603 
3604 static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3605 					  char *buf, size_t nbytes,
3606 					  loff_t off)
3607 {
3608 	return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3609 }
3610 
3611 static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3612 					  char *buf, size_t nbytes,
3613 					  loff_t off)
3614 {
3615 	return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3616 }
3617 
3618 static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3619 					  char *buf, size_t nbytes,
3620 					  loff_t off)
3621 {
3622 	return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3623 }
3624 
3625 static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3626 					  poll_table *pt)
3627 {
3628 	return psi_trigger_poll(&of->priv, of->file, pt);
3629 }
3630 
3631 static void cgroup_pressure_release(struct kernfs_open_file *of)
3632 {
3633 	psi_trigger_replace(&of->priv, NULL);
3634 }
3635 #endif /* CONFIG_PSI */
3636 
3637 static int cgroup_freeze_show(struct seq_file *seq, void *v)
3638 {
3639 	struct cgroup *cgrp = seq_css(seq)->cgroup;
3640 
3641 	seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3642 
3643 	return 0;
3644 }
3645 
3646 static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3647 				   char *buf, size_t nbytes, loff_t off)
3648 {
3649 	struct cgroup *cgrp;
3650 	ssize_t ret;
3651 	int freeze;
3652 
3653 	ret = kstrtoint(strstrip(buf), 0, &freeze);
3654 	if (ret)
3655 		return ret;
3656 
3657 	if (freeze < 0 || freeze > 1)
3658 		return -ERANGE;
3659 
3660 	cgrp = cgroup_kn_lock_live(of->kn, false);
3661 	if (!cgrp)
3662 		return -ENOENT;
3663 
3664 	cgroup_freeze(cgrp, freeze);
3665 
3666 	cgroup_kn_unlock(of->kn);
3667 
3668 	return nbytes;
3669 }
3670 
3671 static int cgroup_file_open(struct kernfs_open_file *of)
3672 {
3673 	struct cftype *cft = of_cft(of);
3674 
3675 	if (cft->open)
3676 		return cft->open(of);
3677 	return 0;
3678 }
3679 
3680 static void cgroup_file_release(struct kernfs_open_file *of)
3681 {
3682 	struct cftype *cft = of_cft(of);
3683 
3684 	if (cft->release)
3685 		cft->release(of);
3686 }
3687 
3688 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3689 				 size_t nbytes, loff_t off)
3690 {
3691 	struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3692 	struct cgroup *cgrp = of->kn->parent->priv;
3693 	struct cftype *cft = of_cft(of);
3694 	struct cgroup_subsys_state *css;
3695 	int ret;
3696 
3697 	if (!nbytes)
3698 		return 0;
3699 
3700 	/*
3701 	 * If namespaces are delegation boundaries, disallow writes to
3702 	 * files in an non-init namespace root from inside the namespace
3703 	 * except for the files explicitly marked delegatable -
3704 	 * cgroup.procs and cgroup.subtree_control.
3705 	 */
3706 	if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3707 	    !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3708 	    ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3709 		return -EPERM;
3710 
3711 	if (cft->write)
3712 		return cft->write(of, buf, nbytes, off);
3713 
3714 	/*
3715 	 * kernfs guarantees that a file isn't deleted with operations in
3716 	 * flight, which means that the matching css is and stays alive and
3717 	 * doesn't need to be pinned.  The RCU locking is not necessary
3718 	 * either.  It's just for the convenience of using cgroup_css().
3719 	 */
3720 	rcu_read_lock();
3721 	css = cgroup_css(cgrp, cft->ss);
3722 	rcu_read_unlock();
3723 
3724 	if (cft->write_u64) {
3725 		unsigned long long v;
3726 		ret = kstrtoull(buf, 0, &v);
3727 		if (!ret)
3728 			ret = cft->write_u64(css, cft, v);
3729 	} else if (cft->write_s64) {
3730 		long long v;
3731 		ret = kstrtoll(buf, 0, &v);
3732 		if (!ret)
3733 			ret = cft->write_s64(css, cft, v);
3734 	} else {
3735 		ret = -EINVAL;
3736 	}
3737 
3738 	return ret ?: nbytes;
3739 }
3740 
3741 static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3742 {
3743 	struct cftype *cft = of_cft(of);
3744 
3745 	if (cft->poll)
3746 		return cft->poll(of, pt);
3747 
3748 	return kernfs_generic_poll(of, pt);
3749 }
3750 
3751 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3752 {
3753 	return seq_cft(seq)->seq_start(seq, ppos);
3754 }
3755 
3756 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3757 {
3758 	return seq_cft(seq)->seq_next(seq, v, ppos);
3759 }
3760 
3761 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3762 {
3763 	if (seq_cft(seq)->seq_stop)
3764 		seq_cft(seq)->seq_stop(seq, v);
3765 }
3766 
3767 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3768 {
3769 	struct cftype *cft = seq_cft(m);
3770 	struct cgroup_subsys_state *css = seq_css(m);
3771 
3772 	if (cft->seq_show)
3773 		return cft->seq_show(m, arg);
3774 
3775 	if (cft->read_u64)
3776 		seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3777 	else if (cft->read_s64)
3778 		seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3779 	else
3780 		return -EINVAL;
3781 	return 0;
3782 }
3783 
3784 static struct kernfs_ops cgroup_kf_single_ops = {
3785 	.atomic_write_len	= PAGE_SIZE,
3786 	.open			= cgroup_file_open,
3787 	.release		= cgroup_file_release,
3788 	.write			= cgroup_file_write,
3789 	.poll			= cgroup_file_poll,
3790 	.seq_show		= cgroup_seqfile_show,
3791 };
3792 
3793 static struct kernfs_ops cgroup_kf_ops = {
3794 	.atomic_write_len	= PAGE_SIZE,
3795 	.open			= cgroup_file_open,
3796 	.release		= cgroup_file_release,
3797 	.write			= cgroup_file_write,
3798 	.poll			= cgroup_file_poll,
3799 	.seq_start		= cgroup_seqfile_start,
3800 	.seq_next		= cgroup_seqfile_next,
3801 	.seq_stop		= cgroup_seqfile_stop,
3802 	.seq_show		= cgroup_seqfile_show,
3803 };
3804 
3805 /* set uid and gid of cgroup dirs and files to that of the creator */
3806 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3807 {
3808 	struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3809 			       .ia_uid = current_fsuid(),
3810 			       .ia_gid = current_fsgid(), };
3811 
3812 	if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3813 	    gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3814 		return 0;
3815 
3816 	return kernfs_setattr(kn, &iattr);
3817 }
3818 
3819 static void cgroup_file_notify_timer(struct timer_list *timer)
3820 {
3821 	cgroup_file_notify(container_of(timer, struct cgroup_file,
3822 					notify_timer));
3823 }
3824 
3825 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3826 			   struct cftype *cft)
3827 {
3828 	char name[CGROUP_FILE_NAME_MAX];
3829 	struct kernfs_node *kn;
3830 	struct lock_class_key *key = NULL;
3831 	int ret;
3832 
3833 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3834 	key = &cft->lockdep_key;
3835 #endif
3836 	kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3837 				  cgroup_file_mode(cft),
3838 				  GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
3839 				  0, cft->kf_ops, cft,
3840 				  NULL, key);
3841 	if (IS_ERR(kn))
3842 		return PTR_ERR(kn);
3843 
3844 	ret = cgroup_kn_set_ugid(kn);
3845 	if (ret) {
3846 		kernfs_remove(kn);
3847 		return ret;
3848 	}
3849 
3850 	if (cft->file_offset) {
3851 		struct cgroup_file *cfile = (void *)css + cft->file_offset;
3852 
3853 		timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
3854 
3855 		spin_lock_irq(&cgroup_file_kn_lock);
3856 		cfile->kn = kn;
3857 		spin_unlock_irq(&cgroup_file_kn_lock);
3858 	}
3859 
3860 	return 0;
3861 }
3862 
3863 /**
3864  * cgroup_addrm_files - add or remove files to a cgroup directory
3865  * @css: the target css
3866  * @cgrp: the target cgroup (usually css->cgroup)
3867  * @cfts: array of cftypes to be added
3868  * @is_add: whether to add or remove
3869  *
3870  * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3871  * For removals, this function never fails.
3872  */
3873 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3874 			      struct cgroup *cgrp, struct cftype cfts[],
3875 			      bool is_add)
3876 {
3877 	struct cftype *cft, *cft_end = NULL;
3878 	int ret = 0;
3879 
3880 	lockdep_assert_held(&cgroup_mutex);
3881 
3882 restart:
3883 	for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3884 		/* does cft->flags tell us to skip this file on @cgrp? */
3885 		if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3886 			continue;
3887 		if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3888 			continue;
3889 		if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3890 			continue;
3891 		if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3892 			continue;
3893 		if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
3894 			continue;
3895 		if (is_add) {
3896 			ret = cgroup_add_file(css, cgrp, cft);
3897 			if (ret) {
3898 				pr_warn("%s: failed to add %s, err=%d\n",
3899 					__func__, cft->name, ret);
3900 				cft_end = cft;
3901 				is_add = false;
3902 				goto restart;
3903 			}
3904 		} else {
3905 			cgroup_rm_file(cgrp, cft);
3906 		}
3907 	}
3908 	return ret;
3909 }
3910 
3911 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3912 {
3913 	struct cgroup_subsys *ss = cfts[0].ss;
3914 	struct cgroup *root = &ss->root->cgrp;
3915 	struct cgroup_subsys_state *css;
3916 	int ret = 0;
3917 
3918 	lockdep_assert_held(&cgroup_mutex);
3919 
3920 	/* add/rm files for all cgroups created before */
3921 	css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3922 		struct cgroup *cgrp = css->cgroup;
3923 
3924 		if (!(css->flags & CSS_VISIBLE))
3925 			continue;
3926 
3927 		ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3928 		if (ret)
3929 			break;
3930 	}
3931 
3932 	if (is_add && !ret)
3933 		kernfs_activate(root->kn);
3934 	return ret;
3935 }
3936 
3937 static void cgroup_exit_cftypes(struct cftype *cfts)
3938 {
3939 	struct cftype *cft;
3940 
3941 	for (cft = cfts; cft->name[0] != '\0'; cft++) {
3942 		/* free copy for custom atomic_write_len, see init_cftypes() */
3943 		if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3944 			kfree(cft->kf_ops);
3945 		cft->kf_ops = NULL;
3946 		cft->ss = NULL;
3947 
3948 		/* revert flags set by cgroup core while adding @cfts */
3949 		cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3950 	}
3951 }
3952 
3953 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3954 {
3955 	struct cftype *cft;
3956 
3957 	for (cft = cfts; cft->name[0] != '\0'; cft++) {
3958 		struct kernfs_ops *kf_ops;
3959 
3960 		WARN_ON(cft->ss || cft->kf_ops);
3961 
3962 		if (cft->seq_start)
3963 			kf_ops = &cgroup_kf_ops;
3964 		else
3965 			kf_ops = &cgroup_kf_single_ops;
3966 
3967 		/*
3968 		 * Ugh... if @cft wants a custom max_write_len, we need to
3969 		 * make a copy of kf_ops to set its atomic_write_len.
3970 		 */
3971 		if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3972 			kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3973 			if (!kf_ops) {
3974 				cgroup_exit_cftypes(cfts);
3975 				return -ENOMEM;
3976 			}
3977 			kf_ops->atomic_write_len = cft->max_write_len;
3978 		}
3979 
3980 		cft->kf_ops = kf_ops;
3981 		cft->ss = ss;
3982 	}
3983 
3984 	return 0;
3985 }
3986 
3987 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3988 {
3989 	lockdep_assert_held(&cgroup_mutex);
3990 
3991 	if (!cfts || !cfts[0].ss)
3992 		return -ENOENT;
3993 
3994 	list_del(&cfts->node);
3995 	cgroup_apply_cftypes(cfts, false);
3996 	cgroup_exit_cftypes(cfts);
3997 	return 0;
3998 }
3999 
4000 /**
4001  * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4002  * @cfts: zero-length name terminated array of cftypes
4003  *
4004  * Unregister @cfts.  Files described by @cfts are removed from all
4005  * existing cgroups and all future cgroups won't have them either.  This
4006  * function can be called anytime whether @cfts' subsys is attached or not.
4007  *
4008  * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4009  * registered.
