xref: /linux/include/linux/cgroup-defs.h (revision 6aee5aed2edd0a156bf060abce1bdbbc38171c10)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * linux/cgroup-defs.h - basic definitions for cgroup
4  *
5  * This file provides basic type and interface.  Include this file directly
6  * only if necessary to avoid cyclic dependencies.
7  */
8 #ifndef _LINUX_CGROUP_DEFS_H
9 #define _LINUX_CGROUP_DEFS_H
10 
11 #include <linux/limits.h>
12 #include <linux/list.h>
13 #include <linux/idr.h>
14 #include <linux/wait.h>
15 #include <linux/mutex.h>
16 #include <linux/rcupdate.h>
17 #include <linux/refcount.h>
18 #include <linux/percpu-refcount.h>
19 #include <linux/percpu-rwsem.h>
20 #include <linux/u64_stats_sync.h>
21 #include <linux/workqueue.h>
22 #include <linux/bpf-cgroup-defs.h>
23 #include <linux/psi_types.h>
24 
25 #ifdef CONFIG_CGROUPS
26 
27 struct cgroup;
28 struct cgroup_root;
29 struct cgroup_subsys;
30 struct cgroup_taskset;
31 struct kernfs_node;
32 struct kernfs_ops;
33 struct kernfs_open_file;
34 struct seq_file;
35 struct poll_table_struct;
36 
37 #define MAX_CGROUP_TYPE_NAMELEN 32
38 #define MAX_CGROUP_ROOT_NAMELEN 64
39 #define MAX_CFTYPE_NAME		64
40 
41 /* define the enumeration of all cgroup subsystems */
42 #define SUBSYS(_x) _x ## _cgrp_id,
43 enum cgroup_subsys_id {
44 #include <linux/cgroup_subsys.h>
45 	CGROUP_SUBSYS_COUNT,
46 };
47 #undef SUBSYS
48 
49 /* bits in struct cgroup_subsys_state flags field */
50 enum {
51 	CSS_NO_REF	= (1 << 0), /* no reference counting for this css */
52 	CSS_ONLINE	= (1 << 1), /* between ->css_online() and ->css_offline() */
53 	CSS_RELEASED	= (1 << 2), /* refcnt reached zero, released */
54 	CSS_VISIBLE	= (1 << 3), /* css is visible to userland */
55 	CSS_DYING	= (1 << 4), /* css is dying */
56 };
57 
58 /* bits in struct cgroup flags field */
59 enum {
60 	/* Control Group requires release notifications to userspace */
61 	CGRP_NOTIFY_ON_RELEASE,
62 	/*
63 	 * Clone the parent's configuration when creating a new child
64 	 * cpuset cgroup.  For historical reasons, this option can be
65 	 * specified at mount time and thus is implemented here.
66 	 */
67 	CGRP_CPUSET_CLONE_CHILDREN,
68 
69 	/* Control group has to be frozen. */
70 	CGRP_FREEZE,
71 
72 	/* Cgroup is frozen. */
73 	CGRP_FROZEN,
74 };
75 
76 /* cgroup_root->flags */
77 enum {
78 	CGRP_ROOT_NOPREFIX	= (1 << 1), /* mounted subsystems have no named prefix */
79 	CGRP_ROOT_XATTR		= (1 << 2), /* supports extended attributes */
80 
81 	/*
82 	 * Consider namespaces as delegation boundaries.  If this flag is
83 	 * set, controller specific interface files in a namespace root
84 	 * aren't writeable from inside the namespace.
85 	 */
86 	CGRP_ROOT_NS_DELEGATE	= (1 << 3),
87 
88 	/*
89 	 * Reduce latencies on dynamic cgroup modifications such as task
90 	 * migrations and controller on/offs by disabling percpu operation on
91 	 * cgroup_threadgroup_rwsem. This makes hot path operations such as
92 	 * forks and exits into the slow path and more expensive.
93 	 *
94 	 * The static usage pattern of creating a cgroup, enabling controllers,
95 	 * and then seeding it with CLONE_INTO_CGROUP doesn't require write
96 	 * locking cgroup_threadgroup_rwsem and thus doesn't benefit from
97 	 * favordynmod.
98 	 */
99 	CGRP_ROOT_FAVOR_DYNMODS = (1 << 4),
100 
101 	/*
102 	 * Enable cpuset controller in v1 cgroup to use v2 behavior.
