xref: /openbmc/linux/include/linux/cgroup-defs.h (revision f3c60ab6)
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 	/* Control group has to be killed. */
76 	CGRP_KILL,
77 };
78 
79 /* cgroup_root->flags */
80 enum {
81 	CGRP_ROOT_NOPREFIX	= (1 << 1), /* mounted subsystems have no named prefix */
82 	CGRP_ROOT_XATTR		= (1 << 2), /* supports extended attributes */
83 
84 	/*
85 	 * Consider namespaces as delegation boundaries.  If this flag is
86 	 * set, controller specific interface files in a namespace root
87 	 * aren't writeable from inside the namespace.
88 	 */
89 	CGRP_ROOT_NS_DELEGATE	= (1 << 3),
90 
91 	/*
92 	 * Reduce latencies on dynamic cgroup modifications such as task
93 	 * migrations and controller on/offs by disabling percpu operation on
94 	 * cgroup_threadgroup_rwsem. This makes hot path operations such as
95 	 * forks and exits into the slow path and more expensive.
96 	 *
97 	 * The static usage pattern of creating a cgroup, enabling controllers,
98 	 * and then seeding it with CLONE_INTO_CGROUP doesn't require write
99 	 * locking cgroup_threadgroup_rwsem and thus doesn't benefit from
100 	 * favordynmod.
101 	 */
102 	CGRP_ROOT_FAVOR_DYNMODS = (1 << 4),
103 
104 	/*
105 	 * Enable cpuset controller in v1 cgroup to use v2 behavior.
106 	 */
107 	CGRP_ROOT_CPUSET_V2_MODE = (1 << 16),
108 
109 	/*
110 	 * Enable legacy local memory.events.
111 	 */
112 	CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 17),
113 
114 	/*
115 	 * Enable recursive subtree protection
116 	 */
117 	CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 18),
118 };
119 
120 /* cftype->flags */
121 enum {
122 	CFTYPE_ONLY_ON_ROOT	= (1 << 0),	/* only create on root cgrp */
123 	CFTYPE_NOT_ON_ROOT	= (1 << 1),	/* don't create on root cgrp */
124 	CFTYPE_NS_DELEGATABLE	= (1 << 2),	/* writeable beyond delegation boundaries */
125 
126 	CFTYPE_NO_PREFIX	= (1 << 3),	/* (DON'T USE FOR NEW FILES) no subsys prefix */
127 	CFTYPE_WORLD_WRITABLE	= (1 << 4),	/* (DON'T USE FOR NEW FILES) S_IWUGO */
128 	CFTYPE_DEBUG		= (1 << 5),	/* create when cgroup_debug */
129 
130 	/* internal flags, do not use outside cgroup core proper */
131 	__CFTYPE_ONLY_ON_DFL	= (1 << 16),	/* only on default hierarchy */
132 	__CFTYPE_NOT_ON_DFL	= (1 << 17),	/* not on default hierarchy */
133 	__CFTYPE_ADDED		= (1 << 18),
134 };
135 
136 /*
137  * cgroup_file is the handle for a file instance created in a cgroup which
138  * is used, for example, to generate file changed notifications.  This can
139  * be obtained by setting cftype->file_offset.
140  */
141 struct cgroup_file {
142 	/* do not access any fields from outside cgroup core */
143 	struct kernfs_node *kn;
144 	unsigned long notified_at;
145 	struct timer_list notify_timer;
146 };
147 
148 /*
149  * Per-subsystem/per-cgroup state maintained by the system.  This is the
150  * fundamental structural building block that controllers deal with.
151  *
152  * Fields marked with "PI:" are public and immutable and may be accessed
153  * directly without synchronization.
154  */
155 struct cgroup_subsys_state {
156 	/* PI: the cgroup that this css is attached to */
157 	struct cgroup *cgroup;
158 
159 	/* PI: the cgroup subsystem that this css is attached to */
160 	struct cgroup_subsys *ss;
161 
162 	/* reference count - access via css_[try]get() and css_put() */
163 	struct percpu_ref refcnt;
164 
165 	/* siblings list anchored at the parent's ->children */
166 	struct list_head sibling;
167 	struct list_head children;
168 
169 	/* flush target list anchored at cgrp->rstat_css_list */
170 	struct list_head rstat_css_node;
171 
172 	/*
173 	 * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
174 	 * matching css can be looked up using css_from_id().