4010  */
4011 int cgroup_rm_cftypes(struct cftype *cfts)
4012 {
4013 	int ret;
4014 
4015 	mutex_lock(&cgroup_mutex);
4016 	ret = cgroup_rm_cftypes_locked(cfts);
4017 	mutex_unlock(&cgroup_mutex);
4018 	return ret;
4019 }
4020 
4021 /**
4022  * cgroup_add_cftypes - add an array of cftypes to a subsystem
4023  * @ss: target cgroup subsystem
4024  * @cfts: zero-length name terminated array of cftypes
4025  *
4026  * Register @cfts to @ss.  Files described by @cfts are created for all
4027  * existing cgroups to which @ss is attached and all future cgroups will
4028  * have them too.  This function can be called anytime whether @ss is
4029  * attached or not.
4030  *
4031  * Returns 0 on successful registration, -errno on failure.  Note that this
4032  * function currently returns 0 as long as @cfts registration is successful
4033  * even if some file creation attempts on existing cgroups fail.
4034  */
4035 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4036 {
4037 	int ret;
4038 
4039 	if (!cgroup_ssid_enabled(ss->id))
4040 		return 0;
4041 
4042 	if (!cfts || cfts[0].name[0] == '\0')
4043 		return 0;
4044 
4045 	ret = cgroup_init_cftypes(ss, cfts);
4046 	if (ret)
4047 		return ret;
4048 
4049 	mutex_lock(&cgroup_mutex);
4050 
4051 	list_add_tail(&cfts->node, &ss->cfts);
4052 	ret = cgroup_apply_cftypes(cfts, true);
4053 	if (ret)
4054 		cgroup_rm_cftypes_locked(cfts);
4055 
4056 	mutex_unlock(&cgroup_mutex);
4057 	return ret;
4058 }
4059 
4060 /**
4061  * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4062  * @ss: target cgroup subsystem
4063  * @cfts: zero-length name terminated array of cftypes
4064  *
4065  * Similar to cgroup_add_cftypes() but the added files are only used for
4066  * the default hierarchy.
4067  */
4068 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4069 {
4070 	struct cftype *cft;
4071 
4072 	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4073 		cft->flags |= __CFTYPE_ONLY_ON_DFL;
4074 	return cgroup_add_cftypes(ss, cfts);
4075 }
4076 
4077 /**
4078  * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4079  * @ss: target cgroup subsystem
4080  * @cfts: zero-length name terminated array of cftypes
4081  *
4082  * Similar to cgroup_add_cftypes() but the added files are only used for
4083  * the legacy hierarchies.
4084  */
4085 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4086 {
4087 	struct cftype *cft;
4088 
4089 	for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4090 		cft->flags |= __CFTYPE_NOT_ON_DFL;
4091 	return cgroup_add_cftypes(ss, cfts);
4092 }
4093 
4094 /**
4095  * cgroup_file_notify - generate a file modified event for a cgroup_file
4096  * @cfile: target cgroup_file
4097  *
4098  * @cfile must have been obtained by setting cftype->file_offset.
4099  */
4100 void cgroup_file_notify(struct cgroup_file *cfile)
4101 {
4102 	unsigned long flags;
4103 
4104 	spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4105 	if (cfile->kn) {
4106 		unsigned long last = cfile->notified_at;
4107 		unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4108 
4109 		if (time_in_range(jiffies, last, next)) {
4110 			timer_reduce(&cfile->notify_timer, next);
4111 		} else {
4112 			kernfs_notify(cfile->kn);
4113 			cfile->notified_at = jiffies;
4114 		}
4115 	}
4116 	spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4117 }
4118 
4119 /**
4120  * css_next_child - find the next child of a given css
4121  * @pos: the current position (%NULL to initiate traversal)
4122  * @parent: css whose children to walk
4123  *
4124  * This function returns the next child of @parent and should be called
4125  * under either cgroup_mutex or RCU read lock.  The only requirement is
4126  * that @parent and @pos are accessible.  The next sibling is guaranteed to
4127  * be returned regardless of their states.
4128  *
4129  * If a subsystem synchronizes ->css_online() and the start of iteration, a
4130  * css which finished ->css_online() is guaranteed to be visible in the
4131  * future iterations and will stay visible until the last reference is put.
4132  * A css which hasn't finished ->css_online() or already finished
4133  * ->css_offline() may show up during traversal.  It's each subsystem's
4134  * responsibility to synchronize against on/offlining.
4135  */
4136 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4137 					   struct cgroup_subsys_state *parent)
4138 {
4139 	struct cgroup_subsys_state *next;
4140 
4141 	cgroup_assert_mutex_or_rcu_locked();
4142 
4143 	/*
4144 	 * @pos could already have been unlinked from the sibling list.
4145 	 * Once a cgroup is removed, its ->sibling.next is no longer
4146 	 * updated when its next sibling changes.  CSS_RELEASED is set when
4147 	 * @pos is taken off list, at which time its next pointer is valid,
4148 	 * and, as releases are serialized, the one pointed to by the next
4149 	 * pointer is guaranteed to not have started release yet.  This
4150 	 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4151 	 * critical section, the one pointed to by its next pointer is
4152 	 * guaranteed to not have finished its RCU grace period even if we
4153 	 * have dropped rcu_read_lock() in-between iterations.
4154 	 *
4155 	 * If @pos has CSS_RELEASED set, its next pointer can't be
4156 	 * dereferenced; however, as each css is given a monotonically
4157 	 * increasing unique serial number and always appended to the
4158 	 * sibling list, the next one can be found by walking the parent's
4159 	 * children until the first css with higher serial number than
4160 	 * @pos's.  While this path can be slower, it happens iff iteration
4161 	 * races against release and the race window is very small.
4162 	 */
4163 	if (!pos) {
4164 		next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4165 	} else if (likely(!(pos->flags & CSS_RELEASED))) {
4166 		next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4167 	} else {
4168 		list_for_each_entry_rcu(next, &parent->children, sibling,
4169 					lockdep_is_held(&cgroup_mutex))
4170 			if (next->serial_nr > pos->serial_nr)
4171 				break;
4172 	}
4173 
4174 	/*
4175 	 * @next, if not pointing to the head, can be dereferenced and is
4176 	 * the next sibling.
4177 	 */
4178 	if (&next->sibling != &parent->children)
4179 		return next;
4180 	return NULL;
4181 }
4182 
4183 /**
4184  * css_next_descendant_pre - find the next descendant for pre-order walk
4185  * @pos: the current position (%NULL to initiate traversal)
4186  * @root: css whose descendants to walk
4187  *
4188  * To be used by css_for_each_descendant_pre().  Find the next descendant
4189  * to visit for pre-order traversal of @root's descendants.  @root is
4190  * included in the iteration and the first node to be visited.
4191  *
4192  * While this function requires cgroup_mutex or RCU read locking, it
4193  * doesn't require the whole traversal to be contained in a single critical
4194  * section.  This function will return the correct next descendant as long
4195  * as both @pos and @root are accessible and @pos is a descendant of @root.
4196  *
4197  * If a subsystem synchronizes ->css_online() and the start of iteration, a
4198  * css which finished ->css_online() is guaranteed to be visible in the
4199  * future iterations and will stay visible until the last reference is put.
4200  * A css which hasn't finished ->css_online() or already finished
4201  * ->css_offline() may show up during traversal.  It's each subsystem's
4202  * responsibility to synchronize against on/offlining.
4203  */
4204 struct cgroup_subsys_state *
4205 css_next_descendant_pre(struct cgroup_subsys_state *pos,
4206 			struct cgroup_subsys_state *root)
4207 {
4208 	struct cgroup_subsys_state *next;
4209 
4210 	cgroup_assert_mutex_or_rcu_locked();
4211 
4212 	/* if first iteration, visit @root */
4213 	if (!pos)
4214 		return root;
4215 
4216 	/* visit the first child if exists */
4217 	next = css_next_child(NULL, pos);
4218 	if (next)
4219 		return next;
4220 
4221 	/* no child, visit my or the closest ancestor's next sibling */
4222 	while (pos != root) {
4223 		next = css_next_child(pos, pos->parent);
4224 		if (next)
4225 			return next;
4226 		pos = pos->parent;
4227 	}
4228 
4229 	return NULL;
4230 }
4231 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4232 
4233 /**
4234  * css_rightmost_descendant - return the rightmost descendant of a css
4235  * @pos: css of interest
4236  *
4237  * Return the rightmost descendant of @pos.  If there's no descendant, @pos
4238  * is returned.  This can be used during pre-order traversal to skip
4239  * subtree of @pos.
4240  *
4241  * While this function requires cgroup_mutex or RCU read locking, it
4242  * doesn't require the whole traversal to be contained in a single critical
4243  * section.  This function will return the correct rightmost descendant as
4244  * long as @pos is accessible.
4245  */
4246 struct cgroup_subsys_state *
4247 css_rightmost_descendant(struct cgroup_subsys_state *pos)
4248 {
4249 	struct cgroup_subsys_state *last, *tmp;
4250 
4251 	cgroup_assert_mutex_or_rcu_locked();
4252 
4253 	do {
4254 		last = pos;
4255 		/* ->prev isn't RCU safe, walk ->next till the end */
4256 		pos = NULL;
4257 		css_for_each_child(tmp, last)
4258 			pos = tmp;
4259 	} while (pos);
4260 
4261 	return last;
4262 }
4263 
4264 static struct cgroup_subsys_state *
4265 css_leftmost_descendant(struct cgroup_subsys_state *pos)
4266 {
4267 	struct cgroup_subsys_state *last;
4268 
4269 	do {
4270 		last = pos;
4271 		pos = css_next_child(NULL, pos);
4272 	} while (pos);
4273 
4274 	return last;
4275 }
4276 
4277 /**
4278  * css_next_descendant_post - find the next descendant for post-order walk
4279  * @pos: the current position (%NULL to initiate traversal)
4280  * @root: css whose descendants to walk
4281  *
4282  * To be used by css_for_each_descendant_post().  Find the next descendant
4283  * to visit for post-order traversal of @root's descendants.  @root is
4284  * included in the iteration and the last node to be visited.
4285  *
4286  * While this function requires cgroup_mutex or RCU read locking, it
4287  * doesn't require the whole traversal to be contained in a single critical
4288  * section.  This function will return the correct next descendant as long
4289  * as both @pos and @cgroup are accessible and @pos is a descendant of
4290  * @cgroup.
4291  *
4292  * If a subsystem synchronizes ->css_online() and the start of iteration, a
4293  * css which finished ->css_online() is guaranteed to be visible in the
4294  * future iterations and will stay visible until the last reference is put.
4295  * A css which hasn't finished ->css_online() or already finished
4296  * ->css_offline() may show up during traversal.  It's each subsystem's
4297  * responsibility to synchronize against on/offlining.
4298  */
4299 struct cgroup_subsys_state *
4300 css_next_descendant_post(struct cgroup_subsys_state *pos,
4301 			 struct cgroup_subsys_state *root)
4302 {
4303 	struct cgroup_subsys_state *next;
4304 
4305 	cgroup_assert_mutex_or_rcu_locked();
4306 
4307 	/* if first iteration, visit leftmost descendant which may be @root */
4308 	if (!pos)
4309 		return css_leftmost_descendant(root);
4310 
4311 	/* if we visited @root, we're done */
4312 	if (pos == root)
4313 		return NULL;
4314 
4315 	/* if there's an unvisited sibling, visit its leftmost descendant */
4316 	next = css_next_child(pos, pos->parent);
4317 	if (next)
4318 		return css_leftmost_descendant(next);
4319 
4320 	/* no sibling left, visit parent */
4321 	return pos->parent;
4322 }
4323 
4324 /**
4325  * css_has_online_children - does a css have online children
4326  * @css: the target css
4327  *
4328  * Returns %true if @css has any online children; otherwise, %false.  This
4329  * function can be called from any context but the caller is responsible
4330  * for synchronizing against on/offlining as necessary.