103 	 */
104 	CGRP_ROOT_CPUSET_V2_MODE = (1 << 16),
105 
106 	/*
107 	 * Enable legacy local memory.events.
108 	 */
109 	CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 17),
110 
111 	/*
112 	 * Enable recursive subtree protection
113 	 */
114 	CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 18),
115 
116 	/*
117 	 * Enable hugetlb accounting for the memory controller.
118 	 */
119 	CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING = (1 << 19),
120 
121 	/*
122 	 * Enable legacy local pids.events.
123 	 */
124 	CGRP_ROOT_PIDS_LOCAL_EVENTS = (1 << 20),
125 };
126 
127 /* cftype->flags */
128 enum {
129 	CFTYPE_ONLY_ON_ROOT	= (1 << 0),	/* only create on root cgrp */
130 	CFTYPE_NOT_ON_ROOT	= (1 << 1),	/* don't create on root cgrp */
131 	CFTYPE_NS_DELEGATABLE	= (1 << 2),	/* writeable beyond delegation boundaries */
132 
133 	CFTYPE_NO_PREFIX	= (1 << 3),	/* (DON'T USE FOR NEW FILES) no subsys prefix */
134 	CFTYPE_WORLD_WRITABLE	= (1 << 4),	/* (DON'T USE FOR NEW FILES) S_IWUGO */
135 	CFTYPE_DEBUG		= (1 << 5),	/* create when cgroup_debug */
136 
137 	/* internal flags, do not use outside cgroup core proper */
138 	__CFTYPE_ONLY_ON_DFL	= (1 << 16),	/* only on default hierarchy */
139 	__CFTYPE_NOT_ON_DFL	= (1 << 17),	/* not on default hierarchy */
140 	__CFTYPE_ADDED		= (1 << 18),
141 };
142 
143 /*
144  * cgroup_file is the handle for a file instance created in a cgroup which
145  * is used, for example, to generate file changed notifications.  This can
146  * be obtained by setting cftype->file_offset.
147  */
148 struct cgroup_file {
149 	/* do not access any fields from outside cgroup core */
150 	struct kernfs_node *kn;
151 	unsigned long notified_at;
152 	struct timer_list notify_timer;
153 };
154 
155 /*
156  * Per-subsystem/per-cgroup state maintained by the system.  This is the
157  * fundamental structural building block that controllers deal with.
158  *
159  * Fields marked with "PI:" are public and immutable and may be accessed
160  * directly without synchronization.
161  */
162 struct cgroup_subsys_state {
163 	/* PI: the cgroup that this css is attached to */
164 	struct cgroup *cgroup;
165 
166 	/* PI: the cgroup subsystem that this css is attached to */
167 	struct cgroup_subsys *ss;
168 
169 	/* reference count - access via css_[try]get() and css_put() */
170 	struct percpu_ref refcnt;
171 
172 	/*
173 	 * Depending on the context, this field is initialized
174 	 * via css_rstat_init() at different places:
175 	 *
176 	 * when css is associated with cgroup::self
177 	 *   when css->cgroup is the root cgroup
178 	 *     performed in cgroup_init()
179 	 *   when css->cgroup is not the root cgroup
180 	 *     performed in cgroup_create()
181 	 * when css is associated with a subsystem
182 	 *   when css->cgroup is the root cgroup
183 	 *     performed in cgroup_init_subsys() in the non-early path
184 	 *   when css->cgroup is not the root cgroup
185 	 *     performed in css_create()
186 	 */
187 	struct css_rstat_cpu __percpu *rstat_cpu;
188 
189 	/*
190 	 * siblings list anchored at the parent's ->children
191 	 *
192 	 * linkage is protected by cgroup_mutex or RCU
193 	 */
194 	struct list_head sibling;
195 	struct list_head children;
196 
197 	/*
198 	 * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
199 	 * matching css can be looked up using css_from_id().
200 	 */
201 	int id;
202 
203 	unsigned int flags;
204 
205 	/*
206 	 * Monotonically increasing unique serial number which defines a
207 	 * uniform order among all csses.  It's guaranteed that all
208 	 * ->children lists are in the ascending order of ->serial_nr and
209 	 * used to allow interrupting and resuming iterations.