175 	 */
176 	int id;
177 
178 	unsigned int flags;
179 
180 	/*
181 	 * Monotonically increasing unique serial number which defines a
182 	 * uniform order among all csses.  It's guaranteed that all
183 	 * ->children lists are in the ascending order of ->serial_nr and
184 	 * used to allow interrupting and resuming iterations.
185 	 */
186 	u64 serial_nr;
187 
188 	/*
189 	 * Incremented by online self and children.  Used to guarantee that
190 	 * parents are not offlined before their children.
191 	 */
192 	atomic_t online_cnt;
193 
194 	/* percpu_ref killing and RCU release */
195 	struct work_struct destroy_work;
196 	struct rcu_work destroy_rwork;
197 
198 	/*
199 	 * PI: the parent css.	Placed here for cache proximity to following
200 	 * fields of the containing structure.
201 	 */
202 	struct cgroup_subsys_state *parent;
203 };
204 
205 /*
206  * A css_set is a structure holding pointers to a set of
207  * cgroup_subsys_state objects. This saves space in the task struct
208  * object and speeds up fork()/exit(), since a single inc/dec and a
209  * list_add()/del() can bump the reference count on the entire cgroup
210  * set for a task.
211  */
212 struct css_set {
213 	/*
214 	 * Set of subsystem states, one for each subsystem. This array is
215 	 * immutable after creation apart from the init_css_set during
216 	 * subsystem registration (at boot time).
217 	 */
218 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
219 
220 	/* reference count */
221 	refcount_t refcount;
222 
223 	/*
224 	 * For a domain cgroup, the following points to self.  If threaded,
225 	 * to the matching cset of the nearest domain ancestor.  The
226 	 * dom_cset provides access to the domain cgroup and its csses to
227 	 * which domain level resource consumptions should be charged.
228 	 */
229 	struct css_set *dom_cset;
230 
231 	/* the default cgroup associated with this css_set */
232 	struct cgroup *dfl_cgrp;
233 
234 	/* internal task count, protected by css_set_lock */
235 	int nr_tasks;
236 
237 	/*
238 	 * Lists running through all tasks using this cgroup group.
239 	 * mg_tasks lists tasks which belong to this cset but are in the
240 	 * process of being migrated out or in.  Protected by
241 	 * css_set_lock, but, during migration, once tasks are moved to
242 	 * mg_tasks, it can be read safely while holding cgroup_mutex.
243 	 */
244 	struct list_head tasks;
245 	struct list_head mg_tasks;
246 	struct list_head dying_tasks;
247 
248 	/* all css_task_iters currently walking this cset */
249 	struct list_head task_iters;
250 
251 	/*
252 	 * On the default hierarchy, ->subsys[ssid] may point to a css
253 	 * attached to an ancestor instead of the cgroup this css_set is
254 	 * associated with.  The following node is anchored at
255 	 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
256 	 * iterate through all css's attached to a given cgroup.
257 	 */
258 	struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
259 
260 	/* all threaded csets whose ->dom_cset points to this cset */
261 	struct list_head threaded_csets;
262 	struct list_head threaded_csets_node;
263 
264 	/*
265 	 * List running through all cgroup groups in the same hash
266 	 * slot. Protected by css_set_lock
267 	 */
268 	struct hlist_node hlist;
269 
270 	/*
271 	 * List of cgrp_cset_links pointing at cgroups referenced from this
272 	 * css_set.  Protected by css_set_lock.
273 	 */
274 	struct list_head cgrp_links;
275 
276 	/*
277 	 * List of csets participating in the on-going migration either as
278 	 * source or destination.  Protected by cgroup_mutex.
279 	 */
280 	struct list_head mg_src_preload_node;
281 	struct list_head mg_dst_preload_node;
282 	struct list_head mg_node;
283 
284 	/*
285 	 * If this cset is acting as the source of migration the following
286 	 * two fields are set.  mg_src_cgrp and mg_dst_cgrp are
287 	 * respectively the source and destination cgroups of the on-going
288 	 * migration.  mg_dst_cset is the destination cset the target tasks
289 	 * on this cset should be migrated to.  Protected by cgroup_mutex.