4331  */
4332 bool css_has_online_children(struct cgroup_subsys_state *css)
4333 {
4334 	struct cgroup_subsys_state *child;
4335 	bool ret = false;
4336 
4337 	rcu_read_lock();
4338 	css_for_each_child(child, css) {
4339 		if (child->flags & CSS_ONLINE) {
4340 			ret = true;
4341 			break;
4342 		}
4343 	}
4344 	rcu_read_unlock();
4345 	return ret;
4346 }
4347 
4348 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4349 {
4350 	struct list_head *l;
4351 	struct cgrp_cset_link *link;
4352 	struct css_set *cset;
4353 
4354 	lockdep_assert_held(&css_set_lock);
4355 
4356 	/* find the next threaded cset */
4357 	if (it->tcset_pos) {
4358 		l = it->tcset_pos->next;
4359 
4360 		if (l != it->tcset_head) {
4361 			it->tcset_pos = l;
4362 			return container_of(l, struct css_set,
4363 					    threaded_csets_node);
4364 		}
4365 
4366 		it->tcset_pos = NULL;
4367 	}
4368 
4369 	/* find the next cset */
4370 	l = it->cset_pos;
4371 	l = l->next;
4372 	if (l == it->cset_head) {
4373 		it->cset_pos = NULL;
4374 		return NULL;
4375 	}
4376 
4377 	if (it->ss) {
4378 		cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4379 	} else {
4380 		link = list_entry(l, struct cgrp_cset_link, cset_link);
4381 		cset = link->cset;
4382 	}
4383 
4384 	it->cset_pos = l;
4385 
4386 	/* initialize threaded css_set walking */
4387 	if (it->flags & CSS_TASK_ITER_THREADED) {
4388 		if (it->cur_dcset)
4389 			put_css_set_locked(it->cur_dcset);
4390 		it->cur_dcset = cset;
4391 		get_css_set(cset);
4392 
4393 		it->tcset_head = &cset->threaded_csets;
4394 		it->tcset_pos = &cset->threaded_csets;
4395 	}
4396 
4397 	return cset;
4398 }
4399 
4400 /**
4401  * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4402  * @it: the iterator to advance
4403  *
4404  * Advance @it to the next css_set to walk.
4405  */
4406 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4407 {
4408 	struct css_set *cset;
4409 
4410 	lockdep_assert_held(&css_set_lock);
4411 
4412 	/* Advance to the next non-empty css_set and find first non-empty tasks list*/
4413 	while ((cset = css_task_iter_next_css_set(it))) {
4414 		if (!list_empty(&cset->tasks)) {
4415 			it->cur_tasks_head = &cset->tasks;
4416 			break;
4417 		} else if (!list_empty(&cset->mg_tasks)) {
4418 			it->cur_tasks_head = &cset->mg_tasks;
4419 			break;
4420 		} else if (!list_empty(&cset->dying_tasks)) {
4421 			it->cur_tasks_head = &cset->dying_tasks;
4422 			break;
4423 		}
4424 	}
4425 	if (!cset) {
4426 		it->task_pos = NULL;
4427 		return;
4428 	}
4429 	it->task_pos = it->cur_tasks_head->next;
4430 
4431 	/*
4432 	 * We don't keep css_sets locked across iteration steps and thus
4433 	 * need to take steps to ensure that iteration can be resumed after
4434 	 * the lock is re-acquired.  Iteration is performed at two levels -
4435 	 * css_sets and tasks in them.
4436 	 *
4437 	 * Once created, a css_set never leaves its cgroup lists, so a
4438 	 * pinned css_set is guaranteed to stay put and we can resume
4439 	 * iteration afterwards.
4440 	 *
4441 	 * Tasks may leave @cset across iteration steps.  This is resolved
4442 	 * by registering each iterator with the css_set currently being
4443 	 * walked and making css_set_move_task() advance iterators whose
4444 	 * next task is leaving.
4445 	 */
4446 	if (it->cur_cset) {
4447 		list_del(&it->iters_node);
4448 		put_css_set_locked(it->cur_cset);
4449 	}
4450 	get_css_set(cset);
4451 	it->cur_cset = cset;
4452 	list_add(&it->iters_node, &cset->task_iters);
4453 }
4454 
4455 static void css_task_iter_skip(struct css_task_iter *it,
4456 			       struct task_struct *task)
4457 {
4458 	lockdep_assert_held(&css_set_lock);
4459 
4460 	if (it->task_pos == &task->cg_list) {
4461 		it->task_pos = it->task_pos->next;
4462 		it->flags |= CSS_TASK_ITER_SKIPPED;
4463 	}
4464 }
4465 
4466 static void css_task_iter_advance(struct css_task_iter *it)
4467 {
4468 	struct task_struct *task;
4469 
4470 	lockdep_assert_held(&css_set_lock);
4471 repeat:
4472 	if (it->task_pos) {
4473 		/*
4474 		 * Advance iterator to find next entry. We go through cset
4475 		 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4476 		 * the next cset.
4477 		 */
4478 		if (it->flags & CSS_TASK_ITER_SKIPPED)
4479 			it->flags &= ~CSS_TASK_ITER_SKIPPED;
4480 		else
4481 			it->task_pos = it->task_pos->next;
4482 
4483 		if (it->task_pos == &it->cur_cset->tasks) {
4484 			it->cur_tasks_head = &it->cur_cset->mg_tasks;
4485 			it->task_pos = it->cur_tasks_head->next;
4486 		}
4487 		if (it->task_pos == &it->cur_cset->mg_tasks) {
4488 			it->cur_tasks_head = &it->cur_cset->dying_tasks;
4489 			it->task_pos = it->cur_tasks_head->next;
4490 		}
4491 		if (it->task_pos == &it->cur_cset->dying_tasks)
4492 			css_task_iter_advance_css_set(it);
4493 	} else {
4494 		/* called from start, proceed to the first cset */
4495 		css_task_iter_advance_css_set(it);
4496 	}
4497 
4498 	if (!it->task_pos)
4499 		return;
4500 
4501 	task = list_entry(it->task_pos, struct task_struct, cg_list);
4502 
4503 	if (it->flags & CSS_TASK_ITER_PROCS) {
4504 		/* if PROCS, skip over tasks which aren't group leaders */
4505 		if (!thread_group_leader(task))
4506 			goto repeat;
4507 
4508 		/* and dying leaders w/o live member threads */
4509 		if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
4510 		    !atomic_read(&task->signal->live))
4511 			goto repeat;
4512 	} else {
4513 		/* skip all dying ones */
4514 		if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
4515 			goto repeat;
4516 	}
4517 }
4518 
4519 /**
4520  * css_task_iter_start - initiate task iteration
4521  * @css: the css to walk tasks of
4522  * @flags: CSS_TASK_ITER_* flags
4523  * @it: the task iterator to use
4524  *
4525  * Initiate iteration through the tasks of @css.  The caller can call
4526  * css_task_iter_next() to walk through the tasks until the function
4527  * returns NULL.  On completion of iteration, css_task_iter_end() must be
4528  * called.
4529  */
4530 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4531 			 struct css_task_iter *it)
4532 {
4533 	memset(it, 0, sizeof(*it));
4534 
4535 	spin_lock_irq(&css_set_lock);
4536 
4537 	it->ss = css->ss;
4538 	it->flags = flags;
4539 
4540 	if (it->ss)
4541 		it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4542 	else
4543 		it->cset_pos = &css->cgroup->cset_links;
4544 
4545 	it->cset_head = it->cset_pos;
4546 
4547 	css_task_iter_advance(it);
4548 
4549 	spin_unlock_irq(&css_set_lock);
4550 }
4551 
4552 /**
4553  * css_task_iter_next - return the next task for the iterator
4554  * @it: the task iterator being iterated
4555  *
4556  * The "next" function for task iteration.  @it should have been
4557  * initialized via css_task_iter_start().  Returns NULL when the iteration
4558  * reaches the end.
4559  */
4560 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4561 {
4562 	if (it->cur_task) {
4563 		put_task_struct(it->cur_task);
4564 		it->cur_task = NULL;
4565 	}
4566 
4567 	spin_lock_irq(&css_set_lock);
4568 
4569 	/* @it may be half-advanced by skips, finish advancing */
4570 	if (it->flags & CSS_TASK_ITER_SKIPPED)
4571 		css_task_iter_advance(it);
4572 
4573 	if (it->task_pos) {
4574 		it->cur_task = list_entry(it->task_pos, struct task_struct,
4575 					  cg_list);
4576 		get_task_struct(it->cur_task);
4577 		css_task_iter_advance(it);
4578 	}
4579 
4580 	spin_unlock_irq(&css_set_lock);
4581 
4582 	return it->cur_task;
4583 }
4584 
4585 /**
4586  * css_task_iter_end - finish task iteration
4587  * @it: the task iterator to finish
4588  *
4589  * Finish task iteration started by css_task_iter_start().
4590  */
4591 void css_task_iter_end(struct css_task_iter *it)
4592 {
4593 	if (it->cur_cset) {
4594 		spin_lock_irq(&css_set_lock);
4595 		list_del(&it->iters_node);
4596 		put_css_set_locked(it->cur_cset);
4597 		spin_unlock_irq(&css_set_lock);
4598 	}
4599 
4600 	if (it->cur_dcset)
4601 		put_css_set(it->cur_dcset);
4602 
4603 	if (it->cur_task)
4604 		put_task_struct(it->cur_task);
4605 }
4606 
4607 static void cgroup_procs_release(struct kernfs_open_file *of)
4608 {
4609 	if (of->priv) {
4610 		css_task_iter_end(of->priv);
4611 		kfree(of->priv);
4612 	}
4613 }
4614 
4615 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4616 {
4617 	struct kernfs_open_file *of = s->private;
4618 	struct css_task_iter *it = of->priv;
4619 
4620 	if (pos)
4621 		(*pos)++;
4622 
4623 	return css_task_iter_next(it);
4624 }
4625 
4626 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4627 				  unsigned int iter_flags)
4628 {
4629 	struct kernfs_open_file *of = s->private;
4630 	struct cgroup *cgrp = seq_css(s)->cgroup;
4631 	struct css_task_iter *it = of->priv;
4632 
4633 	/*
4634 	 * When a seq_file is seeked, it's always traversed sequentially
4635 	 * from position 0, so we can simply keep iterating on !0 *pos.
4636 	 */
4637 	if (!it) {
4638 		if (WARN_ON_ONCE((*pos)))
4639 			return ERR_PTR(-EINVAL);
4640 
4641 		it = kzalloc(sizeof(*it), GFP_KERNEL);
4642 		if (!it)
4643 			return ERR_PTR(-ENOMEM);
4644 		of->priv = it;
4645 		css_task_iter_start(&cgrp->self, iter_flags, it);
4646 	} else if (!(*pos)) {
4647 		css_task_iter_end(it);
4648 		css_task_iter_start(&cgrp->self, iter_flags, it);
4649 	} else
4650 		return it->cur_task;
4651 
4652 	return cgroup_procs_next(s, NULL, NULL);
4653 }
4654 
4655 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4656 {
4657 	struct cgroup *cgrp = seq_css(s)->cgroup;
4658 
4659 	/*
4660 	 * All processes of a threaded subtree belong to the domain cgroup
4661 	 * of the subtree.  Only threads can be distributed across the
4662 	 * subtree.  Reject reads on cgroup.procs in the subtree proper.
4663 	 * They're always empty anyway.
4664 	 */
4665 	if (cgroup_is_threaded(cgrp))
4666 		return ERR_PTR(-EOPNOTSUPP);
4667 
4668 	return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4669 					    CSS_TASK_ITER_THREADED);
4670 }
4671 
4672 static int cgroup_procs_show(struct seq_file *s, void *v)
4673 {
4674 	seq_printf(s, "%d\n", task_pid_vnr(v));
4675 	return 0;
4676 }
4677 
4678 static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
4679 {
4680 	int ret;
4681 	struct inode *inode;
4682 
4683 	lockdep_assert_held(&cgroup_mutex);
4684 
4685 	inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
4686 	if (!inode)
4687 		return -ENOMEM;
4688 
4689 	ret = inode_permission(&init_user_ns, inode, MAY_WRITE);
4690 	iput(inode);
4691 	return ret;
4692 }
4693 
4694 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4695 					 struct cgroup *dst_cgrp,
4696 					 struct super_block *sb)
4697 {
4698 	struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4699 	struct cgroup *com_cgrp = src_cgrp;
4700 	int ret;
4701 
4702 	lockdep_assert_held(&cgroup_mutex);
4703 
4704 	/* find the common ancestor */
4705 	while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4706 		com_cgrp = cgroup_parent(com_cgrp);
4707 
4708 	/* %current should be authorized to migrate to the common ancestor */
4709 	ret = cgroup_may_write(com_cgrp, sb);
4710 	if (ret)
4711 		return ret;
4712 
4713 	/*
4714 	 * If namespaces are delegation boundaries, %current must be able
4715 	 * to see both source and destination cgroups from its namespace.
