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/android_kabi.h>
12 #include <linux/limits.h>
13 #include <linux/list.h>
14 #include <linux/idr.h>
15 #include <linux/wait.h>
16 #include <linux/mutex.h>
17 #include <linux/rcupdate.h>
18 #include <linux/refcount.h>
19 #include <linux/percpu-refcount.h>
20 #include <linux/percpu-rwsem.h>
21 #include <linux/u64_stats_sync.h>
22 #include <linux/workqueue.h>
23 #include <linux/bpf-cgroup-defs.h>
24 #include <linux/psi_types.h>
25
26 #ifdef CONFIG_CGROUPS
27
28 struct cgroup;
29 struct cgroup_root;
30 struct cgroup_subsys;
31 struct cgroup_taskset;
32 struct kernfs_node;
33 struct kernfs_ops;
34 struct kernfs_open_file;
35 struct seq_file;
36 struct poll_table_struct;
37
38 #define MAX_CGROUP_TYPE_NAMELEN 32
39 #define MAX_CGROUP_ROOT_NAMELEN 64
40 #define MAX_CFTYPE_NAME 64
41
42 /* define the enumeration of all cgroup subsystems */
43 #define SUBSYS(_x) _x ## _cgrp_id,
44 enum cgroup_subsys_id {
45 #include <linux/cgroup_subsys.h>
46 CGROUP_SUBSYS_COUNT,
47 };
48 #undef SUBSYS
49
50 /* bits in struct cgroup_subsys_state flags field */
51 enum {
52 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
53 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
54 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
55 CSS_VISIBLE = (1 << 3), /* css is visible to userland */
56 CSS_DYING = (1 << 4), /* css is dying */
57 };
58
59 /* bits in struct cgroup flags field */
60 enum {
61 /* Control Group requires release notifications to userspace */
62 CGRP_NOTIFY_ON_RELEASE,
63 /*
64 * Clone the parent's configuration when creating a new child
65 * cpuset cgroup. For historical reasons, this option can be
66 * specified at mount time and thus is implemented here.
67 */
68 CGRP_CPUSET_CLONE_CHILDREN,
69
70 /* Control group has to be frozen. */
71 CGRP_FREEZE,
72
73 /* Cgroup is frozen. */
74 CGRP_FROZEN,
75 };
76
77 /* cgroup_root->flags */
78 enum {
79 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
80 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
81
82 /*
83 * Consider namespaces as delegation boundaries. If this flag is
84 * set, controller specific interface files in a namespace root
85 * aren't writeable from inside the namespace.
86 */
87 CGRP_ROOT_NS_DELEGATE = (1 << 3),
88
89 /*
90 * Reduce latencies on dynamic cgroup modifications such as task
91 * migrations and controller on/offs by disabling percpu operation on
92 * cgroup_threadgroup_rwsem. This makes hot path operations such as
93 * forks and exits into the slow path and more expensive.
94 *
95 * The static usage pattern of creating a cgroup, enabling controllers,
96 * and then seeding it with CLONE_INTO_CGROUP doesn't require write
97 * locking cgroup_threadgroup_rwsem and thus doesn't benefit from
98 * favordynmod.
99 */
100 CGRP_ROOT_FAVOR_DYNMODS = (1 << 4),
101
102 /*
103 * Enable cpuset controller in v1 cgroup to use v2 behavior.
104 */
105 CGRP_ROOT_CPUSET_V2_MODE = (1 << 16),
106
107 /*
108 * Enable legacy local memory.events.
109 */
110 CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 17),
111
112 /*
113 * Enable recursive subtree protection
114 */
115 CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 18),
116
117 /*
118 * Enable hugetlb accounting for the memory controller.
119 */
120 CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING = (1 << 19),
121
122 /*
123 * Enable legacy local pids.events.