290 	 */
291 	struct cgroup *mg_src_cgrp;
292 	struct cgroup *mg_dst_cgrp;
293 	struct css_set *mg_dst_cset;
294 
295 	/* dead and being drained, ignore for migration */
296 	bool dead;
297 
298 	/* For RCU-protected deletion */
299 	struct rcu_head rcu_head;
300 };
301 
302 struct cgroup_base_stat {
303 	struct task_cputime cputime;
304 
305 #ifdef CONFIG_SCHED_CORE
306 	u64 forceidle_sum;
307 #endif
308 };
309 
310 /*
311  * rstat - cgroup scalable recursive statistics.  Accounting is done
312  * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
313  * hierarchy on reads.
314  *
315  * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
316  * linked into the updated tree.  On the following read, propagation only
317  * considers and consumes the updated tree.  This makes reading O(the
318  * number of descendants which have been active since last read) instead of
319  * O(the total number of descendants).
320  *
321  * This is important because there can be a lot of (draining) cgroups which
322  * aren't active and stat may be read frequently.  The combination can
323  * become very expensive.  By propagating selectively, increasing reading
324  * frequency decreases the cost of each read.
325  *
326  * This struct hosts both the fields which implement the above -
327  * updated_children and updated_next - and the fields which track basic
328  * resource statistics on top of it - bsync, bstat and last_bstat.
329  */
330 struct cgroup_rstat_cpu {
331 	/*
332 	 * ->bsync protects ->bstat.  These are the only fields which get
333 	 * updated in the hot path.
334 	 */
335 	struct u64_stats_sync bsync;
336 	struct cgroup_base_stat bstat;
337 
338 	/*
339 	 * Snapshots at the last reading.  These are used to calculate the
340 	 * deltas to propagate to the global counters.
341 	 */
342 	struct cgroup_base_stat last_bstat;
343 
344 	/*
345 	 * This field is used to record the cumulative per-cpu time of
346 	 * the cgroup and its descendants. Currently it can be read via
347 	 * eBPF/drgn etc, and we are still trying to determine how to
348 	 * expose it in the cgroupfs interface.
349 	 */
350 	struct cgroup_base_stat subtree_bstat;
351 
352 	/*
353 	 * Snapshots at the last reading. These are used to calculate the
354 	 * deltas to propagate to the per-cpu subtree_bstat.
355 	 */
356 	struct cgroup_base_stat last_subtree_bstat;
357 
358 	/*
359 	 * Child cgroups with stat updates on this cpu since the last read
360 	 * are linked on the parent's ->updated_children through
361 	 * ->updated_next.
362 	 *
363 	 * In addition to being more compact, singly-linked list pointing
364 	 * to the cgroup makes it unnecessary for each per-cpu struct to
365 	 * point back to the associated cgroup.
366 	 *
367 	 * Protected by per-cpu cgroup_rstat_cpu_lock.
368 	 */
369 	struct cgroup *updated_children;	/* terminated by self cgroup */
370 	struct cgroup *updated_next;		/* NULL iff not on the list */
371 };
372 
373 struct cgroup_freezer_state {
374 	/* Should the cgroup and its descendants be frozen. */
375 	bool freeze;
376 
377 	/* Should the cgroup actually be frozen? */
378 	int e_freeze;
379 
380 	/* Fields below are protected by css_set_lock */
381 
382 	/* Number of frozen descendant cgroups */
383 	int nr_frozen_descendants;
384 
385 	/*
386 	 * Number of tasks, which are counted as frozen:
387 	 * frozen, SIGSTOPped, and PTRACEd.
388 	 */
389 	int nr_frozen_tasks;
390 };
391 
392 struct cgroup {
393 	/* self css with NULL ->ss, points back to this cgroup */
394 	struct cgroup_subsys_state self;
395 
396 	unsigned long flags;		/* "unsigned long" so bitops work */
397 
398 	/*
399 	 * The depth this cgroup is at.  The root is at depth zero and each
400 	 * step down the hierarchy increments the level.  This along with
401 	 * ancestors[] can determine whether a given cgroup is a
402 	 * descendant of another without traversing the hierarchy.
