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.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 * Enable cpuset controller in v1 cgroup to use v2 behavior.
90 */
91 CGRP_ROOT_CPUSET_V2_MODE = (1 << 4),
92
93 /*
94 * Enable legacy local memory.events.
95 */
96 CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 5),
97
98 /*
99 * Enable recursive subtree protection
100 */
101 CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 6),
102 };
103
104 /* cftype->flags */
105 enum {
106 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
107 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
108 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
109
110 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
111 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
112 CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */
113
114 /* internal flags, do not use outside cgroup core proper */
115 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
116 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
117 };
118
119 /*
120 * cgroup_file is the handle for a file instance created in a cgroup which
121 * is used, for example, to generate file changed notifications. This can
122 * be obtained by setting cftype->file_offset.
123 */
124 struct cgroup_file {
125 /* do not access any fields from outside cgroup core */
126 struct kernfs_node *kn;
127 unsigned long notified_at;
128 struct timer_list notify_timer;
129 };
130
131 /*
132 * Per-subsystem/per-cgroup state maintained by the system. This is the
133 * fundamental structural building block that controllers deal with.
134 *
135 * Fields marked with "PI:" are public and immutable and may be accessed
136 * directly without synchronization.
137 */
138 struct cgroup_subsys_state {
139 /* PI: the cgroup that this css is attached to */
140 struct cgroup *cgroup;
141
142 /* PI: the cgroup subsystem that this css is attached to */
143 struct cgroup_subsys *ss;
144
145 /* reference count - access via css_[try]get() and css_put() */
146 struct percpu_ref refcnt;
147
148 /* siblings list anchored at the parent's ->children */
149 struct list_head sibling;
150 struct list_head children;
151
152 /* flush target list anchored at cgrp->rstat_css_list */
153 struct list_head rstat_css_node;
154
155 /*
156 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
157 * matching css can be looked up using css_from_id().
158 */
159 int id;
160
161 unsigned int flags;
162
163 /*
164 * Monotonically increasing unique serial number which defines a
165 * uniform order among all csses. It's guaranteed that all
166 * ->children lists are in the ascending order of ->serial_nr and
167 * used to allow interrupting and resuming iterations.
168 */
169 u64 serial_nr;
170
171 /*
172 * Incremented by online self and children. Used to guarantee that
173 * parents are not offlined before their children.
174 */
175 atomic_t online_cnt;
176
177 /* percpu_ref killing and RCU release */
178 struct work_struct destroy_work;
179 struct rcu_work destroy_rwork;
180
181 /*
182 * PI: the parent css. Placed here for cache proximity to following
183 * fields of the containing structure.
184 */
185 struct cgroup_subsys_state *parent;
186 };
187
188 /*
189 * A css_set is a structure holding pointers to a set of
190 * cgroup_subsys_state objects. This saves space in the task struct
191 * object and speeds up fork()/exit(), since a single inc/dec and a
192 * list_add()/del() can bump the reference count on the entire cgroup
193 * set for a task.
194 */
195 struct css_set {
196 /*
197 * Set of subsystem states, one for each subsystem. This array is
198 * immutable after creation apart from the init_css_set during
199 * subsystem registration (at boot time).
200 */
201 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
202
203 /* reference count */
204 refcount_t refcount;
205
206 /*
207 * For a domain cgroup, the following points to self. If threaded,
208 * to the matching cset of the nearest domain ancestor. The
209 * dom_cset provides access to the domain cgroup and its csses to
210 * which domain level resource consumptions should be charged.
211 */
212 struct css_set *dom_cset;
213
214 /* the default cgroup associated with this css_set */
215 struct cgroup *dfl_cgrp;
216
217 /* internal task count, protected by css_set_lock */
218 int nr_tasks;
219
220 /*
221 * Lists running through all tasks using this cgroup group.
222 * mg_tasks lists tasks which belong to this cset but are in the
223 * process of being migrated out or in. Protected by
224 * css_set_rwsem, but, during migration, once tasks are moved to
225 * mg_tasks, it can be read safely while holding cgroup_mutex.