4716 	 */
4717 	if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4718 	    (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4719 	     !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4720 		return -ENOENT;
4721 
4722 	return 0;
4723 }
4724 
4725 static int cgroup_attach_permissions(struct cgroup *src_cgrp,
4726 				     struct cgroup *dst_cgrp,
4727 				     struct super_block *sb, bool threadgroup)
4728 {
4729 	int ret = 0;
4730 
4731 	ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb);
4732 	if (ret)
4733 		return ret;
4734 
4735 	ret = cgroup_migrate_vet_dst(dst_cgrp);
4736 	if (ret)
4737 		return ret;
4738 
4739 	if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
4740 		ret = -EOPNOTSUPP;
4741 
4742 	return ret;
4743 }
4744 
4745 static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
4746 				    bool threadgroup)
4747 {
4748 	struct cgroup *src_cgrp, *dst_cgrp;
4749 	struct task_struct *task;
4750 	ssize_t ret;
4751 	bool locked;
4752 
4753 	dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4754 	if (!dst_cgrp)
4755 		return -ENODEV;
4756 
4757 	task = cgroup_procs_write_start(buf, threadgroup, &locked);
4758 	ret = PTR_ERR_OR_ZERO(task);
4759 	if (ret)
4760 		goto out_unlock;
4761 
4762 	/* find the source cgroup */
4763 	spin_lock_irq(&css_set_lock);
4764 	src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4765 	spin_unlock_irq(&css_set_lock);
4766 
4767 	/* process and thread migrations follow same delegation rule */
4768 	ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4769 					of->file->f_path.dentry->d_sb, threadgroup);
4770 	if (ret)
4771 		goto out_finish;
4772 
4773 	ret = cgroup_attach_task(dst_cgrp, task, threadgroup);
4774 
4775 out_finish:
4776 	cgroup_procs_write_finish(task, locked);
4777 out_unlock:
4778 	cgroup_kn_unlock(of->kn);
4779 
4780 	return ret;
4781 }
4782 
4783 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4784 				  char *buf, size_t nbytes, loff_t off)
4785 {
4786 	return __cgroup_procs_write(of, buf, true) ?: nbytes;
4787 }
4788 
4789 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4790 {
4791 	return __cgroup_procs_start(s, pos, 0);
4792 }
4793 
4794 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4795 				    char *buf, size_t nbytes, loff_t off)
4796 {
4797 	return __cgroup_procs_write(of, buf, false) ?: nbytes;
4798 }
4799 
4800 /* cgroup core interface files for the default hierarchy */
4801 static struct cftype cgroup_base_files[] = {
4802 	{
4803 		.name = "cgroup.type",
4804 		.flags = CFTYPE_NOT_ON_ROOT,
4805 		.seq_show = cgroup_type_show,
4806 		.write = cgroup_type_write,
4807 	},
4808 	{
4809 		.name = "cgroup.procs",
4810 		.flags = CFTYPE_NS_DELEGATABLE,
4811 		.file_offset = offsetof(struct cgroup, procs_file),
4812 		.release = cgroup_procs_release,
4813 		.seq_start = cgroup_procs_start,
4814 		.seq_next = cgroup_procs_next,
4815 		.seq_show = cgroup_procs_show,
4816 		.write = cgroup_procs_write,
4817 	},
4818 	{
4819 		.name = "cgroup.threads",
4820 		.flags = CFTYPE_NS_DELEGATABLE,
4821 		.release = cgroup_procs_release,
4822 		.seq_start = cgroup_threads_start,
4823 		.seq_next = cgroup_procs_next,
4824 		.seq_show = cgroup_procs_show,
4825 		.write = cgroup_threads_write,
4826 	},
4827 	{
4828 		.name = "cgroup.controllers",
4829 		.seq_show = cgroup_controllers_show,
4830 	},
4831 	{
4832 		.name = "cgroup.subtree_control",
4833 		.flags = CFTYPE_NS_DELEGATABLE,
4834 		.seq_show = cgroup_subtree_control_show,
4835 		.write = cgroup_subtree_control_write,
4836 	},
4837 	{
4838 		.name = "cgroup.events",
4839 		.flags = CFTYPE_NOT_ON_ROOT,
4840 		.file_offset = offsetof(struct cgroup, events_file),
4841 		.seq_show = cgroup_events_show,
4842 	},
4843 	{
4844 		.name = "cgroup.max.descendants",
4845 		.seq_show = cgroup_max_descendants_show,
4846 		.write = cgroup_max_descendants_write,
4847 	},
4848 	{
4849 		.name = "cgroup.max.depth",
4850 		.seq_show = cgroup_max_depth_show,
4851 		.write = cgroup_max_depth_write,
4852 	},
4853 	{
4854 		.name = "cgroup.stat",
4855 		.seq_show = cgroup_stat_show,
4856 	},
4857 	{
4858 		.name = "cgroup.freeze",
4859 		.flags = CFTYPE_NOT_ON_ROOT,
4860 		.seq_show = cgroup_freeze_show,
4861 		.write = cgroup_freeze_write,
4862 	},
4863 	{
4864 		.name = "cpu.stat",
4865 		.seq_show = cpu_stat_show,
4866 	},
4867 #ifdef CONFIG_PSI
4868 	{
4869 		.name = "io.pressure",
4870 		.seq_show = cgroup_io_pressure_show,
4871 		.write = cgroup_io_pressure_write,
4872 		.poll = cgroup_pressure_poll,
4873 		.release = cgroup_pressure_release,
4874 	},
4875 	{
4876 		.name = "memory.pressure",
4877 		.seq_show = cgroup_memory_pressure_show,
4878 		.write = cgroup_memory_pressure_write,
4879 		.poll = cgroup_pressure_poll,
4880 		.release = cgroup_pressure_release,
4881 	},
4882 	{
4883 		.name = "cpu.pressure",
4884 		.seq_show = cgroup_cpu_pressure_show,
4885 		.write = cgroup_cpu_pressure_write,
4886 		.poll = cgroup_pressure_poll,
4887 		.release = cgroup_pressure_release,
4888 	},
4889 #endif /* CONFIG_PSI */
4890 	{ }	/* terminate */
4891 };
4892 
4893 /*
4894  * css destruction is four-stage process.
4895  *
4896  * 1. Destruction starts.  Killing of the percpu_ref is initiated.
4897  *    Implemented in kill_css().
4898  *
4899  * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4900  *    and thus css_tryget_online() is guaranteed to fail, the css can be
4901  *    offlined by invoking offline_css().  After offlining, the base ref is
4902  *    put.  Implemented in css_killed_work_fn().
4903  *
4904  * 3. When the percpu_ref reaches zero, the only possible remaining
4905  *    accessors are inside RCU read sections.  css_release() schedules the
4906  *    RCU callback.
4907  *
4908  * 4. After the grace period, the css can be freed.  Implemented in
4909  *    css_free_work_fn().
4910  *
4911  * It is actually hairier because both step 2 and 4 require process context
4912  * and thus involve punting to css->destroy_work adding two additional
4913  * steps to the already complex sequence.
4914  */
4915 static void css_free_rwork_fn(struct work_struct *work)
4916 {
4917 	struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
4918 				struct cgroup_subsys_state, destroy_rwork);
4919 	struct cgroup_subsys *ss = css->ss;
4920 	struct cgroup *cgrp = css->cgroup;
4921 
4922 	percpu_ref_exit(&css->refcnt);
4923 
4924 	if (ss) {
4925 		/* css free path */
4926 		struct cgroup_subsys_state *parent = css->parent;
4927 		int id = css->id;
4928 
4929 		ss->css_free(css);
4930 		cgroup_idr_remove(&ss->css_idr, id);
4931 		cgroup_put(cgrp);
4932 
4933 		if (parent)
4934 			css_put(parent);
4935 	} else {
4936 		/* cgroup free path */
4937 		atomic_dec(&cgrp->root->nr_cgrps);
4938 		cgroup1_pidlist_destroy_all(cgrp);
4939 		cancel_work_sync(&cgrp->release_agent_work);
4940 
4941 		if (cgroup_parent(cgrp)) {
4942 			/*
4943 			 * We get a ref to the parent, and put the ref when
4944 			 * this cgroup is being freed, so it's guaranteed
4945 			 * that the parent won't be destroyed before its
4946 			 * children.
4947 			 */
4948 			cgroup_put(cgroup_parent(cgrp));
4949 			kernfs_put(cgrp->kn);
4950 			psi_cgroup_free(cgrp);
4951 			cgroup_rstat_exit(cgrp);
4952 			kfree(cgrp);
4953 		} else {
4954 			/*
4955 			 * This is root cgroup's refcnt reaching zero,
4956 			 * which indicates that the root should be
4957 			 * released.
4958 			 */
4959 			cgroup_destroy_root(cgrp->root);
4960 		}
4961 	}
4962 }
4963 
4964 static void css_release_work_fn(struct work_struct *work)
4965 {
4966 	struct cgroup_subsys_state *css =
4967 		container_of(work, struct cgroup_subsys_state, destroy_work);
4968 	struct cgroup_subsys *ss = css->ss;
4969 	struct cgroup *cgrp = css->cgroup;
4970 
4971 	mutex_lock(&cgroup_mutex);
4972 
4973 	css->flags |= CSS_RELEASED;
4974 	list_del_rcu(&css->sibling);
4975 
4976 	if (ss) {
4977 		/* css release path */
4978 		if (!list_empty(&css->rstat_css_node)) {
4979 			cgroup_rstat_flush(cgrp);
4980 			list_del_rcu(&css->rstat_css_node);
4981 		}
4982 
4983 		cgroup_idr_replace(&ss->css_idr, NULL, css->id);
4984 		if (ss->css_released)
4985 			ss->css_released(css);
4986 	} else {
4987 		struct cgroup *tcgrp;
4988 
4989 		/* cgroup release path */
4990 		TRACE_CGROUP_PATH(release, cgrp);
4991 
4992 		cgroup_rstat_flush(cgrp);
4993 
4994 		spin_lock_irq(&css_set_lock);
4995 		for (tcgrp = cgroup_parent(cgrp); tcgrp;
4996 		     tcgrp = cgroup_parent(tcgrp))
4997 			tcgrp->nr_dying_descendants--;
4998 		spin_unlock_irq(&css_set_lock);
4999 
5000 		/*
5001 		 * There are two control paths which try to determine
5002 		 * cgroup from dentry without going through kernfs -
5003 		 * cgroupstats_build() and css_tryget_online_from_dir().
5004 		 * Those are supported by RCU protecting clearing of
5005 		 * cgrp->kn->priv backpointer.
5006 		 */
5007 		if (cgrp->kn)
5008 			RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5009 					 NULL);
5010 	}
5011 
5012 	mutex_unlock(&cgroup_mutex);
5013 
5014 	INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5015 	queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5016 }
5017 
5018 static void css_release(struct percpu_ref *ref)
5019 {
5020 	struct cgroup_subsys_state *css =
5021 		container_of(ref, struct cgroup_subsys_state, refcnt);
5022 
5023 	INIT_WORK(&css->destroy_work, css_release_work_fn);
5024 	queue_work(cgroup_destroy_wq, &css->destroy_work);
5025 }
5026 
5027 static void init_and_link_css(struct cgroup_subsys_state *css,
5028 			      struct cgroup_subsys *ss, struct cgroup *cgrp)
5029 {
5030 	lockdep_assert_held(&cgroup_mutex);
5031 
5032 	cgroup_get_live(cgrp);
5033 
5034 	memset(css, 0, sizeof(*css));
5035 	css->cgroup = cgrp;
5036 	css->ss = ss;
5037 	css->id = -1;
5038 	INIT_LIST_HEAD(&css->sibling);
5039 	INIT_LIST_HEAD(&css->children);
5040 	INIT_LIST_HEAD(&css->rstat_css_node);
5041 	css->serial_nr = css_serial_nr_next++;
5042 	atomic_set(&css->online_cnt, 0);
5043 
5044 	if (cgroup_parent(cgrp)) {
5045 		css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5046 		css_get(css->parent);
5047 	}
5048 
5049 	if (ss->css_rstat_flush)
5050 		list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5051 
5052 	BUG_ON(cgroup_css(cgrp, ss));
5053 }
5054 
5055 /* invoke ->css_online() on a new CSS and mark it online if successful */
5056 static int online_css(struct cgroup_subsys_state *css)
5057 {
5058 	struct cgroup_subsys *ss = css->ss;
5059 	int ret = 0;
5060 
5061 	lockdep_assert_held(&cgroup_mutex);
5062 
5063 	if (ss->css_online)
5064 		ret = ss->css_online(css);
5065 	if (!ret) {
5066 		css->flags |= CSS_ONLINE;
5067 		rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5068 
5069 		atomic_inc(&css->online_cnt);
5070 		if (css->parent)
5071 			atomic_inc(&css->parent->online_cnt);
5072 	}
5073 	return ret;
5074 }
5075 
5076 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5077 static void offline_css(struct cgroup_subsys_state *css)
5078 {
5079 	struct cgroup_subsys *ss = css->ss;
5080 
5081 	lockdep_assert_held(&cgroup_mutex);
5082 
5083 	if (!(css->flags & CSS_ONLINE))
5084 		return;
5085 
5086 	if (ss->css_offline)
5087 		ss->css_offline(css);
5088 
5089 	css->flags &= ~CSS_ONLINE;
5090 	RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5091 
5092 	wake_up_all(&css->cgroup->offline_waitq);
5093 }
5094 
5095 /**
5096  * css_create - create a cgroup_subsys_state
5097  * @cgrp: the cgroup new css will be associated with
5098  * @ss: the subsys of new css
5099  *
5100  * Create a new css associated with @cgrp - @ss pair.  On success, the new
5101  * css is online and installed in @cgrp.  This function doesn't create the
5102  * interface files.  Returns 0 on success, -errno on failure.
