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 /* cftype->flags */
100 enum {
101 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
102 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
103 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
104
105 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
106 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
107 CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */
108
109 /* internal flags, do not use outside cgroup core proper */
110 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
111 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
112 };
113
114 /*
115 * cgroup_file is the handle for a file instance created in a cgroup which
116 * is used, for example, to generate file changed notifications. This can
117 * be obtained by setting cftype->file_offset.
118 */
119 struct cgroup_file {
120 /* do not access any fields from outside cgroup core */
121 struct kernfs_node *kn;
122 unsigned long notified_at;
123 struct timer_list notify_timer;
124 };
125
126 /*
127 * Per-subsystem/per-cgroup state maintained by the system. This is the
128 * fundamental structural building block that controllers deal with.
129 *
130 * Fields marked with "PI:" are public and immutable and may be accessed
131 * directly without synchronization.
132 */
133 struct cgroup_subsys_state {
134 /* PI: the cgroup that this css is attached to */
135 struct cgroup *cgroup;
136
137 /* PI: the cgroup subsystem that this css is attached to */
138 struct cgroup_subsys *ss;
139
140 /* reference count - access via css_[try]get() and css_put() */
141 struct percpu_ref refcnt;
142
143 /* siblings list anchored at the parent's ->children */
144 struct list_head sibling;
145 struct list_head children;
146
147 /* flush target list anchored at cgrp->rstat_css_list */
148 struct list_head rstat_css_node;
149
150 /*
151 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
152 * matching css can be looked up using css_from_id().
153 */
154 int id;
155
156 unsigned int flags;
157
158 /*
159 * Monotonically increasing unique serial number which defines a
160 * uniform order among all csses. It's guaranteed that all
161 * ->children lists are in the ascending order of ->serial_nr and
162 * used to allow interrupting and resuming iterations.
163 */
164 u64 serial_nr;
165
166 /*
167 * Incremented by online self and children. Used to guarantee that
168 * parents are not offlined before their children.
169 */
170 atomic_t online_cnt;
171
172 /* percpu_ref killing and RCU release */
173 struct work_struct destroy_work;
174 struct rcu_work destroy_rwork;
175
176 /*
177 * PI: the parent css. Placed here for cache proximity to following
178 * fields of the containing structure.
179 */
180 struct cgroup_subsys_state *parent;
181 };
182
183 /*
184 * A css_set is a structure holding pointers to a set of
185 * cgroup_subsys_state objects. This saves space in the task struct
186 * object and speeds up fork()/exit(), since a single inc/dec and a
187 * list_add()/del() can bump the reference count on the entire cgroup
188 * set for a task.
189 */
190 struct css_set {
191 /*
192 * Set of subsystem states, one for each subsystem. This array is
193 * immutable after creation apart from the init_css_set during
194 * subsystem registration (at boot time).
195 */
196 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
197
198 /* reference count */
199 refcount_t refcount;
200
201 /*
202 * For a domain cgroup, the following points to self. If threaded,
203 * to the matching cset of the nearest domain ancestor. The
204 * dom_cset provides access to the domain cgroup and its csses to
205 * which domain level resource consumptions should be charged.
206 */
207 struct css_set *dom_cset;
208
209 /* the default cgroup associated with this css_set */
210 struct cgroup *dfl_cgrp;
211
212 /* internal task count, protected by css_set_lock */
213 int nr_tasks;
214
215 /*
216 * Lists running through all tasks using this cgroup group.
217 * mg_tasks lists tasks which belong to this cset but are in the
218 * process of being migrated out or in. Protected by
219 * css_set_rwsem, but, during migration, once tasks are moved to
220 * mg_tasks, it can be read safely while holding cgroup_mutex.
221 */
222 struct list_head tasks;
223 struct list_head mg_tasks;
224 struct list_head dying_tasks;
225
226 /* all css_task_iters currently walking this cset */
227 struct list_head task_iters;
228
229 /*
230 * On the default hierarhcy, ->subsys[ssid] may point to a css
231 * attached to an ancestor instead of the cgroup this css_set is
232 * associated with. The following node is anchored at
233 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
234 * iterate through all css's attached to a given cgroup.