124 */
125 CGRP_ROOT_PIDS_LOCAL_EVENTS = (1 << 20),
126 };
127
128 /* cftype->flags */
129 enum {
130 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
131 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
132 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
133
134 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
135 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
136 CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */
137
138 /* internal flags, do not use outside cgroup core proper */
139 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
140 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
141 __CFTYPE_ADDED = (1 << 18),
142 };
143
144 /*
145 * cgroup_file is the handle for a file instance created in a cgroup which
146 * is used, for example, to generate file changed notifications. This can
147 * be obtained by setting cftype->file_offset.
148 */
149 struct cgroup_file {
150 /* do not access any fields from outside cgroup core */
151 struct kernfs_node *kn;
152 unsigned long notified_at;
153 struct timer_list notify_timer;
154 };
155
156 /*
157 * Per-subsystem/per-cgroup state maintained by the system. This is the
158 * fundamental structural building block that controllers deal with.
159 *
160 * Fields marked with "PI:" are public and immutable and may be accessed
161 * directly without synchronization.
162 */
163 struct cgroup_subsys_state {
164 /* PI: the cgroup that this css is attached to */
165 struct cgroup *cgroup;
166
167 /* PI: the cgroup subsystem that this css is attached to */
168 struct cgroup_subsys *ss;
169
170 /* reference count - access via css_[try]get() and css_put() */
171 struct percpu_ref refcnt;
172
173 /*
174 * siblings list anchored at the parent's ->children
175 *
176 * linkage is protected by cgroup_mutex or RCU
177 */
178 struct list_head sibling;
179 struct list_head children;
180
181 /* flush target list anchored at cgrp->rstat_css_list */
182 struct list_head rstat_css_node;
183
184 /*
185 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
186 * matching css can be looked up using css_from_id().
187 */
188 int id;
189
190 unsigned int flags;
191
192 /*
193 * Monotonically increasing unique serial number which defines a
194 * uniform order among all csses. It's guaranteed that all
195 * ->children lists are in the ascending order of ->serial_nr and
196 * used to allow interrupting and resuming iterations.
197 */
198 u64 serial_nr;
199
200 /*
201 * Incremented by online self and children. Used to guarantee that
202 * parents are not offlined before their children.
203 */
204 atomic_t online_cnt;
205
206 /* percpu_ref killing and RCU release */
207 struct work_struct destroy_work;
208 struct rcu_work destroy_rwork;
209
210 /*
211 * PI: the parent css. Placed here for cache proximity to following
212 * fields of the containing structure.
213 */
214 struct cgroup_subsys_state *parent;
215
216 /*
217 * Keep track of total numbers of visible descendant CSSes.
218 * The total number of dying CSSes is tracked in
219 * css->cgroup->nr_dying_subsys[ssid].
220 * Protected by cgroup_mutex.
221 */
222 int nr_descendants;
223
224 ANDROID_BACKPORT_RESERVE(1);
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 ANDROID_BACKPORT_RESERVE(1);
324 };
325
326 struct cgroup_base_stat {
327 struct task_cputime cputime;
328
329 #ifdef CONFIG_SCHED_CORE
330 u64 forceidle_sum;
331 #endif
332 u64 ntime;
333 };
334
335 /*
336 * rstat - cgroup scalable recursive statistics. Accounting is done
337 * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
338 * hierarchy on reads.
339 *
340 * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
341 * linked into the updated tree. On the following read, propagation only
342 * considers and consumes the updated tree. This makes reading O(the
343 * number of descendants which have been active since last read) instead of
344 * O(the total number of descendants).
345 *
346 * This is important because there can be a lot of (draining) cgroups which
347 * aren't active and stat may be read frequently. The combination can
348 * become very expensive. By propagating selectively, increasing reading
349 * frequency decreases the cost of each read.
350 *
351 * This struct hosts both the fields which implement the above -
352 * updated_children and updated_next - and the fields which track basic
353 * resource statistics on top of it - bsync, bstat and last_bstat.
354 */
355 struct cgroup_rstat_cpu {
356 /*
357 * ->bsync protects ->bstat. These are the only fields which get
358 * updated in the hot path.