210 	 */
211 	u64 serial_nr;
212 
213 	/*
214 	 * Incremented by online self and children.  Used to guarantee that
215 	 * parents are not offlined before their children.
216 	 */
217 	atomic_t online_cnt;
218 
219 	/* percpu_ref killing and RCU release */
220 	struct work_struct destroy_work;
221 	struct rcu_work destroy_rwork;
222 
223 	/*
224 	 * PI: the parent css.	Placed here for cache proximity to following
225 	 * fields of the containing structure.
226 	 */
227 	struct cgroup_subsys_state *parent;
228 
229 	/*
230 	 * Keep track of total numbers of visible descendant CSSes.
231 	 * The total number of dying CSSes is tracked in
232 	 * css->cgroup->nr_dying_subsys[ssid].
233 	 * Protected by cgroup_mutex.
234 	 */
235 	int nr_descendants;
236 
237 	/*
238 	 * A singly-linked list of css structures to be rstat flushed.
239 	 * This is a scratch field to be used exclusively by
240 	 * css_rstat_flush().
241 	 *
242 	 * Protected by rstat_base_lock when css is cgroup::self.
243 	 * Protected by css->ss->rstat_ss_lock otherwise.
244 	 */
245 	struct cgroup_subsys_state *rstat_flush_next;
246 };
247 
248 /*
249  * A css_set is a structure holding pointers to a set of
250  * cgroup_subsys_state objects. This saves space in the task struct
251  * object and speeds up fork()/exit(), since a single inc/dec and a
252  * list_add()/del() can bump the reference count on the entire cgroup
253  * set for a task.
254  */
255 struct css_set {
256 	/*
257 	 * Set of subsystem states, one for each subsystem. This array is
258 	 * immutable after creation apart from the init_css_set during
259 	 * subsystem registration (at boot time).
260 	 */
261 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
262 
263 	/* reference count */
264 	refcount_t refcount;
265 
266 	/*
267 	 * For a domain cgroup, the following points to self.  If threaded,
268 	 * to the matching cset of the nearest domain ancestor.  The
269 	 * dom_cset provides access to the domain cgroup and its csses to
270 	 * which domain level resource consumptions should be charged.
271 	 */
272 	struct css_set *dom_cset;
273 
274 	/* the default cgroup associated with this css_set */
275 	struct cgroup *dfl_cgrp;
276 
277 	/* internal task count, protected by css_set_lock */
278 	int nr_tasks;
279 
280 	/*
281 	 * Lists running through all tasks using this cgroup group.
282 	 * mg_tasks lists tasks which belong to this cset but are in the
283 	 * process of being migrated out or in.  Protected by
284 	 * css_set_lock, but, during migration, once tasks are moved to
285 	 * mg_tasks, it can be read safely while holding cgroup_mutex.
286 	 */
287 	struct list_head tasks;
288 	struct list_head mg_tasks;
289 	struct list_head dying_tasks;
290 
291 	/* all css_task_iters currently walking this cset */
292 	struct list_head task_iters;
293 
294 	/*
295 	 * On the default hierarchy, ->subsys[ssid] may point to a css
296 	 * attached to an ancestor instead of the cgroup this css_set is
297 	 * associated with.  The following node is anchored at
298 	 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
299 	 * iterate through all css's attached to a given cgroup.
300 	 */
301 	struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
302 
303 	/* all threaded csets whose ->dom_cset points to this cset */
304 	struct list_head threaded_csets;
305 	struct list_head threaded_csets_node;
306 
307 	/*
308 	 * List running through all cgroup groups in the same hash
309 	 * slot. Protected by css_set_lock
310 	 */
311 	struct hlist_node hlist;
312 
313 	/*
314 	 * List of cgrp_cset_links pointing at cgroups referenced from this
315 	 * css_set.  Protected by css_set_lock.
316 	 */
317 	struct list_head cgrp_links;
318 
319 	/*
320 	 * List of csets participating in the on-going migration either as
321 	 * source or destination.  Protected by cgroup_mutex.
322 	 */
323 	struct list_head mg_src_preload_node;
324 	struct list_head mg_dst_preload_node;
325 	struct list_head mg_node;
326 
327 	/*
328 	 * If this cset is acting as the source of migration the following
329 	 * two fields are set.  mg_src_cgrp and mg_dst_cgrp are
330 	 * respectively the source and destination cgroups of the on-going
331 	 * migration.  mg_dst_cset is the destination cset the target tasks
332 	 * on this cset should be migrated to.  Protected by cgroup_mutex.