226 */
227 struct list_head tasks;
228 struct list_head mg_tasks;
229 struct list_head dying_tasks;
230
231 /* all css_task_iters currently walking this cset */
232 struct list_head task_iters;
233
234 /*
235 * On the default hierarhcy, ->subsys[ssid] may point to a css
236 * attached to an ancestor instead of the cgroup this css_set is
237 * associated with. The following node is anchored at
238 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
239 * iterate through all css's attached to a given cgroup.
240 */
241 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
242
243 /* all threaded csets whose ->dom_cset points to this cset */
244 struct list_head threaded_csets;
245 struct list_head threaded_csets_node;
246
247 /*
248 * List running through all cgroup groups in the same hash
249 * slot. Protected by css_set_lock
250 */
251 struct hlist_node hlist;
252
253 /*
254 * List of cgrp_cset_links pointing at cgroups referenced from this
255 * css_set. Protected by css_set_lock.
256 */
257 struct list_head cgrp_links;
258
259 /*
260 * List of csets participating in the on-going migration either as
261 * source or destination. Protected by cgroup_mutex.
262 */
263 struct list_head mg_src_preload_node;
264 struct list_head mg_dst_preload_node;
265 struct list_head mg_node;
266
267 /*
268 * If this cset is acting as the source of migration the following
269 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
270 * respectively the source and destination cgroups of the on-going
271 * migration. mg_dst_cset is the destination cset the target tasks
272 * on this cset should be migrated to. Protected by cgroup_mutex.
273 */
274 struct cgroup *mg_src_cgrp;
275 struct cgroup *mg_dst_cgrp;
276 struct css_set *mg_dst_cset;
277
278 /* dead and being drained, ignore for migration */
279 bool dead;
280
281 /* For RCU-protected deletion */
282 struct rcu_head rcu_head;
283 };
284
285 struct cgroup_base_stat {
286 struct task_cputime cputime;
287 };
288
289 /*
290 * rstat - cgroup scalable recursive statistics. Accounting is done
291 * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
292 * hierarchy on reads.
293 *
294 * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
295 * linked into the updated tree. On the following read, propagation only
296 * considers and consumes the updated tree. This makes reading O(the
297 * number of descendants which have been active since last read) instead of
298 * O(the total number of descendants).
299 *
300 * This is important because there can be a lot of (draining) cgroups which
301 * aren't active and stat may be read frequently. The combination can
302 * become very expensive. By propagating selectively, increasing reading
303 * frequency decreases the cost of each read.
304 *
305 * This struct hosts both the fields which implement the above -
306 * updated_children and updated_next - and the fields which track basic
307 * resource statistics on top of it - bsync, bstat and last_bstat.
308 */
309 struct cgroup_rstat_cpu {
310 /*
311 * ->bsync protects ->bstat. These are the only fields which get
312 * updated in the hot path.
313 */
314 struct u64_stats_sync bsync;
315 struct cgroup_base_stat bstat;
316
317 /*
318 * Snapshots at the last reading. These are used to calculate the
319 * deltas to propagate to the global counters.
320 */
321 struct cgroup_base_stat last_bstat;
322
323 /*
324 * Child cgroups with stat updates on this cpu since the last read
325 * are linked on the parent's ->updated_children through
326 * ->updated_next.
327 *
328 * In addition to being more compact, singly-linked list pointing
329 * to the cgroup makes it unnecessary for each per-cpu struct to
330 * point back to the associated cgroup.
331 *
332 * Protected by per-cpu cgroup_rstat_cpu_lock.
333 */
334 struct cgroup *updated_children; /* terminated by self cgroup */
335 struct cgroup *updated_next; /* NULL iff not on the list */
336 };
337
338 struct cgroup_freezer_state {
339 /* Should the cgroup and its descendants be frozen. */
340 bool freeze;
341
342 /* Should the cgroup actually be frozen? */
343 int e_freeze;
344
345 /* Fields below are protected by css_set_lock */
346
347 /* Number of frozen descendant cgroups */
348 int nr_frozen_descendants;
349
350 /*
351 * Number of tasks, which are counted as frozen:
352 * frozen, SIGSTOPped, and PTRACEd.