235 */
236 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
237
238 /* all threaded csets whose ->dom_cset points to this cset */
239 struct list_head threaded_csets;
240 struct list_head threaded_csets_node;
241
242 /*
243 * List running through all cgroup groups in the same hash
244 * slot. Protected by css_set_lock
245 */
246 struct hlist_node hlist;
247
248 /*
249 * List of cgrp_cset_links pointing at cgroups referenced from this
250 * css_set. Protected by css_set_lock.
251 */
252 struct list_head cgrp_links;
253
254 /*
255 * List of csets participating in the on-going migration either as
256 * source or destination. Protected by cgroup_mutex.
257 */
258 struct list_head mg_preload_node;
259 struct list_head mg_node;
260
261 /*
262 * If this cset is acting as the source of migration the following
263 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
264 * respectively the source and destination cgroups of the on-going
265 * migration. mg_dst_cset is the destination cset the target tasks
266 * on this cset should be migrated to. Protected by cgroup_mutex.
267 */
268 struct cgroup *mg_src_cgrp;
269 struct cgroup *mg_dst_cgrp;
270 struct css_set *mg_dst_cset;
271
272 /* dead and being drained, ignore for migration */
273 bool dead;
274
275 /* For RCU-protected deletion */
276 struct rcu_head rcu_head;
277 };
278
279 struct cgroup_base_stat {
280 struct task_cputime cputime;
281 };
282
283 /*
284 * rstat - cgroup scalable recursive statistics. Accounting is done
285 * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
286 * hierarchy on reads.
287 *
288 * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
289 * linked into the updated tree. On the following read, propagation only
290 * considers and consumes the updated tree. This makes reading O(the
291 * number of descendants which have been active since last read) instead of
292 * O(the total number of descendants).
293 *
294 * This is important because there can be a lot of (draining) cgroups which
295 * aren't active and stat may be read frequently. The combination can
296 * become very expensive. By propagating selectively, increasing reading
297 * frequency decreases the cost of each read.
298 *
299 * This struct hosts both the fields which implement the above -
300 * updated_children and updated_next - and the fields which track basic
301 * resource statistics on top of it - bsync, bstat and last_bstat.
302 */
303 struct cgroup_rstat_cpu {
304 /*
305 * ->bsync protects ->bstat. These are the only fields which get
306 * updated in the hot path.
307 */
308 struct u64_stats_sync bsync;
309 struct cgroup_base_stat bstat;
310
311 /*
312 * Snapshots at the last reading. These are used to calculate the
313 * deltas to propagate to the global counters.
314 */
315 struct cgroup_base_stat last_bstat;
316
317 /*
318 * Child cgroups with stat updates on this cpu since the last read
319 * are linked on the parent's ->updated_children through
320 * ->updated_next.
321 *
322 * In addition to being more compact, singly-linked list pointing
323 * to the cgroup makes it unnecessary for each per-cpu struct to
324 * point back to the associated cgroup.
325 *
326 * Protected by per-cpu cgroup_rstat_cpu_lock.
327 */
328 struct cgroup *updated_children; /* terminated by self cgroup */
329 struct cgroup *updated_next; /* NULL iff not on the list */
330 };
331
332 struct cgroup_freezer_state {
333 /* Should the cgroup and its descendants be frozen. */
334 bool freeze;
335
336 /* Should the cgroup actually be frozen? */
337 int e_freeze;
338
339 /* Fields below are protected by css_set_lock */
340
341 /* Number of frozen descendant cgroups */
342 int nr_frozen_descendants;
343
344 /*
345 * Number of tasks, which are counted as frozen:
346 * frozen, SIGSTOPped, and PTRACEd.
347 */
348 int nr_frozen_tasks;
349 };
350
351 struct cgroup {
352 /* self css with NULL ->ss, points back to this cgroup */
353 struct cgroup_subsys_state self;
354
355 unsigned long flags; /* "unsigned long" so bitops work */
356
357 /*
358 * idr allocated in-hierarchy ID.
359 *
360 * ID 0 is not used, the ID of the root cgroup is always 1, and a
361 * new cgroup will be assigned with a smallest available ID.