5103  */
5104 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5105 					      struct cgroup_subsys *ss)
5106 {
5107 	struct cgroup *parent = cgroup_parent(cgrp);
5108 	struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5109 	struct cgroup_subsys_state *css;
5110 	int err;
5111 
5112 	lockdep_assert_held(&cgroup_mutex);
5113 
5114 	css = ss->css_alloc(parent_css);
5115 	if (!css)
5116 		css = ERR_PTR(-ENOMEM);
5117 	if (IS_ERR(css))
5118 		return css;
5119 
5120 	init_and_link_css(css, ss, cgrp);
5121 
5122 	err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5123 	if (err)
5124 		goto err_free_css;
5125 
5126 	err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5127 	if (err < 0)
5128 		goto err_free_css;
5129 	css->id = err;
5130 
5131 	/* @css is ready to be brought online now, make it visible */
5132 	list_add_tail_rcu(&css->sibling, &parent_css->children);
5133 	cgroup_idr_replace(&ss->css_idr, css, css->id);
5134 
5135 	err = online_css(css);
5136 	if (err)
5137 		goto err_list_del;
5138 
5139 	return css;
5140 
5141 err_list_del:
5142 	list_del_rcu(&css->sibling);
5143 err_free_css:
5144 	list_del_rcu(&css->rstat_css_node);
5145 	INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5146 	queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5147 	return ERR_PTR(err);
5148 }
5149 
5150 /*
5151  * The returned cgroup is fully initialized including its control mask, but
5152  * it isn't associated with its kernfs_node and doesn't have the control
5153  * mask applied.
5154  */
5155 static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5156 				    umode_t mode)
5157 {
5158 	struct cgroup_root *root = parent->root;
5159 	struct cgroup *cgrp, *tcgrp;
5160 	struct kernfs_node *kn;
5161 	int level = parent->level + 1;
5162 	int ret;
5163 
5164 	/* allocate the cgroup and its ID, 0 is reserved for the root */
5165 	cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5166 		       GFP_KERNEL);
5167 	if (!cgrp)
5168 		return ERR_PTR(-ENOMEM);
5169 
5170 	ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5171 	if (ret)
5172 		goto out_free_cgrp;
5173 
5174 	ret = cgroup_rstat_init(cgrp);
5175 	if (ret)
5176 		goto out_cancel_ref;
5177 
5178 	/* create the directory */
5179 	kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5180 	if (IS_ERR(kn)) {
5181 		ret = PTR_ERR(kn);
5182 		goto out_stat_exit;
5183 	}
5184 	cgrp->kn = kn;
5185 
5186 	init_cgroup_housekeeping(cgrp);
5187 
5188 	cgrp->self.parent = &parent->self;
5189 	cgrp->root = root;
5190 	cgrp->level = level;
5191 
5192 	ret = psi_cgroup_alloc(cgrp);
5193 	if (ret)
5194 		goto out_kernfs_remove;
5195 
5196 	ret = cgroup_bpf_inherit(cgrp);
5197 	if (ret)
5198 		goto out_psi_free;
5199 
5200 	/*
5201 	 * New cgroup inherits effective freeze counter, and
5202 	 * if the parent has to be frozen, the child has too.
5203 	 */
5204 	cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5205 	if (cgrp->freezer.e_freeze) {
5206 		/*
5207 		 * Set the CGRP_FREEZE flag, so when a process will be
5208 		 * attached to the child cgroup, it will become frozen.
5209 		 * At this point the new cgroup is unpopulated, so we can
5210 		 * consider it frozen immediately.
5211 		 */
5212 		set_bit(CGRP_FREEZE, &cgrp->flags);
5213 		set_bit(CGRP_FROZEN, &cgrp->flags);
5214 	}
5215 
5216 	spin_lock_irq(&css_set_lock);
5217 	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5218 		cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
5219 
5220 		if (tcgrp != cgrp) {
5221 			tcgrp->nr_descendants++;
5222 
5223 			/*
5224 			 * If the new cgroup is frozen, all ancestor cgroups
5225 			 * get a new frozen descendant, but their state can't
5226 			 * change because of this.
5227 			 */
5228 			if (cgrp->freezer.e_freeze)
5229 				tcgrp->freezer.nr_frozen_descendants++;
5230 		}
5231 	}
5232 	spin_unlock_irq(&css_set_lock);
5233 
5234 	if (notify_on_release(parent))
5235 		set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5236 
5237 	if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5238 		set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5239 
5240 	cgrp->self.serial_nr = css_serial_nr_next++;
5241 
5242 	/* allocation complete, commit to creation */
5243 	list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5244 	atomic_inc(&root->nr_cgrps);
5245 	cgroup_get_live(parent);
5246 
5247 	/*
5248 	 * On the default hierarchy, a child doesn't automatically inherit
5249 	 * subtree_control from the parent.  Each is configured manually.
5250 	 */
5251 	if (!cgroup_on_dfl(cgrp))
5252 		cgrp->subtree_control = cgroup_control(cgrp);
5253 
5254 	cgroup_propagate_control(cgrp);
5255 
5256 	return cgrp;
5257 
5258 out_psi_free:
5259 	psi_cgroup_free(cgrp);
5260 out_kernfs_remove:
5261 	kernfs_remove(cgrp->kn);
5262 out_stat_exit:
5263 	cgroup_rstat_exit(cgrp);
5264 out_cancel_ref:
5265 	percpu_ref_exit(&cgrp->self.refcnt);
5266 out_free_cgrp:
5267 	kfree(cgrp);
5268 	return ERR_PTR(ret);
5269 }
5270 
5271 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5272 {
5273 	struct cgroup *cgroup;
5274 	int ret = false;
5275 	int level = 1;
5276 
5277 	lockdep_assert_held(&cgroup_mutex);
5278 
5279 	for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5280 		if (cgroup->nr_descendants >= cgroup->max_descendants)
5281 			goto fail;
5282 
5283 		if (level > cgroup->max_depth)
5284 			goto fail;
5285 
5286 		level++;
5287 	}
5288 
5289 	ret = true;
5290 fail:
5291 	return ret;
5292 }
5293 
5294 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5295 {
5296 	struct cgroup *parent, *cgrp;
5297 	int ret;
5298 
5299 	/* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5300 	if (strchr(name, '\n'))
5301 		return -EINVAL;
5302 
5303 	parent = cgroup_kn_lock_live(parent_kn, false);
5304 	if (!parent)
5305 		return -ENODEV;
5306 
5307 	if (!cgroup_check_hierarchy_limits(parent)) {
5308 		ret = -EAGAIN;
5309 		goto out_unlock;
5310 	}
5311 
5312 	cgrp = cgroup_create(parent, name, mode);
5313 	if (IS_ERR(cgrp)) {
5314 		ret = PTR_ERR(cgrp);
5315 		goto out_unlock;
5316 	}
5317 
5318 	/*
5319 	 * This extra ref will be put in cgroup_free_fn() and guarantees
5320 	 * that @cgrp->kn is always accessible.
5321 	 */
5322 	kernfs_get(cgrp->kn);
5323 
5324 	ret = cgroup_kn_set_ugid(cgrp->kn);
5325 	if (ret)
5326 		goto out_destroy;
5327 
5328 	ret = css_populate_dir(&cgrp->self);
5329 	if (ret)
5330 		goto out_destroy;
5331 
5332 	ret = cgroup_apply_control_enable(cgrp);
5333 	if (ret)
5334 		goto out_destroy;
5335 
5336 	TRACE_CGROUP_PATH(mkdir, cgrp);
5337 
5338 	/* let's create and online css's */
5339 	kernfs_activate(cgrp->kn);
5340 
5341 	ret = 0;
5342 	goto out_unlock;
5343 
5344 out_destroy:
5345 	cgroup_destroy_locked(cgrp);
5346 out_unlock:
5347 	cgroup_kn_unlock(parent_kn);
5348 	return ret;
5349 }
5350 
5351 /*
5352  * This is called when the refcnt of a css is confirmed to be killed.
5353  * css_tryget_online() is now guaranteed to fail.  Tell the subsystem to
5354  * initiate destruction and put the css ref from kill_css().
5355  */
5356 static void css_killed_work_fn(struct work_struct *work)
5357 {
5358 	struct cgroup_subsys_state *css =
5359 		container_of(work, struct cgroup_subsys_state, destroy_work);
5360 
5361 	mutex_lock(&cgroup_mutex);
5362 
5363 	do {
5364 		offline_css(css);
5365 		css_put(css);
5366 		/* @css can't go away while we're holding cgroup_mutex */
5367 		css = css->parent;
5368 	} while (css && atomic_dec_and_test(&css->online_cnt));
5369 
5370 	mutex_unlock(&cgroup_mutex);
5371 }
5372 
5373 /* css kill confirmation processing requires process context, bounce */
5374 static void css_killed_ref_fn(struct percpu_ref *ref)
5375 {
5376 	struct cgroup_subsys_state *css =
5377 		container_of(ref, struct cgroup_subsys_state, refcnt);
5378 
5379 	if (atomic_dec_and_test(&css->online_cnt)) {
5380 		INIT_WORK(&css->destroy_work, css_killed_work_fn);
5381 		queue_work(cgroup_destroy_wq, &css->destroy_work);
5382 	}
5383 }
5384 
5385 /**
5386  * kill_css - destroy a css
5387  * @css: css to destroy
5388  *
5389  * This function initiates destruction of @css by removing cgroup interface
5390  * files and putting its base reference.  ->css_offline() will be invoked
5391  * asynchronously once css_tryget_online() is guaranteed to fail and when
5392  * the reference count reaches zero, @css will be released.
5393  */
5394 static void kill_css(struct cgroup_subsys_state *css)
5395 {
5396 	lockdep_assert_held(&cgroup_mutex);
5397 
5398 	if (css->flags & CSS_DYING)
5399 		return;
5400 
5401 	css->flags |= CSS_DYING;
5402 
5403 	/*
5404 	 * This must happen before css is disassociated with its cgroup.
5405 	 * See seq_css() for details.
5406 	 */
5407 	css_clear_dir(css);
5408 
5409 	/*
5410 	 * Killing would put the base ref, but we need to keep it alive
5411 	 * until after ->css_offline().
5412 	 */
5413 	css_get(css);
5414 
5415 	/*
5416 	 * cgroup core guarantees that, by the time ->css_offline() is
5417 	 * invoked, no new css reference will be given out via
5418 	 * css_tryget_online().  We can't simply call percpu_ref_kill() and
5419 	 * proceed to offlining css's because percpu_ref_kill() doesn't
5420 	 * guarantee that the ref is seen as killed on all CPUs on return.
5421 	 *
5422 	 * Use percpu_ref_kill_and_confirm() to get notifications as each
5423 	 * css is confirmed to be seen as killed on all CPUs.
5424 	 */
5425 	percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5426 }
5427 
5428 /**
5429  * cgroup_destroy_locked - the first stage of cgroup destruction
5430  * @cgrp: cgroup to be destroyed
5431  *
5432  * css's make use of percpu refcnts whose killing latency shouldn't be
5433  * exposed to userland and are RCU protected.  Also, cgroup core needs to
5434  * guarantee that css_tryget_online() won't succeed by the time
5435  * ->css_offline() is invoked.  To satisfy all the requirements,
5436  * destruction is implemented in the following two steps.
5437  *
5438  * s1. Verify @cgrp can be destroyed and mark it dying.  Remove all
5439  *     userland visible parts and start killing the percpu refcnts of
5440  *     css's.  Set up so that the next stage will be kicked off once all
5441  *     the percpu refcnts are confirmed to be killed.