333 	 */
334 	struct cgroup *mg_src_cgrp;
335 	struct cgroup *mg_dst_cgrp;
336 	struct css_set *mg_dst_cset;
337 
338 	/* dead and being drained, ignore for migration */
339 	bool dead;
340 
341 	/* For RCU-protected deletion */
342 	struct rcu_head rcu_head;
343 };
344 
345 struct cgroup_base_stat {
346 	struct task_cputime cputime;
347 
348 #ifdef CONFIG_SCHED_CORE
349 	u64 forceidle_sum;
350 #endif
351 	u64 ntime;
352 };
353 
354 /*
355  * rstat - cgroup scalable recursive statistics.  Accounting is done
356  * per-cpu in css_rstat_cpu which is then lazily propagated up the
357  * hierarchy on reads.
358  *
359  * When a stat gets updated, the css_rstat_cpu and its ancestors are
360  * linked into the updated tree.  On the following read, propagation only
361  * considers and consumes the updated tree.  This makes reading O(the
362  * number of descendants which have been active since last read) instead of
363  * O(the total number of descendants).
364  *
365  * This is important because there can be a lot of (draining) cgroups which
366  * aren't active and stat may be read frequently.  The combination can
367  * become very expensive.  By propagating selectively, increasing reading
368  * frequency decreases the cost of each read.
369  *
370  * This struct hosts both the fields which implement the above -
371  * updated_children and updated_next.
372  */
373 struct css_rstat_cpu {
374 	/*
375 	 * Child cgroups with stat updates on this cpu since the last read
376 	 * are linked on the parent's ->updated_children through
377 	 * ->updated_next. updated_children is terminated by its container css.
378 	 */
379 	struct cgroup_subsys_state *updated_children;
380 	struct cgroup_subsys_state *updated_next;	/* NULL if not on the list */
381 
382 	struct llist_node lnode;		/* lockless list for update */
383 	struct cgroup_subsys_state *owner;	/* back pointer */
384 };
385 
386 /*
387  * This struct hosts the fields which track basic resource statistics on
388  * top of it - bsync, bstat and last_bstat.
389  */
390 struct cgroup_rstat_base_cpu {
391 	/*
392 	 * ->bsync protects ->bstat.  These are the only fields which get
393 	 * updated in the hot path.
394 	 */
395 	struct u64_stats_sync bsync;
396 	struct cgroup_base_stat bstat;
397 
398 	/*
399 	 * Snapshots at the last reading.  These are used to calculate the
400 	 * deltas to propagate to the global counters.
401 	 */
402 	struct cgroup_base_stat last_bstat;
403 
404 	/*
405 	 * This field is used to record the cumulative per-cpu time of
406 	 * the cgroup and its descendants. Currently it can be read via
407 	 * eBPF/drgn etc, and we are still trying to determine how to
408 	 * expose it in the cgroupfs interface.
409 	 */
410 	struct cgroup_base_stat subtree_bstat;
411 
412 	/*
413 	 * Snapshots at the last reading. These are used to calculate the
414 	 * deltas to propagate to the per-cpu subtree_bstat.
415 	 */
416 	struct cgroup_base_stat last_subtree_bstat;
417 };
418 
419 struct cgroup_freezer_state {
420 	/* Should the cgroup and its descendants be frozen. */
421 	bool freeze;
422 
423 	/* Should the cgroup actually be frozen? */
424 	bool e_freeze;
425 
426 	/* Fields below are protected by css_set_lock */
427 
428 	/* Number of frozen descendant cgroups */
429 	int nr_frozen_descendants;
430 
431 	/*
432 	 * Number of tasks, which are counted as frozen:
433 	 * frozen, SIGSTOPped, and PTRACEd.
434 	 */
435 	int nr_frozen_tasks;
436 };
437 
438 struct cgroup {
439 	/* self css with NULL ->ss, points back to this cgroup */
440 	struct cgroup_subsys_state self;
441 
442 	unsigned long flags;		/* "unsigned long" so bitops work */
443 
444 	/*
445 	 * The depth this cgroup is at.  The root is at depth zero and each
446 	 * step down the hierarchy increments the level.  This along with
447 	 * ancestors[] can determine whether a given cgroup is a
448 	 * descendant of another without traversing the hierarchy.