362 *
363 * Allocating/Removing ID must be protected by cgroup_mutex.
364 */
365 int id;
366
367 /*
368 * The depth this cgroup is at. The root is at depth zero and each
369 * step down the hierarchy increments the level. This along with
370 * ancestor_ids[] can determine whether a given cgroup is a
371 * descendant of another without traversing the hierarchy.
372 */
373 int level;
374
375 /* Maximum allowed descent tree depth */
376 int max_depth;
377
378 /*
379 * Keep track of total numbers of visible and dying descent cgroups.
380 * Dying cgroups are cgroups which were deleted by a user,
381 * but are still existing because someone else is holding a reference.
382 * max_descendants is a maximum allowed number of descent cgroups.
383 *
384 * nr_descendants and nr_dying_descendants are protected
385 * by cgroup_mutex and css_set_lock. It's fine to read them holding
386 * any of cgroup_mutex and css_set_lock; for writing both locks
387 * should be held.
388 */
389 int nr_descendants;
390 int nr_dying_descendants;
391 int max_descendants;
392
393 /*
394 * Each non-empty css_set associated with this cgroup contributes
395 * one to nr_populated_csets. The counter is zero iff this cgroup
396 * doesn't have any tasks.
397 *
398 * All children which have non-zero nr_populated_csets and/or
399 * nr_populated_children of their own contribute one to either
400 * nr_populated_domain_children or nr_populated_threaded_children
401 * depending on their type. Each counter is zero iff all cgroups
402 * of the type in the subtree proper don't have any tasks.
403 */
404 int nr_populated_csets;
405 int nr_populated_domain_children;
406 int nr_populated_threaded_children;
407
408 int nr_threaded_children; /* # of live threaded child cgroups */
409
410 struct kernfs_node *kn; /* cgroup kernfs entry */
411 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
412 struct cgroup_file events_file; /* handle for "cgroup.events" */
413
414 /*
415 * The bitmask of subsystems enabled on the child cgroups.
416 * ->subtree_control is the one configured through
417 * "cgroup.subtree_control" while ->child_ss_mask is the effective
418 * one which may have more subsystems enabled. Controller knobs
419 * are made available iff it's enabled in ->subtree_control.
420 */
421 u16 subtree_control;
422 u16 subtree_ss_mask;
423 u16 old_subtree_control;
424 u16 old_subtree_ss_mask;
425
426 /* Private pointers for each registered subsystem */
427 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
428
429 struct cgroup_root *root;
430
431 /*
432 * List of cgrp_cset_links pointing at css_sets with tasks in this
433 * cgroup. Protected by css_set_lock.
434 */
435 struct list_head cset_links;
436
437 /*
438 * On the default hierarchy, a css_set for a cgroup with some
439 * susbsys disabled will point to css's which are associated with
440 * the closest ancestor which has the subsys enabled. The
441 * following lists all css_sets which point to this cgroup's css
442 * for the given subsystem.
443 */
444 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
445
446 /*
447 * If !threaded, self. If threaded, it points to the nearest
448 * domain ancestor. Inside a threaded subtree, cgroups are exempt
449 * from process granularity and no-internal-task constraint.
450 * Domain level resource consumptions which aren't tied to a
451 * specific task are charged to the dom_cgrp.
452 */
453 struct cgroup *dom_cgrp;
454 struct cgroup *old_dom_cgrp; /* used while enabling threaded */
455
456 /* per-cpu recursive resource statistics */
457 struct cgroup_rstat_cpu __percpu *rstat_cpu;
458 struct list_head rstat_css_list;
459
460 /* cgroup basic resource statistics */
461 struct cgroup_base_stat pending_bstat; /* pending from children */
462 struct cgroup_base_stat bstat;
463 struct prev_cputime prev_cputime; /* for printing out cputime */
464
465 /*
466 * list of pidlists, up to two for each namespace (one for procs, one
467 * for tasks); created on demand.