353 */
354 int nr_frozen_tasks;
355 };
356
357 struct cgroup {
358 /* self css with NULL ->ss, points back to this cgroup */
359 struct cgroup_subsys_state self;
360
361 unsigned long flags; /* "unsigned long" so bitops work */
362
363 /*
364 * The depth this cgroup is at. The root is at depth zero and each
365 * step down the hierarchy increments the level. This along with
366 * ancestor_ids[] can determine whether a given cgroup is a
367 * descendant of another without traversing the hierarchy.
368 */
369 int level;
370
371 /* Maximum allowed descent tree depth */
372 int max_depth;
373
374 /*
375 * Keep track of total numbers of visible and dying descent cgroups.
376 * Dying cgroups are cgroups which were deleted by a user,
377 * but are still existing because someone else is holding a reference.
378 * max_descendants is a maximum allowed number of descent cgroups.
379 *
380 * nr_descendants and nr_dying_descendants are protected
381 * by cgroup_mutex and css_set_lock. It's fine to read them holding
382 * any of cgroup_mutex and css_set_lock; for writing both locks
383 * should be held.
384 */
385 int nr_descendants;
386 int nr_dying_descendants;
387 int max_descendants;
388
389 /*
390 * Each non-empty css_set associated with this cgroup contributes
391 * one to nr_populated_csets. The counter is zero iff this cgroup
392 * doesn't have any tasks.
393 *
394 * All children which have non-zero nr_populated_csets and/or
395 * nr_populated_children of their own contribute one to either
396 * nr_populated_domain_children or nr_populated_threaded_children
397 * depending on their type. Each counter is zero iff all cgroups
398 * of the type in the subtree proper don't have any tasks.
399 */
400 int nr_populated_csets;
401 int nr_populated_domain_children;
402 int nr_populated_threaded_children;
403
404 int nr_threaded_children; /* # of live threaded child cgroups */
405
406 struct kernfs_node *kn; /* cgroup kernfs entry */
407 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
408 struct cgroup_file events_file; /* handle for "cgroup.events" */
409
410 /*
411 * The bitmask of subsystems enabled on the child cgroups.
412 * ->subtree_control is the one configured through
413 * "cgroup.subtree_control" while ->child_ss_mask is the effective
414 * one which may have more subsystems enabled. Controller knobs
415 * are made available iff it's enabled in ->subtree_control.
416 */
417 u16 subtree_control;
418 u16 subtree_ss_mask;
419 u16 old_subtree_control;
420 u16 old_subtree_ss_mask;
421
422 /* Private pointers for each registered subsystem */
423 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
424
425 struct cgroup_root *root;
426
427 /*
428 * List of cgrp_cset_links pointing at css_sets with tasks in this
429 * cgroup. Protected by css_set_lock.
430 */
431 struct list_head cset_links;
432
433 /*
434 * On the default hierarchy, a css_set for a cgroup with some
435 * susbsys disabled will point to css's which are associated with
436 * the closest ancestor which has the subsys enabled. The
437 * following lists all css_sets which point to this cgroup's css
438 * for the given subsystem.
439 */
440 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
441
442 /*
443 * If !threaded, self. If threaded, it points to the nearest
444 * domain ancestor. Inside a threaded subtree, cgroups are exempt
445 * from process granularity and no-internal-task constraint.
446 * Domain level resource consumptions which aren't tied to a
447 * specific task are charged to the dom_cgrp.
448 */
449 struct cgroup *dom_cgrp;
450 struct cgroup *old_dom_cgrp; /* used while enabling threaded */
451
452 /* per-cpu recursive resource statistics */
453 struct cgroup_rstat_cpu __percpu *rstat_cpu;
454 struct list_head rstat_css_list;
455
456 /* cgroup basic resource statistics */
457 struct cgroup_base_stat last_bstat;
458 struct cgroup_base_stat bstat;
459 struct prev_cputime prev_cputime; /* for printing out cputime */
460
461 /*
462 * list of pidlists, up to two for each namespace (one for procs, one
463 * for tasks); created on demand.
