]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blob - include/linux/cgroup-defs.h
Merge tag 'pci-v4.20-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaa...
[mirror_ubuntu-focal-kernel.git] / include / linux / cgroup-defs.h
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
24 #ifdef CONFIG_CGROUPS
25
26 struct cgroup;
27 struct cgroup_root;
28 struct cgroup_subsys;
29 struct cgroup_taskset;
30 struct kernfs_node;
31 struct kernfs_ops;
32 struct kernfs_open_file;
33 struct seq_file;
34
35 #define MAX_CGROUP_TYPE_NAMELEN 32
36 #define MAX_CGROUP_ROOT_NAMELEN 64
37 #define MAX_CFTYPE_NAME 64
38
39 /* define the enumeration of all cgroup subsystems */
40 #define SUBSYS(_x) _x ## _cgrp_id,
41 enum cgroup_subsys_id {
42 #include <linux/cgroup_subsys.h>
43 CGROUP_SUBSYS_COUNT,
44 };
45 #undef SUBSYS
46
47 /* bits in struct cgroup_subsys_state flags field */
48 enum {
49 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
50 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
51 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
52 CSS_VISIBLE = (1 << 3), /* css is visible to userland */
53 CSS_DYING = (1 << 4), /* css is dying */
54 };
55
56 /* bits in struct cgroup flags field */
57 enum {
58 /* Control Group requires release notifications to userspace */
59 CGRP_NOTIFY_ON_RELEASE,
60 /*
61 * Clone the parent's configuration when creating a new child
62 * cpuset cgroup. For historical reasons, this option can be
63 * specified at mount time and thus is implemented here.
64 */
65 CGRP_CPUSET_CLONE_CHILDREN,
66 };
67
68 /* cgroup_root->flags */
69 enum {
70 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
71 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
72
73 /*
74 * Consider namespaces as delegation boundaries. If this flag is
75 * set, controller specific interface files in a namespace root
76 * aren't writeable from inside the namespace.
77 */
78 CGRP_ROOT_NS_DELEGATE = (1 << 3),
79
80 /*
81 * Enable cpuset controller in v1 cgroup to use v2 behavior.
82 */
83 CGRP_ROOT_CPUSET_V2_MODE = (1 << 4),
84 };
85
86 /* cftype->flags */
87 enum {
88 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
89 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
90 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
91
92 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
93 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
94
95 /* internal flags, do not use outside cgroup core proper */
96 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
97 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
98 };
99
100 /*
101 * cgroup_file is the handle for a file instance created in a cgroup which
102 * is used, for example, to generate file changed notifications. This can
103 * be obtained by setting cftype->file_offset.
104 */
105 struct cgroup_file {
106 /* do not access any fields from outside cgroup core */
107 struct kernfs_node *kn;
108 unsigned long notified_at;
109 struct timer_list notify_timer;
110 };
111
112 /*
113 * Per-subsystem/per-cgroup state maintained by the system. This is the
114 * fundamental structural building block that controllers deal with.
115 *
116 * Fields marked with "PI:" are public and immutable and may be accessed
117 * directly without synchronization.
118 */
119 struct cgroup_subsys_state {
120 /* PI: the cgroup that this css is attached to */
121 struct cgroup *cgroup;
122
123 /* PI: the cgroup subsystem that this css is attached to */
124 struct cgroup_subsys *ss;
125
126 /* reference count - access via css_[try]get() and css_put() */
127 struct percpu_ref refcnt;
128
129 /* siblings list anchored at the parent's ->children */
130 struct list_head sibling;
131 struct list_head children;
132
133 /* flush target list anchored at cgrp->rstat_css_list */
134 struct list_head rstat_css_node;
135
136 /*
137 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
138 * matching css can be looked up using css_from_id().
139 */
140 int id;
141
142 unsigned int flags;
143
144 /*
145 * Monotonically increasing unique serial number which defines a
146 * uniform order among all csses. It's guaranteed that all
147 * ->children lists are in the ascending order of ->serial_nr and
148 * used to allow interrupting and resuming iterations.
149 */
150 u64 serial_nr;
151
152 /*
153 * Incremented by online self and children. Used to guarantee that
154 * parents are not offlined before their children.
155 */
156 atomic_t online_cnt;
157
158 /* percpu_ref killing and RCU release */
159 struct work_struct destroy_work;
160 struct rcu_work destroy_rwork;
161
162 /*
163 * PI: the parent css. Placed here for cache proximity to following
164 * fields of the containing structure.