359 */
360 struct u64_stats_sync bsync;
361 struct cgroup_base_stat bstat;
362
363 /*
364 * Snapshots at the last reading. These are used to calculate the
365 * deltas to propagate to the global counters.
366 */
367 struct cgroup_base_stat last_bstat;
368
369 /*
370 * This field is used to record the cumulative per-cpu time of
371 * the cgroup and its descendants. Currently it can be read via
372 * eBPF/drgn etc, and we are still trying to determine how to
373 * expose it in the cgroupfs interface.
374 */
375 struct cgroup_base_stat subtree_bstat;
376
377 /*
378 * Snapshots at the last reading. These are used to calculate the
379 * deltas to propagate to the per-cpu subtree_bstat.
380 */
381 struct cgroup_base_stat last_subtree_bstat;
382
383 /*
384 * Child cgroups with stat updates on this cpu since the last read
385 * are linked on the parent's ->updated_children through
386 * ->updated_next.
387 *
388 * In addition to being more compact, singly-linked list pointing
389 * to the cgroup makes it unnecessary for each per-cpu struct to
390 * point back to the associated cgroup.
391 *
392 * Protected by per-cpu cgroup_rstat_cpu_lock.
393 */
394 struct cgroup *updated_children; /* terminated by self cgroup */
395 struct cgroup *updated_next; /* NULL iff not on the list */
396 };
397
398 struct cgroup_freezer_state {
399 /* Should the cgroup and its descendants be frozen. */
400 bool freeze;
401
402 /* Should the cgroup actually be frozen? */
403 int e_freeze;
404
405 /* Fields below are protected by css_set_lock */
406
407 /* Number of frozen descendant cgroups */
408 int nr_frozen_descendants;
409
410 /*
411 * Number of tasks, which are counted as frozen:
412 * frozen, SIGSTOPped, and PTRACEd.
413 */
414 int nr_frozen_tasks;
415
416 };
417
418 /**
419 * struct cgroup_kmi_ext_info is meant to hold extensions to struct cgroup while
420 * maintaining KMI stability. This type is meant to be opaque to vendor modules.
421 */
422 struct cgroup_kmi_ext_info {
423 /*
424 * Metadata for cgroup v2 freeze time. Writes protected
425 * by css_set_lock.
426 */
427 struct {
428 /* Freeze time data consistency protection */
429 seqcount_t freeze_seq;
430
431 /*
432 * Most recent time the cgroup was requested to freeze.
433 * Accesses guarded by freeze_seq counter. Writes serialized
434 * by css_set_lock.
435 */
436 u64 freeze_start_nsec;
437
438 /*
439 * Total duration the cgroup has spent freezing.
440 * Accesses guarded by freeze_seq counter. Writes serialized
441 * by css_set_lock.
442 */
443 u64 frozen_nsec;
444 } freezer;
445 };
446 /*
447 * Hide the definition of struct cgroup_kmi_ext_info from the ABI so that it
448 * can be modified to accommodate additional backports in the future.
449 */
450 ANDROID_KABI_DECLONLY(cgroup_kmi_ext_info);
451
452 struct cgroup {
453 /* self css with NULL ->ss, points back to this cgroup */
454 struct cgroup_subsys_state self;
455
456 unsigned long flags; /* "unsigned long" so bitops work */
457
458 /*
459 * The depth this cgroup is at. The root is at depth zero and each
460 * step down the hierarchy increments the level. This along with
461 * ancestors[] can determine whether a given cgroup is a
462 * descendant of another without traversing the hierarchy.
463 */
464 int level;
465
466 /* Maximum allowed descent tree depth */
467 int max_depth;
468
469 /*
470 * Keep track of total numbers of visible and dying descent cgroups.
471 * Dying cgroups are cgroups which were deleted by a user,
472 * but are still existing because someone else is holding a reference.
473 * max_descendants is a maximum allowed number of descent cgroups.
474 *
475 * nr_descendants and nr_dying_descendants are protected
476 * by cgroup_mutex and css_set_lock. It's fine to read them holding
477 * any of cgroup_mutex and css_set_lock; for writing both locks
478 * should be held.