464 */
465 struct list_head pidlists;
466 struct mutex pidlist_mutex;
467
468 /* used to wait for offlining of csses */
469 wait_queue_head_t offline_waitq;
470
471 /* used to schedule release agent */
472 struct work_struct release_agent_work;
473
474 /* used to track pressure stalls */
475 struct psi_group psi;
476
477 /* used to store eBPF programs */
478 struct cgroup_bpf bpf;
479
480 /* If there is block congestion on this cgroup. */
481 atomic_t congestion_count;
482
483 /* Used to store internal freezer state */
484 struct cgroup_freezer_state freezer;
485
486 /* ids of the ancestors at each level including self */
487 u64 ancestor_ids[];
488 };
489
490 /*
491 * A cgroup_root represents the root of a cgroup hierarchy, and may be
492 * associated with a kernfs_root to form an active hierarchy. This is
493 * internal to cgroup core. Don't access directly from controllers.
494 */
495 struct cgroup_root {
496 struct kernfs_root *kf_root;
497
498 /* The bitmask of subsystems attached to this hierarchy */
499 unsigned int subsys_mask;
500
501 /* Unique id for this hierarchy. */
502 int hierarchy_id;
503
504 /* The root cgroup. Root is destroyed on its release. */
505 struct cgroup cgrp;
506
507 /* for cgrp->ancestor_ids[0] */
508 u64 cgrp_ancestor_id_storage;
509
510 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
511 atomic_t nr_cgrps;
512
513 /* Wait while cgroups are being destroyed */
514 wait_queue_head_t wait;
515
516 /* A list running through the active hierarchies */
517 struct list_head root_list;
518
519 /* Hierarchy-specific flags */
520 unsigned int flags;
521
522 /* The path to use for release notifications. */
523 char release_agent_path[PATH_MAX];
524
525 /* The name for this hierarchy - may be empty */
526 char name[MAX_CGROUP_ROOT_NAMELEN];
527 };
528
529 /*
530 * struct cftype: handler definitions for cgroup control files
531 *
532 * When reading/writing to a file:
533 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
534 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
535 */
536 struct cftype {
537 /*
538 * By convention, the name should begin with the name of the
539 * subsystem, followed by a period. Zero length string indicates
540 * end of cftype array.
541 */
542 char name[MAX_CFTYPE_NAME];
543 unsigned long private;
544
545 /*
546 * The maximum length of string, excluding trailing nul, that can
547 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
548 */
549 size_t max_write_len;
550
551 /* CFTYPE_* flags */
552 unsigned int flags;
553
554 /*
555 * If non-zero, should contain the offset from the start of css to
556 * a struct cgroup_file field. cgroup will record the handle of
557 * the created file into it. The recorded handle can be used as
558 * long as the containing css remains accessible.
559 */
560 unsigned int file_offset;
561
562 /*
563 * Fields used for internal bookkeeping. Initialized automatically
564 * during registration.
565 */
566 struct cgroup_subsys *ss; /* NULL for cgroup core files */
567 struct list_head node; /* anchored at ss->cfts */
568 struct kernfs_ops *kf_ops;
569
570 int (*open)(struct kernfs_open_file *of);
571 void (*release)(struct kernfs_open_file *of);
572
573 /*
574 * read_u64() is a shortcut for the common case of returning a
575 * single integer. Use it in place of read()
576 */
577 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
578 /*
579 * read_s64() is a signed version of read_u64()
580 */
581 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
582
583 /* generic seq_file read interface */
584 int (*seq_show)(struct seq_file *sf, void *v);
585
586 /* optional ops, implement all or none */
587 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
588 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
589 void (*seq_stop)(struct seq_file *sf, void *v);
590
591 /*
592 * write_u64() is a shortcut for the common case of accepting
593 * a single integer (as parsed by simple_strtoull) from
594 * userspace. Use in place of write(); return 0 or error.
