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