479 */
480 int nr_descendants;
481 int nr_dying_descendants;
482 int max_descendants;
483
484 /*
485 * Each non-empty css_set associated with this cgroup contributes
486 * one to nr_populated_csets. The counter is zero iff this cgroup
487 * doesn't have any tasks.
488 *
489 * All children which have non-zero nr_populated_csets and/or
490 * nr_populated_children of their own contribute one to either
491 * nr_populated_domain_children or nr_populated_threaded_children
492 * depending on their type. Each counter is zero iff all cgroups
493 * of the type in the subtree proper don't have any tasks.
494 */
495 int nr_populated_csets;
496 int nr_populated_domain_children;
497 int nr_populated_threaded_children;
498
499 int nr_threaded_children; /* # of live threaded child cgroups */
500
501 /* sequence number for cgroup.kill, serialized by css_set_lock. */
502 unsigned int kill_seq;
503
504 struct kernfs_node *kn; /* cgroup kernfs entry */
505 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
506 struct cgroup_file events_file; /* handle for "cgroup.events" */
507
508 /* handles for "{cpu,memory,io,irq}.pressure" */
509 struct cgroup_file psi_files[NR_PSI_RESOURCES];
510
511 /*
512 * The bitmask of subsystems enabled on the child cgroups.
513 * ->subtree_control is the one configured through
514 * "cgroup.subtree_control" while ->subtree_ss_mask is the effective
515 * one which may have more subsystems enabled. Controller knobs
516 * are made available iff it's enabled in ->subtree_control.
517 */
518 u16 subtree_control;
519 u16 subtree_ss_mask;
520 u16 old_subtree_control;
521 u16 old_subtree_ss_mask;
522
523 /* Private pointers for each registered subsystem */
524 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
525
526 /*
527 * Keep track of total number of dying CSSes at and below this cgroup.
528 * Protected by cgroup_mutex.
529 */
530 int nr_dying_subsys[CGROUP_SUBSYS_COUNT];
531
532 struct cgroup_root *root;
533
534 /*
535 * List of cgrp_cset_links pointing at css_sets with tasks in this
536 * cgroup. Protected by css_set_lock.
537 */
538 struct list_head cset_links;
539
540 /*
541 * On the default hierarchy, a css_set for a cgroup with some
542 * susbsys disabled will point to css's which are associated with
543 * the closest ancestor which has the subsys enabled. The
544 * following lists all css_sets which point to this cgroup's css
545 * for the given subsystem.
546 */
547 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
548
549 /*
550 * If !threaded, self. If threaded, it points to the nearest
551 * domain ancestor. Inside a threaded subtree, cgroups are exempt
552 * from process granularity and no-internal-task constraint.
553 * Domain level resource consumptions which aren't tied to a
554 * specific task are charged to the dom_cgrp.
555 */
556 struct cgroup *dom_cgrp;
557 struct cgroup *old_dom_cgrp; /* used while enabling threaded */
558
559 /* per-cpu recursive resource statistics */
560 struct cgroup_rstat_cpu __percpu *rstat_cpu;
561 struct list_head rstat_css_list;
562
563 /*
564 * Add padding to separate the read mostly rstat_cpu and
565 * rstat_css_list into a different cacheline from the following
566 * rstat_flush_next and *bstat fields which can have frequent updates.
567 */
568 CACHELINE_PADDING(_pad_);
569
570 /*
571 * A singly-linked list of cgroup structures to be rstat flushed.
572 * This is a scratch field to be used exclusively by
573 * cgroup_rstat_flush_locked() and protected by cgroup_rstat_lock.
574 */
575 struct cgroup *rstat_flush_next;
576
577 /* cgroup basic resource statistics */
578 struct cgroup_base_stat last_bstat;
579 struct cgroup_base_stat bstat;
580 struct prev_cputime prev_cputime; /* for printing out cputime */
581
582 /*
583 * list of pidlists, up to two for each namespace (one for procs, one
584 * for tasks); created on demand.