449 	 */
450 	int level;
451 
452 	/* Maximum allowed descent tree depth */
453 	int max_depth;
454 
455 	/*
456 	 * Keep track of total numbers of visible and dying descent cgroups.
457 	 * Dying cgroups are cgroups which were deleted by a user,
458 	 * but are still existing because someone else is holding a reference.
459 	 * max_descendants is a maximum allowed number of descent cgroups.
460 	 *
461 	 * nr_descendants and nr_dying_descendants are protected
462 	 * by cgroup_mutex and css_set_lock. It's fine to read them holding
463 	 * any of cgroup_mutex and css_set_lock; for writing both locks
464 	 * should be held.
465 	 */
466 	int nr_descendants;
467 	int nr_dying_descendants;
468 	int max_descendants;
469 
470 	/*
471 	 * Each non-empty css_set associated with this cgroup contributes
472 	 * one to nr_populated_csets.  The counter is zero iff this cgroup
473 	 * doesn't have any tasks.
474 	 *
475 	 * All children which have non-zero nr_populated_csets and/or
476 	 * nr_populated_children of their own contribute one to either
477 	 * nr_populated_domain_children or nr_populated_threaded_children
478 	 * depending on their type.  Each counter is zero iff all cgroups
479 	 * of the type in the subtree proper don't have any tasks.
480 	 */
481 	int nr_populated_csets;
482 	int nr_populated_domain_children;
483 	int nr_populated_threaded_children;
484 
485 	int nr_threaded_children;	/* # of live threaded child cgroups */
486 
487 	/* sequence number for cgroup.kill, serialized by css_set_lock. */
488 	unsigned int kill_seq;
489 
490 	struct kernfs_node *kn;		/* cgroup kernfs entry */
491 	struct cgroup_file procs_file;	/* handle for "cgroup.procs" */
492 	struct cgroup_file events_file;	/* handle for "cgroup.events" */
493 
494 	/* handles for "{cpu,memory,io,irq}.pressure" */
495 	struct cgroup_file psi_files[NR_PSI_RESOURCES];
496 
497 	/*
498 	 * The bitmask of subsystems enabled on the child cgroups.
499 	 * ->subtree_control is the one configured through
500 	 * "cgroup.subtree_control" while ->subtree_ss_mask is the effective
501 	 * one which may have more subsystems enabled.  Controller knobs
502 	 * are made available iff it's enabled in ->subtree_control.
503 	 */
504 	u16 subtree_control;
505 	u16 subtree_ss_mask;
506 	u16 old_subtree_control;
507 	u16 old_subtree_ss_mask;
508 
509 	/* Private pointers for each registered subsystem */
510 	struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
511 
512 	/*
513 	 * Keep track of total number of dying CSSes at and below this cgroup.
514 	 * Protected by cgroup_mutex.
515 	 */
516 	int nr_dying_subsys[CGROUP_SUBSYS_COUNT];
517 
518 	struct cgroup_root *root;
519 
520 	/*
521 	 * List of cgrp_cset_links pointing at css_sets with tasks in this
522 	 * cgroup.  Protected by css_set_lock.
523 	 */
524 	struct list_head cset_links;
525 
526 	/*
527 	 * On the default hierarchy, a css_set for a cgroup with some
528 	 * susbsys disabled will point to css's which are associated with
529 	 * the closest ancestor which has the subsys enabled.  The
530 	 * following lists all css_sets which point to this cgroup's css
531 	 * for the given subsystem.
532 	 */
533 	struct list_head e_csets[CGROUP_SUBSYS_COUNT];
534 
535 	/*
536 	 * If !threaded, self.  If threaded, it points to the nearest
537 	 * domain ancestor.  Inside a threaded subtree, cgroups are exempt
538 	 * from process granularity and no-internal-task constraint.
539 	 * Domain level resource consumptions which aren't tied to a
540 	 * specific task are charged to the dom_cgrp.
541 	 */
542 	struct cgroup *dom_cgrp;
543 	struct cgroup *old_dom_cgrp;		/* used while enabling threaded */
544 
545 	/*
546 	 * Depending on the context, this field is initialized via
547 	 * css_rstat_init() at different places:
548 	 *
549 	 * when cgroup is the root cgroup
550 	 *   performed in cgroup_setup_root()
551 	 * otherwise
552 	 *   performed in cgroup_create()
553 	 */
554 	struct cgroup_rstat_base_cpu __percpu *rstat_base_cpu;
555 
556 	/*
557 	 * Add padding to keep the read mostly rstat per-cpu pointer on a
558 	 * different cacheline than the following *bstat fields which can have
559 	 * frequent updates.