165 */
166 struct cgroup_subsys_state *parent;
167 };
168
169 /*
170 * A css_set is a structure holding pointers to a set of
171 * cgroup_subsys_state objects. This saves space in the task struct
172 * object and speeds up fork()/exit(), since a single inc/dec and a
173 * list_add()/del() can bump the reference count on the entire cgroup
174 * set for a task.
175 */
176 struct css_set {
177 /*
178 * Set of subsystem states, one for each subsystem. This array is
179 * immutable after creation apart from the init_css_set during
180 * subsystem registration (at boot time).
181 */
182 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
183
184 /* reference count */
185 refcount_t refcount;
186
187 /*
188 * For a domain cgroup, the following points to self. If threaded,
189 * to the matching cset of the nearest domain ancestor. The
190 * dom_cset provides access to the domain cgroup and its csses to
191 * which domain level resource consumptions should be charged.
192 */
193 struct css_set *dom_cset;
194
195 /* the default cgroup associated with this css_set */
196 struct cgroup *dfl_cgrp;
197
198 /* internal task count, protected by css_set_lock */
199 int nr_tasks;
200
201 /*
202 * Lists running through all tasks using this cgroup group.
203 * mg_tasks lists tasks which belong to this cset but are in the
204 * process of being migrated out or in. Protected by
205 * css_set_rwsem, but, during migration, once tasks are moved to
206 * mg_tasks, it can be read safely while holding cgroup_mutex.
207 */
208 struct list_head tasks;
209 struct list_head mg_tasks;
210
211 /* all css_task_iters currently walking this cset */
212 struct list_head task_iters;
213
214 /*
215 * On the default hierarhcy, ->subsys[ssid] may point to a css
216 * attached to an ancestor instead of the cgroup this css_set is
217 * associated with. The following node is anchored at
218 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
219 * iterate through all css's attached to a given cgroup.
220 */
221 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
222
223 /* all threaded csets whose ->dom_cset points to this cset */
224 struct list_head threaded_csets;
225 struct list_head threaded_csets_node;
226
227 /*
228 * List running through all cgroup groups in the same hash
229 * slot. Protected by css_set_lock
230 */
231 struct hlist_node hlist;
232
233 /*
234 * List of cgrp_cset_links pointing at cgroups referenced from this
235 * css_set. Protected by css_set_lock.
236 */
237 struct list_head cgrp_links;
238
239 /*
240 * List of csets participating in the on-going migration either as
241 * source or destination. Protected by cgroup_mutex.
242 */
243 struct list_head mg_preload_node;
244 struct list_head mg_node;
245
246 /*
247 * If this cset is acting as the source of migration the following
248 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
249 * respectively the source and destination cgroups of the on-going
250 * migration. mg_dst_cset is the destination cset the target tasks
251 * on this cset should be migrated to. Protected by cgroup_mutex.
252 */
253 struct cgroup *mg_src_cgrp;
254 struct cgroup *mg_dst_cgrp;
255 struct css_set *mg_dst_cset;
256
257 /* dead and being drained, ignore for migration */
258 bool dead;
259
260 /* For RCU-protected deletion */
261 struct rcu_head rcu_head;
262 };
263
264 struct cgroup_base_stat {
265 struct task_cputime cputime;
266 };
267
268 /*
269 * rstat - cgroup scalable recursive statistics. Accounting is done
270 * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
271 * hierarchy on reads.
272 *
273 * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
274 * linked into the updated tree. On the following read, propagation only
275 * considers and consumes the updated tree. This makes reading O(the
276 * number of descendants which have been active since last read) instead of
277 * O(the total number of descendants).
278 *
279 * This is important because there can be a lot of (draining) cgroups which
280 * aren't active and stat may be read frequently. The combination can
281 * become very expensive. By propagating selectively, increasing reading
282 * frequency decreases the cost of each read.
283 *
284 * This struct hosts both the fields which implement the above -
285 * updated_children and updated_next - and the fields which track basic
286 * resource statistics on top of it - bsync, bstat and last_bstat.
287 */
288 struct cgroup_rstat_cpu {
289 /*
290 * ->bsync protects ->bstat. These are the only fields which get
291 * updated in the hot path.
292 */
293 struct u64_stats_sync bsync;
294 struct cgroup_base_stat bstat;
295
296 /*
297 * Snapshots at the last reading. These are used to calculate the
298 * deltas to propagate to the global counters.
299 */
300 struct cgroup_base_stat last_bstat;
301
302 /*
303 * Child cgroups with stat updates on this cpu since the last read
304 * are linked on the parent's ->updated_children through
305 * ->updated_next.
306 *
307 * In addition to being more compact, singly-linked list pointing
308 * to the cgroup makes it unnecessary for each per-cpu struct to
309 * point back to the associated cgroup.