468 */
469 struct list_head pidlists;
470 struct mutex pidlist_mutex;
471
472 /* used to wait for offlining of csses */
473 wait_queue_head_t offline_waitq;
474
475 /* used to schedule release agent */
476 struct work_struct release_agent_work;
477
478 /* used to track pressure stalls */
479 struct psi_group psi;
480
481 /* used to store eBPF programs */
482 struct cgroup_bpf bpf;
483
484 /* If there is block congestion on this cgroup. */
485 atomic_t congestion_count;
486
487 /* Used to store internal freezer state */
488 struct cgroup_freezer_state freezer;
489
490 /* ids of the ancestors at each level including self */
491 int ancestor_ids[];
492 };
493
494 /*
495 * A cgroup_root represents the root of a cgroup hierarchy, and may be
496 * associated with a kernfs_root to form an active hierarchy. This is
497 * internal to cgroup core. Don't access directly from controllers.
498 */
499 struct cgroup_root {
500 struct kernfs_root *kf_root;
501
502 /* The bitmask of subsystems attached to this hierarchy */
503 unsigned int subsys_mask;
504
505 /* Unique id for this hierarchy. */
506 int hierarchy_id;
507
508 /* The root cgroup. Root is destroyed on its release. */
509 struct cgroup cgrp;
510
511 /* for cgrp->ancestor_ids[0] */
512 int cgrp_ancestor_id_storage;
513
514 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
515 atomic_t nr_cgrps;
516
517 /* A list running through the active hierarchies */
518 struct list_head root_list;
519
520 /* Hierarchy-specific flags */
521 unsigned int flags;
522
523 /* IDs for cgroups in this hierarchy */
524 struct idr cgroup_idr;
525
526 /* The path to use for release notifications. */
527 char release_agent_path[PATH_MAX];
528
529 /* The name for this hierarchy - may be empty */
530 char name[MAX_CGROUP_ROOT_NAMELEN];
531 };
532
533 /*
534 * struct cftype: handler definitions for cgroup control files
535 *
536 * When reading/writing to a file:
537 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
538 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
539 */
540 struct cftype {
541 /*
542 * By convention, the name should begin with the name of the
543 * subsystem, followed by a period. Zero length string indicates
544 * end of cftype array.
545 */
546 char name[MAX_CFTYPE_NAME];
547 unsigned long private;
548
549 /*
550 * The maximum length of string, excluding trailing nul, that can
551 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
552 */
553 size_t max_write_len;
554
555 /* CFTYPE_* flags */
556 unsigned int flags;
557
558 /*
559 * If non-zero, should contain the offset from the start of css to
560 * a struct cgroup_file field. cgroup will record the handle of
561 * the created file into it. The recorded handle can be used as
562 * long as the containing css remains accessible.
563 */
564 unsigned int file_offset;
565
566 /*
567 * Fields used for internal bookkeeping. Initialized automatically
568 * during registration.
569 */
570 struct cgroup_subsys *ss; /* NULL for cgroup core files */
571 struct list_head node; /* anchored at ss->cfts */
572 struct kernfs_ops *kf_ops;
573
574 int (*open)(struct kernfs_open_file *of);
575 void (*release)(struct kernfs_open_file *of);
576
577 /*
578 * read_u64() is a shortcut for the common case of returning a
579 * single integer. Use it in place of read()
580 */
581 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
582 /*
583 * read_s64() is a signed version of read_u64()
584 */
585 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
586
587 /* generic seq_file read interface */
588 int (*seq_show)(struct seq_file *sf, void *v);
589
590 /* optional ops, implement all or none */
591 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
592 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
593 void (*seq_stop)(struct seq_file *sf, void *v);
594
595 /*
596 * write_u64() is a shortcut for the common case of accepting
597 * a single integer (as parsed by simple_strtoull) from
598 * userspace. Use in place of write(); return 0 or error.
599 */
600 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
601 u64 val);
602 /*
603 * write_s64() is a signed version of write_u64()
604 */
605 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
606 s64 val);
607
608 /*
609 * write() is the generic write callback which maps directly to
610 * kernfs write operation and overrides all other operations.
611 * Maximum write size is determined by ->max_write_len. Use
612 * of_css/cft() to access the associated css and cft.