560 	 */
561 	CACHELINE_PADDING(_pad_);
562 
563 	/* cgroup basic resource statistics */
564 	struct cgroup_base_stat last_bstat;
565 	struct cgroup_base_stat bstat;
566 	struct prev_cputime prev_cputime;	/* for printing out cputime */
567 
568 	/*
569 	 * list of pidlists, up to two for each namespace (one for procs, one
570 	 * for tasks); created on demand.
571 	 */
572 	struct list_head pidlists;
573 	struct mutex pidlist_mutex;
574 
575 	/* used to wait for offlining of csses */
576 	wait_queue_head_t offline_waitq;
577 
578 	/* used to schedule release agent */
579 	struct work_struct release_agent_work;
580 
581 	/* used to track pressure stalls */
582 	struct psi_group *psi;
583 
584 	/* used to store eBPF programs */
585 	struct cgroup_bpf bpf;
586 
587 	/* Used to store internal freezer state */
588 	struct cgroup_freezer_state freezer;
589 
590 #ifdef CONFIG_BPF_SYSCALL
591 	struct bpf_local_storage __rcu  *bpf_cgrp_storage;
592 #endif
593 
594 	/* All ancestors including self */
595 	struct cgroup *ancestors[];
596 };
597 
598 /*
599  * A cgroup_root represents the root of a cgroup hierarchy, and may be
600  * associated with a kernfs_root to form an active hierarchy.  This is
601  * internal to cgroup core.  Don't access directly from controllers.
602  */
603 struct cgroup_root {
604 	struct kernfs_root *kf_root;
605 
606 	/* The bitmask of subsystems attached to this hierarchy */
607 	unsigned int subsys_mask;
608 
609 	/* Unique id for this hierarchy. */
610 	int hierarchy_id;
611 
612 	/* A list running through the active hierarchies */
613 	struct list_head root_list;
614 	struct rcu_head rcu;	/* Must be near the top */
615 
616 	/*
617 	 * The root cgroup. The containing cgroup_root will be destroyed on its
618 	 * release. cgrp->ancestors[0] will be used overflowing into the
619 	 * following field. cgrp_ancestor_storage must immediately follow.
620 	 */
621 	struct cgroup cgrp;
622 
623 	/* must follow cgrp for cgrp->ancestors[0], see above */
624 	struct cgroup *cgrp_ancestor_storage;
625 
626 	/* Number of cgroups in the hierarchy, used only for /proc/cgroups */
627 	atomic_t nr_cgrps;
628 
629 	/* Hierarchy-specific flags */
630 	unsigned int flags;
631 
632 	/* The path to use for release notifications. */
633 	char release_agent_path[PATH_MAX];
634 
635 	/* The name for this hierarchy - may be empty */
636 	char name[MAX_CGROUP_ROOT_NAMELEN];
637 };
638 
639 /*
640  * struct cftype: handler definitions for cgroup control files
641  *
642  * When reading/writing to a file:
643  *	- the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
644  *	- the 'cftype' of the file is file->f_path.dentry->d_fsdata
645  */
646 struct cftype {
647 	/*
648 	 * Name of the subsystem is prepended in cgroup_file_name().
649 	 * Zero length string indicates end of cftype array.
650 	 */
651 	char name[MAX_CFTYPE_NAME];
652 	unsigned long private;
653 
654 	/*
655 	 * The maximum length of string, excluding trailing nul, that can
656 	 * be passed to write.  If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
657 	 */
658 	size_t max_write_len;
659 
660 	/* CFTYPE_* flags */
661 	unsigned int flags;
662 
663 	/*
664 	 * If non-zero, should contain the offset from the start of css to
665 	 * a struct cgroup_file field.  cgroup will record the handle of
666 	 * the created file into it.  The recorded handle can be used as
667 	 * long as the containing css remains accessible.
668 	 */
669 	unsigned int file_offset;
670 
671 	/*
672 	 * Fields used for internal bookkeeping.  Initialized automatically
673 	 * during registration.