5442  *
5443  * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5444  *     rest of destruction.  Once all cgroup references are gone, the
5445  *     cgroup is RCU-freed.
5446  *
5447  * This function implements s1.  After this step, @cgrp is gone as far as
5448  * the userland is concerned and a new cgroup with the same name may be
5449  * created.  As cgroup doesn't care about the names internally, this
5450  * doesn't cause any problem.
5451  */
5452 static int cgroup_destroy_locked(struct cgroup *cgrp)
5453 	__releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5454 {
5455 	struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5456 	struct cgroup_subsys_state *css;
5457 	struct cgrp_cset_link *link;
5458 	int ssid;
5459 
5460 	lockdep_assert_held(&cgroup_mutex);
5461 
5462 	/*
5463 	 * Only migration can raise populated from zero and we're already
5464 	 * holding cgroup_mutex.
5465 	 */
5466 	if (cgroup_is_populated(cgrp))
5467 		return -EBUSY;
5468 
5469 	/*
5470 	 * Make sure there's no live children.  We can't test emptiness of
5471 	 * ->self.children as dead children linger on it while being
5472 	 * drained; otherwise, "rmdir parent/child parent" may fail.
5473 	 */
5474 	if (css_has_online_children(&cgrp->self))
5475 		return -EBUSY;
5476 
5477 	/*
5478 	 * Mark @cgrp and the associated csets dead.  The former prevents
5479 	 * further task migration and child creation by disabling
5480 	 * cgroup_lock_live_group().  The latter makes the csets ignored by
5481 	 * the migration path.
5482 	 */
5483 	cgrp->self.flags &= ~CSS_ONLINE;
5484 
5485 	spin_lock_irq(&css_set_lock);
5486 	list_for_each_entry(link, &cgrp->cset_links, cset_link)
5487 		link->cset->dead = true;
5488 	spin_unlock_irq(&css_set_lock);
5489 
5490 	/* initiate massacre of all css's */
5491 	for_each_css(css, ssid, cgrp)
5492 		kill_css(css);
5493 
5494 	/* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5495 	css_clear_dir(&cgrp->self);
5496 	kernfs_remove(cgrp->kn);
5497 
5498 	if (parent && cgroup_is_threaded(cgrp))
5499 		parent->nr_threaded_children--;
5500 
5501 	spin_lock_irq(&css_set_lock);
5502 	for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5503 		tcgrp->nr_descendants--;
5504 		tcgrp->nr_dying_descendants++;
5505 		/*
5506 		 * If the dying cgroup is frozen, decrease frozen descendants
5507 		 * counters of ancestor cgroups.
5508 		 */
5509 		if (test_bit(CGRP_FROZEN, &cgrp->flags))
5510 			tcgrp->freezer.nr_frozen_descendants--;
5511 	}
5512 	spin_unlock_irq(&css_set_lock);
5513 
5514 	cgroup1_check_for_release(parent);
5515 
5516 	cgroup_bpf_offline(cgrp);
5517 
5518 	/* put the base reference */
5519 	percpu_ref_kill(&cgrp->self.refcnt);
5520 
5521 	return 0;
5522 };
5523 
5524 int cgroup_rmdir(struct kernfs_node *kn)
5525 {
5526 	struct cgroup *cgrp;
5527 	int ret = 0;
5528 
5529 	cgrp = cgroup_kn_lock_live(kn, false);
5530 	if (!cgrp)
5531 		return 0;
5532 
5533 	ret = cgroup_destroy_locked(cgrp);
5534 	if (!ret)
5535 		TRACE_CGROUP_PATH(rmdir, cgrp);
5536 
5537 	cgroup_kn_unlock(kn);
5538 	return ret;
5539 }
5540 
5541 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5542 	.show_options		= cgroup_show_options,
5543 	.mkdir			= cgroup_mkdir,
5544 	.rmdir			= cgroup_rmdir,
5545 	.show_path		= cgroup_show_path,
5546 };
5547 
5548 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5549 {
5550 	struct cgroup_subsys_state *css;
5551 
5552 	pr_debug("Initializing cgroup subsys %s\n", ss->name);
5553 
5554 	mutex_lock(&cgroup_mutex);
5555 
5556 	idr_init(&ss->css_idr);
5557 	INIT_LIST_HEAD(&ss->cfts);
5558 
5559 	/* Create the root cgroup state for this subsystem */
5560 	ss->root = &cgrp_dfl_root;
5561 	css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5562 	/* We don't handle early failures gracefully */
5563 	BUG_ON(IS_ERR(css));
5564 	init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5565 
5566 	/*
5567 	 * Root csses are never destroyed and we can't initialize
5568 	 * percpu_ref during early init.  Disable refcnting.
5569 	 */
5570 	css->flags |= CSS_NO_REF;
5571 
5572 	if (early) {
5573 		/* allocation can't be done safely during early init */
5574 		css->id = 1;
5575 	} else {
5576 		css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5577 		BUG_ON(css->id < 0);
5578 	}
5579 
5580 	/* Update the init_css_set to contain a subsys
5581 	 * pointer to this state - since the subsystem is
5582 	 * newly registered, all tasks and hence the
5583 	 * init_css_set is in the subsystem's root cgroup. */
5584 	init_css_set.subsys[ss->id] = css;
5585 
5586 	have_fork_callback |= (bool)ss->fork << ss->id;
5587 	have_exit_callback |= (bool)ss->exit << ss->id;
5588 	have_release_callback |= (bool)ss->release << ss->id;
5589 	have_canfork_callback |= (bool)ss->can_fork << ss->id;
5590 
5591 	/* At system boot, before all subsystems have been
5592 	 * registered, no tasks have been forked, so we don't
5593 	 * need to invoke fork callbacks here. */
5594 	BUG_ON(!list_empty(&init_task.tasks));
5595 
5596 	BUG_ON(online_css(css));
5597 
5598 	mutex_unlock(&cgroup_mutex);
5599 }
5600 
5601 /**
5602  * cgroup_init_early - cgroup initialization at system boot
5603  *
5604  * Initialize cgroups at system boot, and initialize any
5605  * subsystems that request early init.
5606  */
5607 int __init cgroup_init_early(void)
5608 {
5609 	static struct cgroup_fs_context __initdata ctx;
5610 	struct cgroup_subsys *ss;
5611 	int i;
5612 
5613 	ctx.root = &cgrp_dfl_root;
5614 	init_cgroup_root(&ctx);
5615 	cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5616 
5617 	RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5618 
5619 	for_each_subsys(ss, i) {
5620 		WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5621 		     "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5622 		     i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5623 		     ss->id, ss->name);
5624 		WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5625 		     "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5626 
5627 		ss->id = i;
5628 		ss->name = cgroup_subsys_name[i];
5629 		if (!ss->legacy_name)
5630 			ss->legacy_name = cgroup_subsys_name[i];
5631 
5632 		if (ss->early_init)
5633 			cgroup_init_subsys(ss, true);
5634 	}
5635 	return 0;
5636 }
5637 
5638 /**
5639  * cgroup_init - cgroup initialization
5640  *
5641  * Register cgroup filesystem and /proc file, and initialize
5642  * any subsystems that didn't request early init.
5643  */
5644 int __init cgroup_init(void)
5645 {
5646 	struct cgroup_subsys *ss;
5647 	int ssid;
5648 
5649 	BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5650 	BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5651 	BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5652 
5653 	cgroup_rstat_boot();
5654 
5655 	/*
5656 	 * The latency of the synchronize_rcu() is too high for cgroups,
5657 	 * avoid it at the cost of forcing all readers into the slow path.
5658 	 */
5659 	rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5660 
5661 	get_user_ns(init_cgroup_ns.user_ns);
5662 
5663 	mutex_lock(&cgroup_mutex);
5664 
5665 	/*
5666 	 * Add init_css_set to the hash table so that dfl_root can link to
5667 	 * it during init.
5668 	 */
5669 	hash_add(css_set_table, &init_css_set.hlist,
5670 		 css_set_hash(init_css_set.subsys));
5671 
5672 	BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5673 
5674 	mutex_unlock(&cgroup_mutex);
5675 
5676 	for_each_subsys(ss, ssid) {
5677 		if (ss->early_init) {
5678 			struct cgroup_subsys_state *css =
5679 				init_css_set.subsys[ss->id];
5680 
5681 			css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5682 						   GFP_KERNEL);
5683 			BUG_ON(css->id < 0);
5684 		} else {
5685 			cgroup_init_subsys(ss, false);
5686 		}
5687 
5688 		list_add_tail(&init_css_set.e_cset_node[ssid],
5689 			      &cgrp_dfl_root.cgrp.e_csets[ssid]);
5690 
5691 		/*
5692 		 * Setting dfl_root subsys_mask needs to consider the
5693 		 * disabled flag and cftype registration needs kmalloc,
5694 		 * both of which aren't available during early_init.
5695 		 */
5696 		if (!cgroup_ssid_enabled(ssid))
5697 			continue;
5698 
5699 		if (cgroup1_ssid_disabled(ssid))
5700 			printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5701 			       ss->name);
5702 
5703 		cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5704 
5705 		/* implicit controllers must be threaded too */
5706 		WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5707 
5708 		if (ss->implicit_on_dfl)
5709 			cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5710 		else if (!ss->dfl_cftypes)
5711 			cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5712 
5713 		if (ss->threaded)
5714 			cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5715 
5716 		if (ss->dfl_cftypes == ss->legacy_cftypes) {
5717 			WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5718 		} else {
5719 			WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5720 			WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5721 		}
5722 
5723 		if (ss->bind)
5724 			ss->bind(init_css_set.subsys[ssid]);
5725 
5726 		mutex_lock(&cgroup_mutex);
5727 		css_populate_dir(init_css_set.subsys[ssid]);
5728 		mutex_unlock(&cgroup_mutex);
5729 	}
5730 
5731 	/* init_css_set.subsys[] has been updated, re-hash */
5732 	hash_del(&init_css_set.hlist);
5733 	hash_add(css_set_table, &init_css_set.hlist,
5734 		 css_set_hash(init_css_set.subsys));
5735 
5736 	WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5737 	WARN_ON(register_filesystem(&cgroup_fs_type));
5738 	WARN_ON(register_filesystem(&cgroup2_fs_type));
5739 	WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5740 #ifdef CONFIG_CPUSETS
5741 	WARN_ON(register_filesystem(&cpuset_fs_type));
5742 #endif
5743 
5744 	return 0;
5745 }
5746 
5747 static int __init cgroup_wq_init(void)
5748 {
5749 	/*
5750 	 * There isn't much point in executing destruction path in
5751 	 * parallel.  Good chunk is serialized with cgroup_mutex anyway.
5752 	 * Use 1 for @max_active.
5753 	 *
5754 	 * We would prefer to do this in cgroup_init() above, but that
5755 	 * is called before init_workqueues(): so leave this until after.
5756 	 */
5757 	cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5758 	BUG_ON(!cgroup_destroy_wq);
5759 	return 0;
5760 }
5761 core_initcall(cgroup_wq_init);
5762 
5763 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
5764 {
5765 	struct kernfs_node *kn;
5766 
5767 	kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5768 	if (!kn)
5769 		return;
5770 	kernfs_path(kn, buf, buflen);
5771 	kernfs_put(kn);
5772 }
5773 
5774 /*
5775  * proc_cgroup_show()
5776  *  - Print task's cgroup paths into seq_file, one line for each hierarchy
5777  *  - Used for /proc/<pid>/cgroup.
5778  */
5779 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5780 		     struct pid *pid, struct task_struct *tsk)
5781 {
5782 	char *buf;
5783 	int retval;
5784 	struct cgroup_root *root;
5785 
5786 	retval = -ENOMEM;
5787 	buf = kmalloc(PATH_MAX, GFP_KERNEL);
5788 	if (!buf)
5789 		goto out;
5790 
5791 	mutex_lock(&cgroup_mutex);
5792 	spin_lock_irq(&css_set_lock);
5793 
5794 	for_each_root(root) {
5795 		struct cgroup_subsys *ss;
5796 		struct cgroup *cgrp;
5797 		int ssid, count = 0;
5798 
5799 		if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5800 			continue;
5801 
5802 		seq_printf(m, "%d:", root->hierarchy_id);
5803 		if (root != &cgrp_dfl_root)
5804 			for_each_subsys(ss, ssid)
5805 				if (root->subsys_mask & (1 << ssid))
5806 					seq_printf(m, "%s%s", count++ ? "," : "",
5807 						   ss->legacy_name);
5808 		if (strlen(root->name))
5809 			seq_printf(m, "%sname=%s", count ? "," : "",
5810 				   root->name);
5811 		seq_putc(m, ':');
5812 
5813 		cgrp = task_cgroup_from_root(tsk, root);
5814 
5815 		/*
5816 		 * On traditional hierarchies, all zombie tasks show up as
5817 		 * belonging to the root cgroup.  On the default hierarchy,
5818 		 * while a zombie doesn't show up in "cgroup.procs" and
5819 		 * thus can't be migrated, its /proc/PID/cgroup keeps
5820 		 * reporting the cgroup it belonged to before exiting.  If
5821 		 * the cgroup is removed before the zombie is reaped,
5822 		 * " (deleted)" is appended to the cgroup path.