403 	 */
404 	int level;
405 
406 	/* Maximum allowed descent tree depth */
407 	int max_depth;
408 
409 	/*
410 	 * Keep track of total numbers of visible and dying descent cgroups.
411 	 * Dying cgroups are cgroups which were deleted by a user,
412 	 * but are still existing because someone else is holding a reference.
413 	 * max_descendants is a maximum allowed number of descent cgroups.
414 	 *
415 	 * nr_descendants and nr_dying_descendants are protected
416 	 * by cgroup_mutex and css_set_lock. It's fine to read them holding
417 	 * any of cgroup_mutex and css_set_lock; for writing both locks
418 	 * should be held.
419 	 */
420 	int nr_descendants;
421 	int nr_dying_descendants;
422 	int max_descendants;
423 
424 	/*
425 	 * Each non-empty css_set associated with this cgroup contributes
426 	 * one to nr_populated_csets.  The counter is zero iff this cgroup
427 	 * doesn't have any tasks.
428 	 *
429 	 * All children which have non-zero nr_populated_csets and/or
430 	 * nr_populated_children of their own contribute one to either
431 	 * nr_populated_domain_children or nr_populated_threaded_children
432 	 * depending on their type.  Each counter is zero iff all cgroups
433 	 * of the type in the subtree proper don't have any tasks.
434 	 */
435 	int nr_populated_csets;
436 	int nr_populated_domain_children;
437 	int nr_populated_threaded_children;
438 
439 	int nr_threaded_children;	/* # of live threaded child cgroups */
440 
441 	struct kernfs_node *kn;		/* cgroup kernfs entry */
442 	struct cgroup_file procs_file;	/* handle for "cgroup.procs" */
443 	struct cgroup_file events_file;	/* handle for "cgroup.events" */
444 
445 	/* handles for "{cpu,memory,io,irq}.pressure" */
446 	struct cgroup_file psi_files[NR_PSI_RESOURCES];
447 
448 	/*
449 	 * The bitmask of subsystems enabled on the child cgroups.
450 	 * ->subtree_control is the one configured through
451 	 * "cgroup.subtree_control" while ->subtree_ss_mask is the effective
452 	 * one which may have more subsystems enabled.  Controller knobs
453 	 * are made available iff it's enabled in ->subtree_control.
454 	 */
455 	u16 subtree_control;
456 	u16 subtree_ss_mask;
457 	u16 old_subtree_control;
458 	u16 old_subtree_ss_mask;
459 
460 	/* Private pointers for each registered subsystem */
461 	struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
462 
463 	struct cgroup_root *root;
464 
465 	/*
466 	 * List of cgrp_cset_links pointing at css_sets with tasks in this
467 	 * cgroup.  Protected by css_set_lock.
468 	 */
469 	struct list_head cset_links;
470 
471 	/*
472 	 * On the default hierarchy, a css_set for a cgroup with some
473 	 * susbsys disabled will point to css's which are associated with
474 	 * the closest ancestor which has the subsys enabled.  The
475 	 * following lists all css_sets which point to this cgroup's css
476 	 * for the given subsystem.
477 	 */
478 	struct list_head e_csets[CGROUP_SUBSYS_COUNT];
479 
480 	/*
481 	 * If !threaded, self.  If threaded, it points to the nearest
482 	 * domain ancestor.  Inside a threaded subtree, cgroups are exempt
483 	 * from process granularity and no-internal-task constraint.
484 	 * Domain level resource consumptions which aren't tied to a
485 	 * specific task are charged to the dom_cgrp.
486 	 */
487 	struct cgroup *dom_cgrp;
488 	struct cgroup *old_dom_cgrp;		/* used while enabling threaded */
489 
490 	/* per-cpu recursive resource statistics */
491 	struct cgroup_rstat_cpu __percpu *rstat_cpu;
492 	struct list_head rstat_css_list;
493 
494 	/* cgroup basic resource statistics */
495 	struct cgroup_base_stat last_bstat;
496 	struct cgroup_base_stat bstat;
497 	struct prev_cputime prev_cputime;	/* for printing out cputime */
498 
499 	/*
500 	 * list of pidlists, up to two for each namespace (one for procs, one
501 	 * for tasks); created on demand.