595 */
596 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
597 u64 val);
598 /*
599 * write_s64() is a signed version of write_u64()
600 */
601 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
602 s64 val);
603
604 /*
605 * write() is the generic write callback which maps directly to
606 * kernfs write operation and overrides all other operations.
607 * Maximum write size is determined by ->max_write_len. Use
608 * of_css/cft() to access the associated css and cft.
609 */
610 ssize_t (*write)(struct kernfs_open_file *of,
611 char *buf, size_t nbytes, loff_t off);
612
613 __poll_t (*poll)(struct kernfs_open_file *of,
614 struct poll_table_struct *pt);
615
616 #ifdef CONFIG_DEBUG_LOCK_ALLOC
617 struct lock_class_key lockdep_key;
618 #endif
619 };
620
621 /*
622 * Control Group subsystem type.
623 * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
624 */
625 struct cgroup_subsys {
626 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
627 int (*css_online)(struct cgroup_subsys_state *css);
628 void (*css_offline)(struct cgroup_subsys_state *css);
629 void (*css_released)(struct cgroup_subsys_state *css);
630 void (*css_free)(struct cgroup_subsys_state *css);
631 void (*css_reset)(struct cgroup_subsys_state *css);
632 void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
633 int (*css_extra_stat_show)(struct seq_file *seq,
634 struct cgroup_subsys_state *css);
635
636 int (*can_attach)(struct cgroup_taskset *tset);
637 void (*cancel_attach)(struct cgroup_taskset *tset);
638 void (*attach)(struct cgroup_taskset *tset);
639 void (*post_attach)(void);
640 int (*can_fork)(struct task_struct *task,
641 struct css_set *cset);
642 void (*cancel_fork)(struct task_struct *task, struct css_set *cset);
643 void (*fork)(struct task_struct *task);
644 void (*exit)(struct task_struct *task);
645 void (*release)(struct task_struct *task);
646 void (*bind)(struct cgroup_subsys_state *root_css);
647
648 bool early_init:1;
649
650 /*
651 * If %true, the controller, on the default hierarchy, doesn't show
652 * up in "cgroup.controllers" or "cgroup.subtree_control", is
653 * implicitly enabled on all cgroups on the default hierarchy, and
654 * bypasses the "no internal process" constraint. This is for
655 * utility type controllers which is transparent to userland.
656 *
657 * An implicit controller can be stolen from the default hierarchy
658 * anytime and thus must be okay with offline csses from previous
659 * hierarchies coexisting with csses for the current one.
660 */
661 bool implicit_on_dfl:1;
662
663 /*
664 * If %true, the controller, supports threaded mode on the default
665 * hierarchy. In a threaded subtree, both process granularity and
666 * no-internal-process constraint are ignored and a threaded
667 * controllers should be able to handle that.
668 *
669 * Note that as an implicit controller is automatically enabled on
670 * all cgroups on the default hierarchy, it should also be
671 * threaded. implicit && !threaded is not supported.
672 */
673 bool threaded:1;
674
675 /*
676 * If %false, this subsystem is properly hierarchical -
677 * configuration, resource accounting and restriction on a parent
678 * cgroup cover those of its children. If %true, hierarchy support
679 * is broken in some ways - some subsystems ignore hierarchy
680 * completely while others are only implemented half-way.
681 *
682 * It's now disallowed to create nested cgroups if the subsystem is
683 * broken and cgroup core will emit a warning message on such
684 * cases. Eventually, all subsystems will be made properly
685 * hierarchical and this will go away.
686 */
687 bool broken_hierarchy:1;
688 bool warned_broken_hierarchy:1;
689
690 /* the following two fields are initialized automtically during boot */
691 int id;
692 const char *name;
693
694 /* optional, initialized automatically during boot if not set */
695 const char *legacy_name;
696
697 /* link to parent, protected by cgroup_lock() */
698 struct cgroup_root *root;
699
700 /* idr for css->id */
701 struct idr css_idr;
702
703 /*
704 * List of cftypes. Each entry is the first entry of an array
705 * terminated by zero length name.