5823 		 */
5824 		if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5825 			retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5826 						current->nsproxy->cgroup_ns);
5827 			if (retval >= PATH_MAX)
5828 				retval = -ENAMETOOLONG;
5829 			if (retval < 0)
5830 				goto out_unlock;
5831 
5832 			seq_puts(m, buf);
5833 		} else {
5834 			seq_puts(m, "/");
5835 		}
5836 
5837 		if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5838 			seq_puts(m, " (deleted)\n");
5839 		else
5840 			seq_putc(m, '\n');
5841 	}
5842 
5843 	retval = 0;
5844 out_unlock:
5845 	spin_unlock_irq(&css_set_lock);
5846 	mutex_unlock(&cgroup_mutex);
5847 	kfree(buf);
5848 out:
5849 	return retval;
5850 }
5851 
5852 /**
5853  * cgroup_fork - initialize cgroup related fields during copy_process()
5854  * @child: pointer to task_struct of forking parent process.
5855  *
5856  * A task is associated with the init_css_set until cgroup_post_fork()
5857  * attaches it to the target css_set.
5858  */
5859 void cgroup_fork(struct task_struct *child)
5860 {
5861 	RCU_INIT_POINTER(child->cgroups, &init_css_set);
5862 	INIT_LIST_HEAD(&child->cg_list);
5863 }
5864 
5865 static struct cgroup *cgroup_get_from_file(struct file *f)
5866 {
5867 	struct cgroup_subsys_state *css;
5868 	struct cgroup *cgrp;
5869 
5870 	css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
5871 	if (IS_ERR(css))
5872 		return ERR_CAST(css);
5873 
5874 	cgrp = css->cgroup;
5875 	if (!cgroup_on_dfl(cgrp)) {
5876 		cgroup_put(cgrp);
5877 		return ERR_PTR(-EBADF);
5878 	}
5879 
5880 	return cgrp;
5881 }
5882 
5883 /**
5884  * cgroup_css_set_fork - find or create a css_set for a child process
5885  * @kargs: the arguments passed to create the child process
5886  *
5887  * This functions finds or creates a new css_set which the child
5888  * process will be attached to in cgroup_post_fork(). By default,
5889  * the child process will be given the same css_set as its parent.
5890  *
5891  * If CLONE_INTO_CGROUP is specified this function will try to find an
5892  * existing css_set which includes the requested cgroup and if not create
5893  * a new css_set that the child will be attached to later. If this function
5894  * succeeds it will hold cgroup_threadgroup_rwsem on return. If
5895  * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
5896  * before grabbing cgroup_threadgroup_rwsem and will hold a reference
5897  * to the target cgroup.
5898  */
5899 static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
5900 	__acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
5901 {
5902 	int ret;
5903 	struct cgroup *dst_cgrp = NULL;
5904 	struct css_set *cset;
5905 	struct super_block *sb;
5906 	struct file *f;
5907 
5908 	if (kargs->flags & CLONE_INTO_CGROUP)
5909 		mutex_lock(&cgroup_mutex);
5910 
5911 	cgroup_threadgroup_change_begin(current);
5912 
5913 	spin_lock_irq(&css_set_lock);
5914 	cset = task_css_set(current);
5915 	get_css_set(cset);
5916 	spin_unlock_irq(&css_set_lock);
5917 
5918 	if (!(kargs->flags & CLONE_INTO_CGROUP)) {
5919 		kargs->cset = cset;
5920 		return 0;
5921 	}
5922 
5923 	f = fget_raw(kargs->cgroup);
5924 	if (!f) {
5925 		ret = -EBADF;
5926 		goto err;
5927 	}
5928 	sb = f->f_path.dentry->d_sb;
5929 
5930 	dst_cgrp = cgroup_get_from_file(f);
5931 	if (IS_ERR(dst_cgrp)) {
5932 		ret = PTR_ERR(dst_cgrp);
5933 		dst_cgrp = NULL;
5934 		goto err;
5935 	}
5936 
5937 	if (cgroup_is_dead(dst_cgrp)) {
5938 		ret = -ENODEV;
5939 		goto err;
5940 	}
5941 
5942 	/*
5943 	 * Verify that we the target cgroup is writable for us. This is
5944 	 * usually done by the vfs layer but since we're not going through
5945 	 * the vfs layer here we need to do it "manually".
5946 	 */
5947 	ret = cgroup_may_write(dst_cgrp, sb);
5948 	if (ret)
5949 		goto err;
5950 
5951 	ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
5952 					!(kargs->flags & CLONE_THREAD));
5953 	if (ret)
5954 		goto err;
5955 
5956 	kargs->cset = find_css_set(cset, dst_cgrp);
5957 	if (!kargs->cset) {
5958 		ret = -ENOMEM;
5959 		goto err;
5960 	}
5961 
5962 	put_css_set(cset);
5963 	fput(f);
5964 	kargs->cgrp = dst_cgrp;
5965 	return ret;
5966 
5967 err:
5968 	cgroup_threadgroup_change_end(current);
5969 	mutex_unlock(&cgroup_mutex);
5970 	if (f)
5971 		fput(f);
5972 	if (dst_cgrp)
5973 		cgroup_put(dst_cgrp);
5974 	put_css_set(cset);
5975 	if (kargs->cset)
5976 		put_css_set(kargs->cset);
5977 	return ret;
5978 }
5979 
5980 /**
5981  * cgroup_css_set_put_fork - drop references we took during fork
5982  * @kargs: the arguments passed to create the child process
5983  *
5984  * Drop references to the prepared css_set and target cgroup if
5985  * CLONE_INTO_CGROUP was requested.
5986  */
5987 static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
5988 	__releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
5989 {
5990 	cgroup_threadgroup_change_end(current);
5991 
5992 	if (kargs->flags & CLONE_INTO_CGROUP) {
5993 		struct cgroup *cgrp = kargs->cgrp;
5994 		struct css_set *cset = kargs->cset;
5995 
5996 		mutex_unlock(&cgroup_mutex);
5997 
5998 		if (cset) {
5999 			put_css_set(cset);
6000 			kargs->cset = NULL;
6001 		}
6002 
6003 		if (cgrp) {
6004 			cgroup_put(cgrp);
6005 			kargs->cgrp = NULL;
6006 		}
6007 	}
6008 }
6009 
6010 /**
6011  * cgroup_can_fork - called on a new task before the process is exposed
6012  * @child: the child process
6013  *
6014  * This prepares a new css_set for the child process which the child will
6015  * be attached to in cgroup_post_fork().
6016  * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6017  * callback returns an error, the fork aborts with that error code. This
6018  * allows for a cgroup subsystem to conditionally allow or deny new forks.
6019  */
6020 int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6021 {
6022 	struct cgroup_subsys *ss;
6023 	int i, j, ret;
6024 
6025 	ret = cgroup_css_set_fork(kargs);
6026 	if (ret)
6027 		return ret;
6028 
6029 	do_each_subsys_mask(ss, i, have_canfork_callback) {
6030 		ret = ss->can_fork(child, kargs->cset);
6031 		if (ret)
6032 			goto out_revert;
6033 	} while_each_subsys_mask();
6034 
6035 	return 0;
6036 
6037 out_revert:
6038 	for_each_subsys(ss, j) {
6039 		if (j >= i)
6040 			break;
6041 		if (ss->cancel_fork)
6042 			ss->cancel_fork(child, kargs->cset);
6043 	}
6044 
6045 	cgroup_css_set_put_fork(kargs);
6046 
6047 	return ret;
6048 }
6049 
6050 /**
6051  * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6052  * @child: the child process
6053  * @kargs: the arguments passed to create the child process
6054  *
6055  * This calls the cancel_fork() callbacks if a fork failed *after*
6056  * cgroup_can_fork() succeeded and cleans up references we took to
6057  * prepare a new css_set for the child process in cgroup_can_fork().
6058  */
6059 void cgroup_cancel_fork(struct task_struct *child,
6060 			struct kernel_clone_args *kargs)
6061 {
6062 	struct cgroup_subsys *ss;
6063 	int i;
6064 
6065 	for_each_subsys(ss, i)
6066 		if (ss->cancel_fork)
6067 			ss->cancel_fork(child, kargs->cset);
6068 
6069 	cgroup_css_set_put_fork(kargs);
6070 }
6071 
6072 /**
6073  * cgroup_post_fork - finalize cgroup setup for the child process
6074  * @child: the child process
6075  *
6076  * Attach the child process to its css_set calling the subsystem fork()
6077  * callbacks.
6078  */
6079 void cgroup_post_fork(struct task_struct *child,
6080 		      struct kernel_clone_args *kargs)
6081 	__releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6082 {
6083 	struct cgroup_subsys *ss;
6084 	struct css_set *cset;
6085 	int i;
6086 
6087 	cset = kargs->cset;
6088 	kargs->cset = NULL;
6089 
6090 	spin_lock_irq(&css_set_lock);
6091 
6092 	/* init tasks are special, only link regular threads */
6093 	if (likely(child->pid)) {
6094 		WARN_ON_ONCE(!list_empty(&child->cg_list));
6095 		cset->nr_tasks++;
6096 		css_set_move_task(child, NULL, cset, false);
6097 	} else {
6098 		put_css_set(cset);
6099 		cset = NULL;
6100 	}
6101 
6102 	/*
6103 	 * If the cgroup has to be frozen, the new task has too.  Let's set
6104 	 * the JOBCTL_TRAP_FREEZE jobctl bit to get the task into the
6105 	 * frozen state.
6106 	 */
6107 	if (unlikely(cgroup_task_freeze(child))) {
6108 		spin_lock(&child->sighand->siglock);
6109 		WARN_ON_ONCE(child->frozen);
6110 		child->jobctl |= JOBCTL_TRAP_FREEZE;
6111 		spin_unlock(&child->sighand->siglock);
6112 
6113 		/*
6114 		 * Calling cgroup_update_frozen() isn't required here,
6115 		 * because it will be called anyway a bit later from
6116 		 * do_freezer_trap(). So we avoid cgroup's transient switch
6117 		 * from the frozen state and back.
6118 		 */
6119 	}
6120 
6121 	spin_unlock_irq(&css_set_lock);
6122 
6123 	/*
6124 	 * Call ss->fork().  This must happen after @child is linked on
6125 	 * css_set; otherwise, @child might change state between ->fork()
6126 	 * and addition to css_set.
6127 	 */
6128 	do_each_subsys_mask(ss, i, have_fork_callback) {
6129 		ss->fork(child);
6130 	} while_each_subsys_mask();
6131 
6132 	/* Make the new cset the root_cset of the new cgroup namespace. */
6133 	if (kargs->flags & CLONE_NEWCGROUP) {
6134 		struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6135 
6136 		get_css_set(cset);
6137 		child->nsproxy->cgroup_ns->root_cset = cset;
6138 		put_css_set(rcset);
6139 	}
6140 
6141 	cgroup_css_set_put_fork(kargs);
6142 }
6143 
6144 /**
6145  * cgroup_exit - detach cgroup from exiting task
6146  * @tsk: pointer to task_struct of exiting process
6147  *
6148  * Description: Detach cgroup from @tsk.