613 */
614 ssize_t (*write)(struct kernfs_open_file *of,
615 char *buf, size_t nbytes, loff_t off);
616
617 __poll_t (*poll)(struct kernfs_open_file *of,
618 struct poll_table_struct *pt);
619
620 #ifdef CONFIG_DEBUG_LOCK_ALLOC
621 struct lock_class_key lockdep_key;
622 #endif
623 };
624
625 /*
626 * Control Group subsystem type.
627 * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
628 */
629 struct cgroup_subsys {
630 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
631 int (*css_online)(struct cgroup_subsys_state *css);
632 void (*css_offline)(struct cgroup_subsys_state *css);
633 void (*css_released)(struct cgroup_subsys_state *css);
634 void (*css_free)(struct cgroup_subsys_state *css);
635 void (*css_reset)(struct cgroup_subsys_state *css);
636 void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
637 int (*css_extra_stat_show)(struct seq_file *seq,
638 struct cgroup_subsys_state *css);
639
640 int (*can_attach)(struct cgroup_taskset *tset);
641 void (*cancel_attach)(struct cgroup_taskset *tset);
642 void (*attach)(struct cgroup_taskset *tset);
643 void (*post_attach)(void);
644 int (*can_fork)(struct task_struct *task);
645 void (*cancel_fork)(struct task_struct *task);
646 void (*fork)(struct task_struct *task);
647 void (*exit)(struct task_struct *task);
648 void (*release)(struct task_struct *task);
649 void (*bind)(struct cgroup_subsys_state *root_css);
650
651 bool early_init:1;
652
653 /*
654 * If %true, the controller, on the default hierarchy, doesn't show
655 * up in "cgroup.controllers" or "cgroup.subtree_control", is
656 * implicitly enabled on all cgroups on the default hierarchy, and
657 * bypasses the "no internal process" constraint. This is for
658 * utility type controllers which is transparent to userland.
659 *
660 * An implicit controller can be stolen from the default hierarchy
661 * anytime and thus must be okay with offline csses from previous
662 * hierarchies coexisting with csses for the current one.
663 */
664 bool implicit_on_dfl:1;
665
666 /*
667 * If %true, the controller, supports threaded mode on the default
668 * hierarchy. In a threaded subtree, both process granularity and
669 * no-internal-process constraint are ignored and a threaded
670 * controllers should be able to handle that.
671 *
672 * Note that as an implicit controller is automatically enabled on
673 * all cgroups on the default hierarchy, it should also be
674 * threaded. implicit && !threaded is not supported.
675 */
676 bool threaded:1;
677
678 /*
679 * If %false, this subsystem is properly hierarchical -
680 * configuration, resource accounting and restriction on a parent
681 * cgroup cover those of its children. If %true, hierarchy support
682 * is broken in some ways - some subsystems ignore hierarchy
683 * completely while others are only implemented half-way.
684 *
685 * It's now disallowed to create nested cgroups if the subsystem is
686 * broken and cgroup core will emit a warning message on such
687 * cases. Eventually, all subsystems will be made properly
688 * hierarchical and this will go away.
689 */
690 bool broken_hierarchy:1;
691 bool warned_broken_hierarchy:1;
692
693 /* the following two fields are initialized automtically during boot */
694 int id;
695 const char *name;
696
697 /* optional, initialized automatically during boot if not set */
698 const char *legacy_name;
699
700 /* link to parent, protected by cgroup_lock() */
701 struct cgroup_root *root;
702
703 /* idr for css->id */
704 struct idr css_idr;
705
706 /*
707 * List of cftypes. Each entry is the first entry of an array
708 * terminated by zero length name.
709 */
710 struct list_head cfts;
711
712 /*
713 * Base cftypes which are automatically registered. The two can
714 * point to the same array.
715 */
716 struct cftype *dfl_cftypes; /* for the default hierarchy */
717 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
718
719 /*
720 * A subsystem may depend on other subsystems. When such subsystem
721 * is enabled on a cgroup, the depended-upon subsystems are enabled
722 * together if available. Subsystems enabled due to dependency are
723 * not visible to userland until explicitly enabled. The following
724 * specifies the mask of subsystems that this one depends on.
725 */
726 unsigned int depends_on;
727 };
728
729 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
730
731 /**
732 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
733 * @tsk: target task
734 *
735 * Allows cgroup operations to synchronize against threadgroup changes
736 * using a percpu_rw_semaphore.