585 */
586 struct list_head pidlists;
587 struct mutex pidlist_mutex;
588
589 /* used to wait for offlining of csses */
590 wait_queue_head_t offline_waitq;
591
592 /* used to schedule release agent */
593 struct work_struct release_agent_work;
594
595 /* used to track pressure stalls */
596 struct psi_group *psi;
597
598 /* used to store eBPF programs */
599 struct cgroup_bpf bpf;
600
601 /* Used to store internal freezer state */
602 struct cgroup_freezer_state freezer;
603
604 #ifdef CONFIG_BPF_SYSCALL
605 struct bpf_local_storage __rcu *bpf_cgrp_storage;
606 #endif
607
608 /*
609 * Used to store KMI-compliant extensions to struct cgroup.
610 * For further additions, modify the definition for struct
611 * cgroup_kmi_ext_info.
612 */
613 ANDROID_BACKPORT_USE(1, struct cgroup_kmi_ext_info *kmi_ext_info);
614
615 /* All ancestors including self */
616 struct cgroup *ancestors[];
617 };
618
619 /*
620 * A cgroup_root represents the root of a cgroup hierarchy, and may be
621 * associated with a kernfs_root to form an active hierarchy. This is
622 * internal to cgroup core. Don't access directly from controllers.
623 */
624 struct cgroup_root {
625 struct kernfs_root *kf_root;
626
627 /* The bitmask of subsystems attached to this hierarchy */
628 unsigned int subsys_mask;
629
630 /* Unique id for this hierarchy. */
631 int hierarchy_id;
632
633 /* A list running through the active hierarchies */
634 struct list_head root_list;
635 struct rcu_head rcu; /* Must be near the top */
636
637 /*
638 * The root cgroup. The containing cgroup_root will be destroyed on its
639 * release. cgrp->ancestors[0] will be used overflowing into the
640 * following field. cgrp_ancestor_storage must immediately follow.
641 */
642 struct cgroup cgrp;
643
644 /* must follow cgrp for cgrp->ancestors[0], see above */
645 struct cgroup *cgrp_ancestor_storage;
646
647 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
648 atomic_t nr_cgrps;
649
650 /* Hierarchy-specific flags */
651 unsigned int flags;
652
653 /* The path to use for release notifications. */
654 char release_agent_path[PATH_MAX];
655
656 /* The name for this hierarchy - may be empty */
657 char name[MAX_CGROUP_ROOT_NAMELEN];
658
659 ANDROID_BACKPORT_RESERVE(1);
660 };
661
662 /*
663 * struct cftype: handler definitions for cgroup control files
664 *
665 * When reading/writing to a file:
666 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
667 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
668 */
669 struct cftype {
670 /*
671 * By convention, the name should begin with the name of the
672 * subsystem, followed by a period. Zero length string indicates
673 * end of cftype array.
674 */
675 char name[MAX_CFTYPE_NAME];
676 unsigned long private;
677
678 /*
679 * The maximum length of string, excluding trailing nul, that can
680 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
681 */
682 size_t max_write_len;
683
684 /* CFTYPE_* flags */
685 unsigned int flags;
686
687 /*
688 * If non-zero, should contain the offset from the start of css to
689 * a struct cgroup_file field. cgroup will record the handle of
690 * the created file into it. The recorded handle can be used as
691 * long as the containing css remains accessible.
692 */
693 unsigned int file_offset;
694
695 /*
696 * Fields used for internal bookkeeping. Initialized automatically
697 * during registration.
698 */
699 struct cgroup_subsys *ss; /* NULL for cgroup core files */
700 struct list_head node; /* anchored at ss->cfts */
701 struct kernfs_ops *kf_ops;
702
703 int (*open)(struct kernfs_open_file *of);
704 void (*release)(struct kernfs_open_file *of);
705
706 /*
707 * read_u64() is a shortcut for the common case of returning a
708 * single integer. Use it in place of read()
709 */
710 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
711 /*
712 * read_s64() is a signed version of read_u64()
713 */
714 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
715
716 /* generic seq_file read interface */
717 int (*seq_show)(struct seq_file *sf, void *v);
718
719 /* optional ops, implement all or none */
720 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
721 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
722 void (*seq_stop)(struct seq_file *sf, void *v);
723
724 /*
725 * write_u64() is a shortcut for the common case of accepting
726 * a single integer (as parsed by simple_strtoull) from
727 * userspace. Use in place of write(); return 0 or error.