502 	 */
503 	struct list_head pidlists;
504 	struct mutex pidlist_mutex;
505 
506 	/* used to wait for offlining of csses */
507 	wait_queue_head_t offline_waitq;
508 
509 	/* used to schedule release agent */
510 	struct work_struct release_agent_work;
511 
512 	/* used to track pressure stalls */
513 	struct psi_group *psi;
514 
515 	/* used to store eBPF programs */
516 	struct cgroup_bpf bpf;
517 
518 	/* If there is block congestion on this cgroup. */
519 	atomic_t congestion_count;
520 
521 	/* Used to store internal freezer state */
522 	struct cgroup_freezer_state freezer;
523 
524 #ifdef CONFIG_BPF_SYSCALL
525 	struct bpf_local_storage __rcu  *bpf_cgrp_storage;
526 #endif
527 
528 	/* All ancestors including self */
529 	struct cgroup *ancestors[];
530 };
531 
532 /*
533  * A cgroup_root represents the root of a cgroup hierarchy, and may be
534  * associated with a kernfs_root to form an active hierarchy.  This is
535  * internal to cgroup core.  Don't access directly from controllers.
536  */
537 struct cgroup_root {
538 	struct kernfs_root *kf_root;
539 
540 	/* The bitmask of subsystems attached to this hierarchy */
541 	unsigned int subsys_mask;
542 
543 	/* Unique id for this hierarchy. */
544 	int hierarchy_id;
545 
546 	/* A list running through the active hierarchies */
547 	struct list_head root_list;
548 	struct rcu_head rcu;	/* Must be near the top */
549 
550 	/*
551 	 * The root cgroup. The containing cgroup_root will be destroyed on its
552 	 * release. cgrp->ancestors[0] will be used overflowing into the
553 	 * following field. cgrp_ancestor_storage must immediately follow.
554 	 */
555 	struct cgroup cgrp;
556 
557 	/* must follow cgrp for cgrp->ancestors[0], see above */
558 	struct cgroup *cgrp_ancestor_storage;
559 
560 	/* Number of cgroups in the hierarchy, used only for /proc/cgroups */
561 	atomic_t nr_cgrps;
562 
563 	/* Hierarchy-specific flags */
564 	unsigned int flags;
565 
566 	/* The path to use for release notifications. */
567 	char release_agent_path[PATH_MAX];
568 
569 	/* The name for this hierarchy - may be empty */
570 	char name[MAX_CGROUP_ROOT_NAMELEN];
571 };
572 
573 /*
574  * struct cftype: handler definitions for cgroup control files
575  *
576  * When reading/writing to a file:
577  *	- the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
578  *	- the 'cftype' of the file is file->f_path.dentry->d_fsdata
579  */
580 struct cftype {
581 	/*
582 	 * By convention, the name should begin with the name of the
583 	 * subsystem, followed by a period.  Zero length string indicates
584 	 * end of cftype array.
585 	 */
586 	char name[MAX_CFTYPE_NAME];
587 	unsigned long private;
588 
589 	/*
590 	 * The maximum length of string, excluding trailing nul, that can
591 	 * be passed to write.  If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
592 	 */
593 	size_t max_write_len;
594 
595 	/* CFTYPE_* flags */
596 	unsigned int flags;
597 
598 	/*
599 	 * If non-zero, should contain the offset from the start of css to
600 	 * a struct cgroup_file field.  cgroup will record the handle of
601 	 * the created file into it.  The recorded handle can be used as
602 	 * long as the containing css remains accessible.
603 	 */
604 	unsigned int file_offset;
605 
606 	/*
607 	 * Fields used for internal bookkeeping.  Initialized automatically
608 	 * during registration.
609 	 */
610 	struct cgroup_subsys *ss;	/* NULL for cgroup core files */
611 	struct list_head node;		/* anchored at ss->cfts */
612 	struct kernfs_ops *kf_ops;
613 
614 	int (*open)(struct kernfs_open_file *of);
615 	void (*release)(struct kernfs_open_file *of);
616 
617 	/*
618 	 * read_u64() is a shortcut for the common case of returning a
619 	 * single integer. Use it in place of read()
620 	 */
621 	u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
622 	/*
623 	 * read_s64() is a signed version of read_u64()
624 	 */
625 	s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
626 
627 	/* generic seq_file read interface */
628 	int (*seq_show)(struct seq_file *sf, void *v);
629 
630 	/* optional ops, implement all or none */
631 	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
632 	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
633 	void (*seq_stop)(struct seq_file *sf, void *v);
634 
635 	/*
636 	 * write_u64() is a shortcut for the common case of accepting
637 	 * a single integer (as parsed by simple_strtoull) from
638 	 * userspace. Use in place of write(); return 0 or error.