310 *
311 * Protected by per-cpu cgroup_rstat_cpu_lock.
312 */
313 struct cgroup *updated_children; /* terminated by self cgroup */
314 struct cgroup *updated_next; /* NULL iff not on the list */
315 };
316
317 struct cgroup {
318 /* self css with NULL ->ss, points back to this cgroup */
319 struct cgroup_subsys_state self;
320
321 unsigned long flags; /* "unsigned long" so bitops work */
322
323 /*
324 * idr allocated in-hierarchy ID.
325 *
326 * ID 0 is not used, the ID of the root cgroup is always 1, and a
327 * new cgroup will be assigned with a smallest available ID.
328 *
329 * Allocating/Removing ID must be protected by cgroup_mutex.
330 */
331 int id;
332
333 /*
334 * The depth this cgroup is at. The root is at depth zero and each
335 * step down the hierarchy increments the level. This along with
336 * ancestor_ids[] can determine whether a given cgroup is a
337 * descendant of another without traversing the hierarchy.
338 */
339 int level;
340
341 /* Maximum allowed descent tree depth */
342 int max_depth;
343
344 /*
345 * Keep track of total numbers of visible and dying descent cgroups.
346 * Dying cgroups are cgroups which were deleted by a user,
347 * but are still existing because someone else is holding a reference.
348 * max_descendants is a maximum allowed number of descent cgroups.
349 */
350 int nr_descendants;
351 int nr_dying_descendants;
352 int max_descendants;
353
354 /*
355 * Each non-empty css_set associated with this cgroup contributes
356 * one to nr_populated_csets. The counter is zero iff this cgroup
357 * doesn't have any tasks.
358 *
359 * All children which have non-zero nr_populated_csets and/or
360 * nr_populated_children of their own contribute one to either
361 * nr_populated_domain_children or nr_populated_threaded_children
362 * depending on their type. Each counter is zero iff all cgroups
363 * of the type in the subtree proper don't have any tasks.
364 */
365 int nr_populated_csets;
366 int nr_populated_domain_children;
367 int nr_populated_threaded_children;
368
369 int nr_threaded_children; /* # of live threaded child cgroups */
370
371 struct kernfs_node *kn; /* cgroup kernfs entry */
372 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
373 struct cgroup_file events_file; /* handle for "cgroup.events" */
374
375 /*
376 * The bitmask of subsystems enabled on the child cgroups.
377 * ->subtree_control is the one configured through
378 * "cgroup.subtree_control" while ->child_ss_mask is the effective
379 * one which may have more subsystems enabled. Controller knobs
380 * are made available iff it's enabled in ->subtree_control.
381 */
382 u16 subtree_control;
383 u16 subtree_ss_mask;
384 u16 old_subtree_control;
385 u16 old_subtree_ss_mask;
386
387 /* Private pointers for each registered subsystem */
388 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
389
390 struct cgroup_root *root;
391
392 /*
393 * List of cgrp_cset_links pointing at css_sets with tasks in this
394 * cgroup. Protected by css_set_lock.
395 */
396 struct list_head cset_links;
397
398 /*
399 * On the default hierarchy, a css_set for a cgroup with some
400 * susbsys disabled will point to css's which are associated with
401 * the closest ancestor which has the subsys enabled. The
402 * following lists all css_sets which point to this cgroup's css
403 * for the given subsystem.
404 */
405 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
406
407 /*
408 * If !threaded, self. If threaded, it points to the nearest
409 * domain ancestor. Inside a threaded subtree, cgroups are exempt
410 * from process granularity and no-internal-task constraint.
411 * Domain level resource consumptions which aren't tied to a
412 * specific task are charged to the dom_cgrp.
413 */
414 struct cgroup *dom_cgrp;
415 struct cgroup *old_dom_cgrp; /* used while enabling threaded */
416
417 /* per-cpu recursive resource statistics */
418 struct cgroup_rstat_cpu __percpu *rstat_cpu;
419 struct list_head rstat_css_list;
420
421 /* cgroup basic resource statistics */
422 struct cgroup_base_stat pending_bstat; /* pending from children */
423 struct cgroup_base_stat bstat;
424 struct prev_cputime prev_cputime; /* for printing out cputime */
425
426 /*
427 * list of pidlists, up to two for each namespace (one for procs, one
428 * for tasks); created on demand.