728 */
729 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
730 u64 val);
731 /*
732 * write_s64() is a signed version of write_u64()
733 */
734 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
735 s64 val);
736
737 /*
738 * write() is the generic write callback which maps directly to
739 * kernfs write operation and overrides all other operations.
740 * Maximum write size is determined by ->max_write_len. Use
741 * of_css/cft() to access the associated css and cft.
742 */
743 ssize_t (*write)(struct kernfs_open_file *of,
744 char *buf, size_t nbytes, loff_t off);
745
746 __poll_t (*poll)(struct kernfs_open_file *of,
747 struct poll_table_struct *pt);
748
749 struct lock_class_key lockdep_key;
750 };
751
752 /*
753 * Control Group subsystem type.
754 * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
755 */
756 struct cgroup_subsys {
757 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
758 int (*css_online)(struct cgroup_subsys_state *css);
759 void (*css_offline)(struct cgroup_subsys_state *css);
760 void (*css_released)(struct cgroup_subsys_state *css);
761 void (*css_free)(struct cgroup_subsys_state *css);
762 void (*css_reset)(struct cgroup_subsys_state *css);
763 void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
764 int (*css_extra_stat_show)(struct seq_file *seq,
765 struct cgroup_subsys_state *css);
766 int (*css_local_stat_show)(struct seq_file *seq,
767 struct cgroup_subsys_state *css);
768
769 int (*can_attach)(struct cgroup_taskset *tset);
770 void (*cancel_attach)(struct cgroup_taskset *tset);
771 void (*attach)(struct cgroup_taskset *tset);
772 void (*post_attach)(void);
773 int (*can_fork)(struct task_struct *task,
774 struct css_set *cset);
775 void (*cancel_fork)(struct task_struct *task, struct css_set *cset);
776 void (*fork)(struct task_struct *task);
777 void (*exit)(struct task_struct *task);
778 void (*release)(struct task_struct *task);
779 void (*bind)(struct cgroup_subsys_state *root_css);
780
781 bool early_init:1;
782
783 /*
784 * If %true, the controller, on the default hierarchy, doesn't show
785 * up in "cgroup.controllers" or "cgroup.subtree_control", is
786 * implicitly enabled on all cgroups on the default hierarchy, and
787 * bypasses the "no internal process" constraint. This is for
788 * utility type controllers which is transparent to userland.
789 *
790 * An implicit controller can be stolen from the default hierarchy
791 * anytime and thus must be okay with offline csses from previous
792 * hierarchies coexisting with csses for the current one.
793 */
794 bool implicit_on_dfl:1;
795
796 /*
797 * If %true, the controller, supports threaded mode on the default
798 * hierarchy. In a threaded subtree, both process granularity and
799 * no-internal-process constraint are ignored and a threaded
800 * controllers should be able to handle that.
801 *
802 * Note that as an implicit controller is automatically enabled on
803 * all cgroups on the default hierarchy, it should also be
804 * threaded. implicit && !threaded is not supported.
805 */
806 bool threaded:1;
807
808 /* the following two fields are initialized automatically during boot */
809 int id;
810 const char *name;
811
812 /* optional, initialized automatically during boot if not set */
813 const char *legacy_name;
814
815 /* link to parent, protected by cgroup_lock() */
816 struct cgroup_root *root;
817
818 /* idr for css->id */
819 struct idr css_idr;
820
821 /*
822 * List of cftypes. Each entry is the first entry of an array
823 * terminated by zero length name.