737 */
738 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
739 {
740 percpu_down_read(&cgroup_threadgroup_rwsem);
741 }
742
743 /**
744 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
745 * @tsk: target task
746 *
747 * Counterpart of cgroup_threadcgroup_change_begin().
748 */
749 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
750 {
751 percpu_up_read(&cgroup_threadgroup_rwsem);
752 }
753
754 #else /* CONFIG_CGROUPS */
755
756 #define CGROUP_SUBSYS_COUNT 0
757
758 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
759 {
760 might_sleep();
761 }
762
763 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
764
765 #endif /* CONFIG_CGROUPS */
766
767 #ifdef CONFIG_SOCK_CGROUP_DATA
768
769 /*
770 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
771 * per-socket cgroup information except for memcg association.
772 *
773 * On legacy hierarchies, net_prio and net_cls controllers directly set
774 * attributes on each sock which can then be tested by the network layer.
775 * On the default hierarchy, each sock is associated with the cgroup it was
776 * created in and the networking layer can match the cgroup directly.
777 *
778 * To avoid carrying all three cgroup related fields separately in sock,
779 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
780 * On boot, sock_cgroup_data records the cgroup that the sock was created
781 * in so that cgroup2 matches can be made; however, once either net_prio or
782 * net_cls starts being used, the area is overriden to carry prioidx and/or
783 * classid. The two modes are distinguished by whether the lowest bit is
784 * set. Clear bit indicates cgroup pointer while set bit prioidx and
785 * classid.
786 *
787 * While userland may start using net_prio or net_cls at any time, once
788 * either is used, cgroup2 matching no longer works. There is no reason to
789 * mix the two and this is in line with how legacy and v2 compatibility is
790 * handled. On mode switch, cgroup references which are already being
791 * pointed to by socks may be leaked. While this can be remedied by adding
792 * synchronization around sock_cgroup_data, given that the number of leaked
793 * cgroups is bound and highly unlikely to be high, this seems to be the
794 * better trade-off.
795 */
796 struct sock_cgroup_data {
797 union {
798 #ifdef __LITTLE_ENDIAN
799 struct {
800 u8 is_data;
801 u8 padding;
802 u16 prioidx;
803 u32 classid;
804 } __packed;
805 #else
806 struct {
807 u32 classid;
808 u16 prioidx;
809 u8 padding;
810 u8 is_data;
811 } __packed;
812 #endif
813 u64 val;
814 };
815 };
816
817 /*
818 * There's a theoretical window where the following accessors race with
819 * updaters and return part of the previous pointer as the prioidx or
820 * classid. Such races are short-lived and the result isn't critical.
821 */
822 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
823 {
824 /* fallback to 1 which is always the ID of the root cgroup */
825 return (skcd->is_data & 1) ? skcd->prioidx : 1;
826 }
827
828 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
829 {
830 /* fallback to 0 which is the unconfigured default classid */
831 return (skcd->is_data & 1) ? skcd->classid : 0;
832 }
833
834 /*
835 * If invoked concurrently, the updaters may clobber each other. The
836 * caller is responsible for synchronization.
837 */
838 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
839 u16 prioidx)
840 {
841 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
842
843 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
844 return;
845
846 if (!(skcd_buf.is_data & 1)) {
847 skcd_buf.val = 0;
848 skcd_buf.is_data = 1;
849 }
850
851 skcd_buf.prioidx = prioidx;
852 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
853 }
854
855 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
856 u32 classid)
857 {
858 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
859
860 if (sock_cgroup_classid(&skcd_buf) == classid)
861 return;
862
863 if (!(skcd_buf.is_data & 1)) {
864 skcd_buf.val = 0;
865 skcd_buf.is_data = 1;
866 }
867
868 skcd_buf.classid = classid;
869 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
870 }
871
872 #else /* CONFIG_SOCK_CGROUP_DATA */
873
874 struct sock_cgroup_data {
875 };
876
877 #endif /* CONFIG_SOCK_CGROUP_DATA */
878
879 #endif /* _LINUX_CGROUP_DEFS_H */