674 	 */
675 	struct cgroup_subsys *ss;	/* NULL for cgroup core files */
676 	struct list_head node;		/* anchored at ss->cfts */
677 	struct kernfs_ops *kf_ops;
678 
679 	int (*open)(struct kernfs_open_file *of);
680 	void (*release)(struct kernfs_open_file *of);
681 
682 	/*
683 	 * read_u64() is a shortcut for the common case of returning a
684 	 * single integer. Use it in place of read()
685 	 */
686 	u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
687 	/*
688 	 * read_s64() is a signed version of read_u64()
689 	 */
690 	s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
691 
692 	/* generic seq_file read interface */
693 	int (*seq_show)(struct seq_file *sf, void *v);
694 
695 	/* optional ops, implement all or none */
696 	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
697 	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
698 	void (*seq_stop)(struct seq_file *sf, void *v);
699 
700 	/*
701 	 * write_u64() is a shortcut for the common case of accepting
702 	 * a single integer (as parsed by simple_strtoull) from
703 	 * userspace. Use in place of write(); return 0 or error.
704 	 */
705 	int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
706 			 u64 val);
707 	/*
708 	 * write_s64() is a signed version of write_u64()
709 	 */
710 	int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
711 			 s64 val);
712 
713 	/*
714 	 * write() is the generic write callback which maps directly to
715 	 * kernfs write operation and overrides all other operations.
716 	 * Maximum write size is determined by ->max_write_len.  Use
717 	 * of_css/cft() to access the associated css and cft.
718 	 */
719 	ssize_t (*write)(struct kernfs_open_file *of,
720 			 char *buf, size_t nbytes, loff_t off);
721 
722 	__poll_t (*poll)(struct kernfs_open_file *of,
723 			 struct poll_table_struct *pt);
724 
725 	struct lock_class_key	lockdep_key;
726 };
727 
728 /*
729  * Control Group subsystem type.
730  * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
731  */
732 struct cgroup_subsys {
733 	struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
734 	int (*css_online)(struct cgroup_subsys_state *css);
735 	void (*css_offline)(struct cgroup_subsys_state *css);
736 	void (*css_released)(struct cgroup_subsys_state *css);
737 	void (*css_free)(struct cgroup_subsys_state *css);
738 	void (*css_reset)(struct cgroup_subsys_state *css);
739 	void (*css_killed)(struct cgroup_subsys_state *css);
740 	void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
741 	int (*css_extra_stat_show)(struct seq_file *seq,
742 				   struct cgroup_subsys_state *css);
743 	int (*css_local_stat_show)(struct seq_file *seq,
744 				   struct cgroup_subsys_state *css);
745 
746 	int (*can_attach)(struct cgroup_taskset *tset);
747 	void (*cancel_attach)(struct cgroup_taskset *tset);
748 	void (*attach)(struct cgroup_taskset *tset);
749 	void (*post_attach)(void);
750 	int (*can_fork)(struct task_struct *task,
751 			struct css_set *cset);
752 	void (*cancel_fork)(struct task_struct *task, struct css_set *cset);
753 	void (*fork)(struct task_struct *task);
754 	void (*exit)(struct task_struct *task);
755 	void (*release)(struct task_struct *task);
756 	void (*bind)(struct cgroup_subsys_state *root_css);
757 
758 	bool early_init:1;
759 
760 	/*
761 	 * If %true, the controller, on the default hierarchy, doesn't show
762 	 * up in "cgroup.controllers" or "cgroup.subtree_control", is
763 	 * implicitly enabled on all cgroups on the default hierarchy, and
764 	 * bypasses the "no internal process" constraint.  This is for
765 	 * utility type controllers which is transparent to userland.
766 	 *
767 	 * An implicit controller can be stolen from the default hierarchy
768 	 * anytime and thus must be okay with offline csses from previous
769 	 * hierarchies coexisting with csses for the current one.
770 	 */
771 	bool implicit_on_dfl:1;
772 
773 	/*
774 	 * If %true, the controller, supports threaded mode on the default
775 	 * hierarchy.  In a threaded subtree, both process granularity and
776 	 * no-internal-process constraint are ignored and a threaded
777 	 * controllers should be able to handle that.
778 	 *
779 	 * Note that as an implicit controller is automatically enabled on
780 	 * all cgroups on the default hierarchy, it should also be
781 	 * threaded.  implicit && !threaded is not supported.