824 */
825 struct list_head cfts;
826
827 /*
828 * Base cftypes which are automatically registered. The two can
829 * point to the same array.
830 */
831 struct cftype *dfl_cftypes; /* for the default hierarchy */
832 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
833
834 /*
835 * A subsystem may depend on other subsystems. When such subsystem
836 * is enabled on a cgroup, the depended-upon subsystems are enabled
837 * together if available. Subsystems enabled due to dependency are
838 * not visible to userland until explicitly enabled. The following
839 * specifies the mask of subsystems that this one depends on.
840 */
841 unsigned int depends_on;
842
843 ANDROID_BACKPORT_RESERVE(1);
844 };
845
846 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
847
848 struct cgroup_of_peak {
849 unsigned long value;
850 struct list_head list;
851 };
852
853 /**
854 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
855 * @tsk: target task
856 *
857 * Allows cgroup operations to synchronize against threadgroup changes
858 * using a percpu_rw_semaphore.
859 */
cgroup_threadgroup_change_begin(struct task_struct * tsk)860 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
861 {
862 percpu_down_read(&cgroup_threadgroup_rwsem);
863 }
864
865 /**
866 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
867 * @tsk: target task
868 *
869 * Counterpart of cgroup_threadcgroup_change_begin().
870 */
cgroup_threadgroup_change_end(struct task_struct * tsk)871 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
872 {
873 percpu_up_read(&cgroup_threadgroup_rwsem);
874 }
875
876 #else /* CONFIG_CGROUPS */
877
878 #define CGROUP_SUBSYS_COUNT 0
879
cgroup_threadgroup_change_begin(struct task_struct * tsk)880 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
881 {
882 might_sleep();
883 }
884
cgroup_threadgroup_change_end(struct task_struct * tsk)885 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
886
887 #endif /* CONFIG_CGROUPS */
888
889 #ifdef CONFIG_SOCK_CGROUP_DATA
890
891 /*
892 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
893 * per-socket cgroup information except for memcg association.
894 *
895 * On legacy hierarchies, net_prio and net_cls controllers directly
896 * set attributes on each sock which can then be tested by the network
897 * layer. On the default hierarchy, each sock is associated with the
898 * cgroup it was created in and the networking layer can match the
899 * cgroup directly.
900 */
901 struct sock_cgroup_data {
902 struct cgroup *cgroup; /* v2 */
903 #ifdef CONFIG_CGROUP_NET_CLASSID
904 u32 classid; /* v1 */
905 #endif
906 #ifdef CONFIG_CGROUP_NET_PRIO
907 u16 prioidx; /* v1 */
908 #endif
909 };
910
sock_cgroup_prioidx(const struct sock_cgroup_data * skcd)911 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
912 {
913 #ifdef CONFIG_CGROUP_NET_PRIO
914 return READ_ONCE(skcd->prioidx);
915 #else
916 return 1;
917 #endif
918 }
919
sock_cgroup_classid(const struct sock_cgroup_data * skcd)920 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
921 {
922 #ifdef CONFIG_CGROUP_NET_CLASSID
923 return READ_ONCE(skcd->classid);
924 #else
925 return 0;
926 #endif
927 }
928
sock_cgroup_set_prioidx(struct sock_cgroup_data * skcd,u16 prioidx)929 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
930 u16 prioidx)
931 {
932 #ifdef CONFIG_CGROUP_NET_PRIO
933 WRITE_ONCE(skcd->prioidx, prioidx);
934 #endif
935 }
936
sock_cgroup_set_classid(struct sock_cgroup_data * skcd,u32 classid)937 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
938 u32 classid)
939 {
940 #ifdef CONFIG_CGROUP_NET_CLASSID
941 WRITE_ONCE(skcd->classid, classid);
942 #endif
943 }
944
945 #else /* CONFIG_SOCK_CGROUP_DATA */
946
947 struct sock_cgroup_data {
948 };
949
950 #endif /* CONFIG_SOCK_CGROUP_DATA */
951
952 #endif /* _LINUX_CGROUP_DEFS_H */
953