6149  *
6150  */
6151 void cgroup_exit(struct task_struct *tsk)
6152 {
6153 	struct cgroup_subsys *ss;
6154 	struct css_set *cset;
6155 	int i;
6156 
6157 	spin_lock_irq(&css_set_lock);
6158 
6159 	WARN_ON_ONCE(list_empty(&tsk->cg_list));
6160 	cset = task_css_set(tsk);
6161 	css_set_move_task(tsk, cset, NULL, false);
6162 	list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6163 	cset->nr_tasks--;
6164 
6165 	WARN_ON_ONCE(cgroup_task_frozen(tsk));
6166 	if (unlikely(cgroup_task_freeze(tsk)))
6167 		cgroup_update_frozen(task_dfl_cgroup(tsk));
6168 
6169 	spin_unlock_irq(&css_set_lock);
6170 
6171 	/* see cgroup_post_fork() for details */
6172 	do_each_subsys_mask(ss, i, have_exit_callback) {
6173 		ss->exit(tsk);
6174 	} while_each_subsys_mask();
6175 }
6176 
6177 void cgroup_release(struct task_struct *task)
6178 {
6179 	struct cgroup_subsys *ss;
6180 	int ssid;
6181 
6182 	do_each_subsys_mask(ss, ssid, have_release_callback) {
6183 		ss->release(task);
6184 	} while_each_subsys_mask();
6185 
6186 	spin_lock_irq(&css_set_lock);
6187 	css_set_skip_task_iters(task_css_set(task), task);
6188 	list_del_init(&task->cg_list);
6189 	spin_unlock_irq(&css_set_lock);
6190 }
6191 
6192 void cgroup_free(struct task_struct *task)
6193 {
6194 	struct css_set *cset = task_css_set(task);
6195 	put_css_set(cset);
6196 }
6197 
6198 static int __init cgroup_disable(char *str)
6199 {
6200 	struct cgroup_subsys *ss;
6201 	char *token;
6202 	int i;
6203 
6204 	while ((token = strsep(&str, ",")) != NULL) {
6205 		if (!*token)
6206 			continue;
6207 
6208 		for_each_subsys(ss, i) {
6209 			if (strcmp(token, ss->name) &&
6210 			    strcmp(token, ss->legacy_name))
6211 				continue;
6212 
6213 			static_branch_disable(cgroup_subsys_enabled_key[i]);
6214 			pr_info("Disabling %s control group subsystem\n",
6215 				ss->name);
6216 		}
6217 	}
6218 	return 1;
6219 }
6220 __setup("cgroup_disable=", cgroup_disable);
6221 
6222 void __init __weak enable_debug_cgroup(void) { }
6223 
6224 static int __init enable_cgroup_debug(char *str)
6225 {
6226 	cgroup_debug = true;
6227 	enable_debug_cgroup();
6228 	return 1;
6229 }
6230 __setup("cgroup_debug", enable_cgroup_debug);
6231 
6232 /**
6233  * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6234  * @dentry: directory dentry of interest
6235  * @ss: subsystem of interest
6236  *
6237  * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6238  * to get the corresponding css and return it.  If such css doesn't exist
6239  * or can't be pinned, an ERR_PTR value is returned.
6240  */
6241 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6242 						       struct cgroup_subsys *ss)
6243 {
6244 	struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6245 	struct file_system_type *s_type = dentry->d_sb->s_type;
6246 	struct cgroup_subsys_state *css = NULL;
6247 	struct cgroup *cgrp;
6248 
6249 	/* is @dentry a cgroup dir? */
6250 	if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6251 	    !kn || kernfs_type(kn) != KERNFS_DIR)
6252 		return ERR_PTR(-EBADF);
6253 
6254 	rcu_read_lock();
6255 
6256 	/*
6257 	 * This path doesn't originate from kernfs and @kn could already
6258 	 * have been or be removed at any point.  @kn->priv is RCU
6259 	 * protected for this access.  See css_release_work_fn() for details.
6260 	 */
6261 	cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6262 	if (cgrp)
6263 		css = cgroup_css(cgrp, ss);
6264 
6265 	if (!css || !css_tryget_online(css))
6266 		css = ERR_PTR(-ENOENT);
6267 
6268 	rcu_read_unlock();
6269 	return css;
6270 }
6271 
6272 /**
6273  * css_from_id - lookup css by id
6274  * @id: the cgroup id
6275  * @ss: cgroup subsys to be looked into
6276  *
6277  * Returns the css if there's valid one with @id, otherwise returns NULL.
6278  * Should be called under rcu_read_lock().
6279  */
6280 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6281 {
6282 	WARN_ON_ONCE(!rcu_read_lock_held());
6283 	return idr_find(&ss->css_idr, id);
6284 }
6285 
6286 /**
6287  * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6288  * @path: path on the default hierarchy
6289  *
6290  * Find the cgroup at @path on the default hierarchy, increment its
6291  * reference count and return it.  Returns pointer to the found cgroup on
6292  * success, ERR_PTR(-ENOENT) if @path doesn't exist and ERR_PTR(-ENOTDIR)
6293  * if @path points to a non-directory.
6294  */
6295 struct cgroup *cgroup_get_from_path(const char *path)
6296 {
6297 	struct kernfs_node *kn;
6298 	struct cgroup *cgrp;
6299 
6300 	mutex_lock(&cgroup_mutex);
6301 
6302 	kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6303 	if (kn) {
6304 		if (kernfs_type(kn) == KERNFS_DIR) {
6305 			cgrp = kn->priv;
6306 			cgroup_get_live(cgrp);
6307 		} else {
6308 			cgrp = ERR_PTR(-ENOTDIR);
6309 		}
6310 		kernfs_put(kn);
6311 	} else {
6312 		cgrp = ERR_PTR(-ENOENT);
6313 	}
6314 
6315 	mutex_unlock(&cgroup_mutex);
6316 	return cgrp;
6317 }
6318 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6319 
6320 /**
6321  * cgroup_get_from_fd - get a cgroup pointer from a fd
6322  * @fd: fd obtained by open(cgroup2_dir)
6323  *
6324  * Find the cgroup from a fd which should be obtained
6325  * by opening a cgroup directory.  Returns a pointer to the
6326  * cgroup on success. ERR_PTR is returned if the cgroup
6327  * cannot be found.
6328  */
6329 struct cgroup *cgroup_get_from_fd(int fd)
6330 {
6331 	struct cgroup *cgrp;
6332 	struct file *f;
6333 
6334 	f = fget_raw(fd);
6335 	if (!f)
6336 		return ERR_PTR(-EBADF);
6337 
6338 	cgrp = cgroup_get_from_file(f);
6339 	fput(f);
6340 	return cgrp;
6341 }
6342 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6343 
6344 static u64 power_of_ten(int power)
6345 {
6346 	u64 v = 1;
6347 	while (power--)
6348 		v *= 10;
6349 	return v;
6350 }
6351 
6352 /**
6353  * cgroup_parse_float - parse a floating number
6354  * @input: input string
6355  * @dec_shift: number of decimal digits to shift
6356  * @v: output
6357  *
6358  * Parse a decimal floating point number in @input and store the result in
6359  * @v with decimal point right shifted @dec_shift times.  For example, if
6360  * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6361  * Returns 0 on success, -errno otherwise.
6362  *
6363  * There's nothing cgroup specific about this function except that it's
6364  * currently the only user.
6365  */
6366 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6367 {
6368 	s64 whole, frac = 0;
6369 	int fstart = 0, fend = 0, flen;
6370 
6371 	if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6372 		return -EINVAL;
6373 	if (frac < 0)
6374 		return -EINVAL;
6375 
6376 	flen = fend > fstart ? fend - fstart : 0;
6377 	if (flen < dec_shift)
6378 		frac *= power_of_ten(dec_shift - flen);
6379 	else
6380 		frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
6381 
6382 	*v = whole * power_of_ten(dec_shift) + frac;
6383 	return 0;
6384 }
6385 
6386 /*
6387  * sock->sk_cgrp_data handling.  For more info, see sock_cgroup_data
6388  * definition in cgroup-defs.h.
6389  */
6390 #ifdef CONFIG_SOCK_CGROUP_DATA
6391 
6392 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6393 
6394 DEFINE_SPINLOCK(cgroup_sk_update_lock);
6395 static bool cgroup_sk_alloc_disabled __read_mostly;
6396 
6397 void cgroup_sk_alloc_disable(void)
6398 {
6399 	if (cgroup_sk_alloc_disabled)
6400 		return;
6401 	pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6402 	cgroup_sk_alloc_disabled = true;
6403 }
6404 
6405 #else
6406 
6407 #define cgroup_sk_alloc_disabled	false
6408 
6409 #endif
6410 
6411 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6412 {
6413 	if (cgroup_sk_alloc_disabled) {
6414 		skcd->no_refcnt = 1;
6415 		return;
6416 	}
6417 
6418 	/* Don't associate the sock with unrelated interrupted task's cgroup. */
6419 	if (in_interrupt())
6420 		return;
6421 
6422 	rcu_read_lock();
6423 
6424 	while (true) {
6425 		struct css_set *cset;
6426 
6427 		cset = task_css_set(current);
6428 		if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6429 			skcd->val = (unsigned long)cset->dfl_cgrp;
6430 			cgroup_bpf_get(cset->dfl_cgrp);
6431 			break;
6432 		}
6433 		cpu_relax();
6434 	}
6435 
6436 	rcu_read_unlock();
6437 }
6438 
6439 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6440 {
6441 	if (skcd->val) {
6442 		if (skcd->no_refcnt)
6443 			return;
6444 		/*
6445 		 * We might be cloning a socket which is left in an empty
6446 		 * cgroup and the cgroup might have already been rmdir'd.
6447 		 * Don't use cgroup_get_live().
6448 		 */
6449 		cgroup_get(sock_cgroup_ptr(skcd));
6450 		cgroup_bpf_get(sock_cgroup_ptr(skcd));
6451 	}
6452 }
6453 
6454 void cgroup_sk_free(struct sock_cgroup_data *skcd)
6455 {
6456 	struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6457 
6458 	if (skcd->no_refcnt)
6459 		return;
6460 	cgroup_bpf_put(cgrp);
6461 	cgroup_put(cgrp);
6462 }
6463 
6464 #endif	/* CONFIG_SOCK_CGROUP_DATA */
6465 
6466 #ifdef CONFIG_CGROUP_BPF
6467 int cgroup_bpf_attach(struct cgroup *cgrp,
6468 		      struct bpf_prog *prog, struct bpf_prog *replace_prog,
6469 		      struct bpf_cgroup_link *link,
6470 		      enum bpf_attach_type type,
6471 		      u32 flags)
6472 {
6473 	int ret;
6474 
6475 	mutex_lock(&cgroup_mutex);
6476 	ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
6477 	mutex_unlock(&cgroup_mutex);
6478 	return ret;
6479 }
6480 
6481 int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
6482 		      enum bpf_attach_type type)
6483 {
6484 	int ret;
6485 
6486 	mutex_lock(&cgroup_mutex);
6487 	ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
6488 	mutex_unlock(&cgroup_mutex);
6489 	return ret;
6490 }
6491 
6492 int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
6493 		     union bpf_attr __user *uattr)
6494 {
6495 	int ret;
6496 
6497 	mutex_lock(&cgroup_mutex);
6498 	ret = __cgroup_bpf_query(cgrp, attr, uattr);
6499 	mutex_unlock(&cgroup_mutex);
6500 	return ret;
6501 }
6502 #endif /* CONFIG_CGROUP_BPF */
6503 
6504 #ifdef CONFIG_SYSFS
6505 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6506 				      ssize_t size, const char *prefix)
6507 {
6508 	struct cftype *cft;
6509 	ssize_t ret = 0;
6510 
6511 	for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6512 		if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6513 			continue;
6514 
6515 		if (prefix)
6516 			ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6517 
6518 		ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6519 
6520 		if (WARN_ON(ret >= size))
6521 			break;
6522 	}
6523 
6524 	return ret;
6525 }
6526 
6527 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6528 			      char *buf)
6529 {
6530 	struct cgroup_subsys *ss;
6531 	int ssid;
6532 	ssize_t ret = 0;
6533 
6534 	ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6535 				     NULL);
6536 
6537 	for_each_subsys(ss, ssid)
6538 		ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6539 					      PAGE_SIZE - ret,
6540 					      cgroup_subsys_name[ssid]);
6541 
6542 	return ret;
6543 }
6544 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6545 
6546 static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6547 			     char *buf)
6548 {
6549 	return snprintf(buf, PAGE_SIZE,
6550 			"nsdelegate\n"
6551 			"memory_localevents\n"
6552 			"memory_recursiveprot\n");
6553 }
6554 static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6555 
6556 static struct attribute *cgroup_sysfs_attrs[] = {
6557 	&cgroup_delegate_attr.attr,
6558 	&cgroup_features_attr.attr,
6559 	NULL,
6560 };
6561 
6562 static const struct attribute_group cgroup_sysfs_attr_group = {
6563 	.attrs = cgroup_sysfs_attrs,
6564 	.name = "cgroup",
6565 };
6566 
6567 static int __init cgroup_sysfs_init(void)
6568 {
6569 	return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6570 }
6571 subsys_initcall(cgroup_sysfs_init);
6572 
6573 #endif /* CONFIG_SYSFS */
6574