639 	 */
640 	int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
641 			 u64 val);
642 	/*
643 	 * write_s64() is a signed version of write_u64()
644 	 */
645 	int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
646 			 s64 val);
647 
648 	/*
649 	 * write() is the generic write callback which maps directly to
650 	 * kernfs write operation and overrides all other operations.
651 	 * Maximum write size is determined by ->max_write_len.  Use
652 	 * of_css/cft() to access the associated css and cft.
653 	 */
654 	ssize_t (*write)(struct kernfs_open_file *of,
655 			 char *buf, size_t nbytes, loff_t off);
656 
657 	__poll_t (*poll)(struct kernfs_open_file *of,
658 			 struct poll_table_struct *pt);
659 
660 #ifdef CONFIG_DEBUG_LOCK_ALLOC
661 	struct lock_class_key	lockdep_key;
662 #endif
663 };
664 
665 /*
666  * Control Group subsystem type.
667  * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
668  */
669 struct cgroup_subsys {
670 	struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
671 	int (*css_online)(struct cgroup_subsys_state *css);
672 	void (*css_offline)(struct cgroup_subsys_state *css);
673 	void (*css_released)(struct cgroup_subsys_state *css);
674 	void (*css_free)(struct cgroup_subsys_state *css);
675 	void (*css_reset)(struct cgroup_subsys_state *css);
676 	void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
677 	int (*css_extra_stat_show)(struct seq_file *seq,
678 				   struct cgroup_subsys_state *css);
679 	int (*css_local_stat_show)(struct seq_file *seq,
680 				   struct cgroup_subsys_state *css);
681 
682 	int (*can_attach)(struct cgroup_taskset *tset);
683 	void (*cancel_attach)(struct cgroup_taskset *tset);
684 	void (*attach)(struct cgroup_taskset *tset);
685 	void (*post_attach)(void);
686 	int (*can_fork)(struct task_struct *task,
687 			struct css_set *cset);
688 	void (*cancel_fork)(struct task_struct *task, struct css_set *cset);
689 	void (*fork)(struct task_struct *task);
690 	void (*exit)(struct task_struct *task);
691 	void (*release)(struct task_struct *task);
692 	void (*bind)(struct cgroup_subsys_state *root_css);
693 
694 	bool early_init:1;
695 
696 	/*
697 	 * If %true, the controller, on the default hierarchy, doesn't show
698 	 * up in "cgroup.controllers" or "cgroup.subtree_control", is
699 	 * implicitly enabled on all cgroups on the default hierarchy, and
700 	 * bypasses the "no internal process" constraint.  This is for
701 	 * utility type controllers which is transparent to userland.
702 	 *
703 	 * An implicit controller can be stolen from the default hierarchy
704 	 * anytime and thus must be okay with offline csses from previous
705 	 * hierarchies coexisting with csses for the current one.
706 	 */
707 	bool implicit_on_dfl:1;
708 
709 	/*
710 	 * If %true, the controller, supports threaded mode on the default
711 	 * hierarchy.  In a threaded subtree, both process granularity and
712 	 * no-internal-process constraint are ignored and a threaded
713 	 * controllers should be able to handle that.
714 	 *
715 	 * Note that as an implicit controller is automatically enabled on
716 	 * all cgroups on the default hierarchy, it should also be
717 	 * threaded.  implicit && !threaded is not supported.
718 	 */
719 	bool threaded:1;
720 
721 	/* the following two fields are initialized automatically during boot */
722 	int id;
723 	const char *name;
724 
725 	/* optional, initialized automatically during boot if not set */
726 	const char *legacy_name;
727 
728 	/* link to parent, protected by cgroup_lock() */
729 	struct cgroup_root *root;
730 
731 	/* idr for css->id */
732 	struct idr css_idr;
733 
734 	/*
735 	 * List of cftypes.  Each entry is the first entry of an array
736 	 * terminated by zero length name.