706 */
707 struct list_head cfts;
708
709 /*
710 * Base cftypes which are automatically registered. The two can
711 * point to the same array.
712 */
713 struct cftype *dfl_cftypes; /* for the default hierarchy */
714 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
715
716 /*
717 * A subsystem may depend on other subsystems. When such subsystem
718 * is enabled on a cgroup, the depended-upon subsystems are enabled
719 * together if available. Subsystems enabled due to dependency are
720 * not visible to userland until explicitly enabled. The following
721 * specifies the mask of subsystems that this one depends on.
722 */
723 unsigned int depends_on;
724 };
725
726 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
727
728 /**
729 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
730 * @tsk: target task
731 *
732 * Allows cgroup operations to synchronize against threadgroup changes
733 * using a percpu_rw_semaphore.
734 */
cgroup_threadgroup_change_begin(struct task_struct * tsk)735 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
736 {
737 percpu_down_read(&cgroup_threadgroup_rwsem);
738 }
739
740 /**
741 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
742 * @tsk: target task
743 *
744 * Counterpart of cgroup_threadcgroup_change_begin().
745 */
cgroup_threadgroup_change_end(struct task_struct * tsk)746 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
747 {
748 percpu_up_read(&cgroup_threadgroup_rwsem);
749 }
750
751 #else /* CONFIG_CGROUPS */
752
753 #define CGROUP_SUBSYS_COUNT 0
754
cgroup_threadgroup_change_begin(struct task_struct * tsk)755 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
756 {
757 might_sleep();
758 }
759
cgroup_threadgroup_change_end(struct task_struct * tsk)760 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
761
762 #endif /* CONFIG_CGROUPS */
763
764 #ifdef CONFIG_SOCK_CGROUP_DATA
765
766 /*
767 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
768 * per-socket cgroup information except for memcg association.
769 *
770 * On legacy hierarchies, net_prio and net_cls controllers directly
771 * set attributes on each sock which can then be tested by the network
772 * layer. On the default hierarchy, each sock is associated with the
773 * cgroup it was created in and the networking layer can match the
774 * cgroup directly.
775 */
776 struct sock_cgroup_data {
777 struct cgroup *cgroup; /* v2 */
778 #ifdef CONFIG_CGROUP_NET_CLASSID
779 u32 classid; /* v1 */
780 #endif
781 #ifdef CONFIG_CGROUP_NET_PRIO
782 u16 prioidx; /* v1 */
783 #endif
784 };
785
sock_cgroup_prioidx(const struct sock_cgroup_data * skcd)786 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
787 {
788 #ifdef CONFIG_CGROUP_NET_PRIO
789 return READ_ONCE(skcd->prioidx);
790 #else
791 return 1;
792 #endif
793 }
794
sock_cgroup_classid(const struct sock_cgroup_data * skcd)795 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
796 {
797 #ifdef CONFIG_CGROUP_NET_CLASSID
798 return READ_ONCE(skcd->classid);
799 #else
800 return 0;
801 #endif
802 }
803
sock_cgroup_set_prioidx(struct sock_cgroup_data * skcd,u16 prioidx)804 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
805 u16 prioidx)
806 {
807 #ifdef CONFIG_CGROUP_NET_PRIO
808 WRITE_ONCE(skcd->prioidx, prioidx);
809 #endif
810 }
811
sock_cgroup_set_classid(struct sock_cgroup_data * skcd,u32 classid)812 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
813 u32 classid)
814 {
815 #ifdef CONFIG_CGROUP_NET_CLASSID
816 WRITE_ONCE(skcd->classid, classid);
817 #endif
818 }
819
820 #else /* CONFIG_SOCK_CGROUP_DATA */
821
822 struct sock_cgroup_data {
823 };
824
825 #endif /* CONFIG_SOCK_CGROUP_DATA */
826
827 #endif /* _LINUX_CGROUP_DEFS_H */
828