782 	 */
783 	bool threaded:1;
784 
785 	/* the following two fields are initialized automatically during boot */
786 	int id;
787 	const char *name;
788 
789 	/* optional, initialized automatically during boot if not set */
790 	const char *legacy_name;
791 
792 	/* link to parent, protected by cgroup_lock() */
793 	struct cgroup_root *root;
794 
795 	/* idr for css->id */
796 	struct idr css_idr;
797 
798 	/*
799 	 * List of cftypes.  Each entry is the first entry of an array
800 	 * terminated by zero length name.
801 	 */
802 	struct list_head cfts;
803 
804 	/*
805 	 * Base cftypes which are automatically registered.  The two can
806 	 * point to the same array.
807 	 */
808 	struct cftype *dfl_cftypes;	/* for the default hierarchy */
809 	struct cftype *legacy_cftypes;	/* for the legacy hierarchies */
810 
811 	/*
812 	 * A subsystem may depend on other subsystems.  When such subsystem
813 	 * is enabled on a cgroup, the depended-upon subsystems are enabled
814 	 * together if available.  Subsystems enabled due to dependency are
815 	 * not visible to userland until explicitly enabled.  The following
816 	 * specifies the mask of subsystems that this one depends on.
817 	 */
818 	unsigned int depends_on;
819 
820 	spinlock_t rstat_ss_lock;
821 	struct llist_head __percpu *lhead; /* lockless update list head */
822 };
823 
824 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
825 
826 struct cgroup_of_peak {
827 	unsigned long		value;
828 	struct list_head	list;
829 };
830 
831 /**
832  * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
833  * @tsk: target task
834  *
835  * Allows cgroup operations to synchronize against threadgroup changes
836  * using a percpu_rw_semaphore.
837  */
838 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
839 {
840 	percpu_down_read(&cgroup_threadgroup_rwsem);
841 }
842 
843 /**
844  * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
845  * @tsk: target task
846  *
847  * Counterpart of cgroup_threadcgroup_change_begin().
848  */
849 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
850 {
851 	percpu_up_read(&cgroup_threadgroup_rwsem);
852 }
853 
854 #else	/* CONFIG_CGROUPS */
855 
856 #define CGROUP_SUBSYS_COUNT 0
857 
858 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
859 {
860 	might_sleep();
861 }
862 
863 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
864 
865 #endif	/* CONFIG_CGROUPS */
866 
867 #ifdef CONFIG_SOCK_CGROUP_DATA
868 
869 /*
870  * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
871  * per-socket cgroup information except for memcg association.
872  *
873  * On legacy hierarchies, net_prio and net_cls controllers directly
874  * set attributes on each sock which can then be tested by the network
875  * layer. On the default hierarchy, each sock is associated with the
876  * cgroup it was created in and the networking layer can match the
877  * cgroup directly.
878  */
879 struct sock_cgroup_data {
880 	struct cgroup	*cgroup; /* v2 */
881 #ifdef CONFIG_CGROUP_NET_CLASSID
882 	u32		classid; /* v1 */
883 #endif
884 #ifdef CONFIG_CGROUP_NET_PRIO
885 	u16		prioidx; /* v1 */
886 #endif
887 };
888 
889 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
890 {
891 #ifdef CONFIG_CGROUP_NET_PRIO
892 	return READ_ONCE(skcd->prioidx);
893 #else
894 	return 1;
895 #endif
896 }
897 
898 #ifdef CONFIG_CGROUP_NET_CLASSID
899 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
900 {
901 	return READ_ONCE(skcd->classid);
902 }
903 #endif
904 
905 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
906 					   u16 prioidx)
907 {
908 #ifdef CONFIG_CGROUP_NET_PRIO
909 	WRITE_ONCE(skcd->prioidx, prioidx);
910 #endif
911 }
912 
913 #ifdef CONFIG_CGROUP_NET_CLASSID
914 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
915 					   u32 classid)
916 {
917 	WRITE_ONCE(skcd->classid, classid);
918 }
919 #endif
920 
921 #else	/* CONFIG_SOCK_CGROUP_DATA */
922 
923 struct sock_cgroup_data {
924 };
925 
926 #endif	/* CONFIG_SOCK_CGROUP_DATA */
927 
928 #endif	/* _LINUX_CGROUP_DEFS_H */
929