737 	 */
738 	struct list_head cfts;
739 
740 	/*
741 	 * Base cftypes which are automatically registered.  The two can
742 	 * point to the same array.
743 	 */
744 	struct cftype *dfl_cftypes;	/* for the default hierarchy */
745 	struct cftype *legacy_cftypes;	/* for the legacy hierarchies */
746 
747 	/*
748 	 * A subsystem may depend on other subsystems.  When such subsystem
749 	 * is enabled on a cgroup, the depended-upon subsystems are enabled
750 	 * together if available.  Subsystems enabled due to dependency are
751 	 * not visible to userland until explicitly enabled.  The following
752 	 * specifies the mask of subsystems that this one depends on.
753 	 */
754 	unsigned int depends_on;
755 };
756 
757 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
758 
759 /**
760  * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
761  * @tsk: target task
762  *
763  * Allows cgroup operations to synchronize against threadgroup changes
764  * using a percpu_rw_semaphore.
765  */
cgroup_threadgroup_change_begin(struct task_struct * tsk)766 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
767 {
768 	percpu_down_read(&cgroup_threadgroup_rwsem);
769 }
770 
771 /**
772  * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
773  * @tsk: target task
774  *
775  * Counterpart of cgroup_threadcgroup_change_begin().
776  */
cgroup_threadgroup_change_end(struct task_struct * tsk)777 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
778 {
779 	percpu_up_read(&cgroup_threadgroup_rwsem);
780 }
781 
782 #else	/* CONFIG_CGROUPS */
783 
784 #define CGROUP_SUBSYS_COUNT 0
785 
cgroup_threadgroup_change_begin(struct task_struct * tsk)786 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
787 {
788 	might_sleep();
789 }
790 
cgroup_threadgroup_change_end(struct task_struct * tsk)791 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
792 
793 #endif	/* CONFIG_CGROUPS */
794 
795 #ifdef CONFIG_SOCK_CGROUP_DATA
796 
797 /*
798  * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
799  * per-socket cgroup information except for memcg association.
800  *
801  * On legacy hierarchies, net_prio and net_cls controllers directly
802  * set attributes on each sock which can then be tested by the network
803  * layer. On the default hierarchy, each sock is associated with the
804  * cgroup it was created in and the networking layer can match the
805  * cgroup directly.
806  */
807 struct sock_cgroup_data {
808 	struct cgroup	*cgroup; /* v2 */
809 #ifdef CONFIG_CGROUP_NET_CLASSID
810 	u32		classid; /* v1 */
811 #endif
812 #ifdef CONFIG_CGROUP_NET_PRIO
813 	u16		prioidx; /* v1 */
814 #endif
815 };
816 
sock_cgroup_prioidx(const struct sock_cgroup_data * skcd)817 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
818 {
819 #ifdef CONFIG_CGROUP_NET_PRIO
820 	return READ_ONCE(skcd->prioidx);
821 #else
822 	return 1;
823 #endif
824 }
825 
sock_cgroup_classid(const struct sock_cgroup_data * skcd)826 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
827 {
828 #ifdef CONFIG_CGROUP_NET_CLASSID
829 	return READ_ONCE(skcd->classid);
830 #else
831 	return 0;
832 #endif
833 }
834 
sock_cgroup_set_prioidx(struct sock_cgroup_data * skcd,u16 prioidx)835 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
836 					   u16 prioidx)
837 {
838 #ifdef CONFIG_CGROUP_NET_PRIO
839 	WRITE_ONCE(skcd->prioidx, prioidx);
840 #endif
841 }
842 
sock_cgroup_set_classid(struct sock_cgroup_data * skcd,u32 classid)843 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
844 					   u32 classid)
845 {
846 #ifdef CONFIG_CGROUP_NET_CLASSID
847 	WRITE_ONCE(skcd->classid, classid);
848 #endif
849 }
850 
851 #else	/* CONFIG_SOCK_CGROUP_DATA */
852 
853 struct sock_cgroup_data {
854 };
855 
856 #endif	/* CONFIG_SOCK_CGROUP_DATA */
857 
858 #endif	/* _LINUX_CGROUP_DEFS_H */
859