429 */
430 struct list_head pidlists;
431 struct mutex pidlist_mutex;
432
433 /* used to wait for offlining of csses */
434 wait_queue_head_t offline_waitq;
435
436 /* used to schedule release agent */
437 struct work_struct release_agent_work;
438
439 /* used to store eBPF programs */
440 struct cgroup_bpf bpf;
441
442 /* If there is block congestion on this cgroup. */
443 atomic_t congestion_count;
444
445 /* ids of the ancestors at each level including self */
446 int ancestor_ids[];
447 };
448
449 /*
450 * A cgroup_root represents the root of a cgroup hierarchy, and may be
451 * associated with a kernfs_root to form an active hierarchy. This is
452 * internal to cgroup core. Don't access directly from controllers.
453 */
454 struct cgroup_root {
455 struct kernfs_root *kf_root;
456
457 /* The bitmask of subsystems attached to this hierarchy */
458 unsigned int subsys_mask;
459
460 /* Unique id for this hierarchy. */
461 int hierarchy_id;
462
463 /* The root cgroup. Root is destroyed on its release. */
464 struct cgroup cgrp;
465
466 /* for cgrp->ancestor_ids[0] */
467 int cgrp_ancestor_id_storage;
468
469 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
470 atomic_t nr_cgrps;
471
472 /* A list running through the active hierarchies */
473 struct list_head root_list;
474
475 /* Hierarchy-specific flags */
476 unsigned int flags;
477
478 /* IDs for cgroups in this hierarchy */
479 struct idr cgroup_idr;
480
481 /* The path to use for release notifications. */
482 char release_agent_path[PATH_MAX];
483
484 /* The name for this hierarchy - may be empty */
485 char name[MAX_CGROUP_ROOT_NAMELEN];
486 };
487
488 /*
489 * struct cftype: handler definitions for cgroup control files
490 *
491 * When reading/writing to a file:
492 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
493 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
494 */
495 struct cftype {
496 /*
497 * By convention, the name should begin with the name of the
498 * subsystem, followed by a period. Zero length string indicates
499 * end of cftype array.
500 */
501 char name[MAX_CFTYPE_NAME];
502 unsigned long private;
503
504 /*
505 * The maximum length of string, excluding trailing nul, that can
506 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
507 */
508 size_t max_write_len;
509
510 /* CFTYPE_* flags */
511 unsigned int flags;
512
513 /*
514 * If non-zero, should contain the offset from the start of css to
515 * a struct cgroup_file field. cgroup will record the handle of
516 * the created file into it. The recorded handle can be used as
517 * long as the containing css remains accessible.
518 */
519 unsigned int file_offset;
520
521 /*
522 * Fields used for internal bookkeeping. Initialized automatically
523 * during registration.
524 */
525 struct cgroup_subsys *ss; /* NULL for cgroup core files */
526 struct list_head node; /* anchored at ss->cfts */
527 struct kernfs_ops *kf_ops;
528
529 int (*open)(struct kernfs_open_file *of);
530 void (*release)(struct kernfs_open_file *of);
531
532 /*
533 * read_u64() is a shortcut for the common case of returning a
534 * single integer. Use it in place of read()
535 */
536 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
537 /*
538 * read_s64() is a signed version of read_u64()
539 */
540 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
541
542 /* generic seq_file read interface */
543 int (*seq_show)(struct seq_file *sf, void *v);
544
545 /* optional ops, implement all or none */
546 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
547 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
548 void (*seq_stop)(struct seq_file *sf, void *v);
549
550 /*
551 * write_u64() is a shortcut for the common case of accepting
552 * a single integer (as parsed by simple_strtoull) from
553 * userspace. Use in place of write(); return 0 or error.
554 */
555 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
556 u64 val);
557 /*
558 * write_s64() is a signed version of write_u64()
559 */
560 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
561 s64 val);
562
563 /*
564 * write() is the generic write callback which maps directly to
565 * kernfs write operation and overrides all other operations.
566 * Maximum write size is determined by ->max_write_len. Use
567 * of_css/cft() to access the associated css and cft.
568 */
569 ssize_t (*write)(struct kernfs_open_file *of,
570 char *buf, size_t nbytes, loff_t off);
571
572 #ifdef CONFIG_DEBUG_LOCK_ALLOC
573 struct lock_class_key lockdep_key;
574 #endif
575 };
576
577 /*
578 * Control Group subsystem type.
579 * See Documentation/cgroup-v1/cgroups.txt for details
580 */
581 struct cgroup_subsys {
582 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
583 int (*css_online)(struct cgroup_subsys_state *css);
584 void (*css_offline)(struct cgroup_subsys_state *css);
585 void (*css_released)(struct cgroup_subsys_state *css);
586 void (*css_free)(struct cgroup_subsys_state *css);
587 void (*css_reset)(struct cgroup_subsys_state *css);
588 void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
589 int (*css_extra_stat_show)(struct seq_file *seq,
590 struct cgroup_subsys_state *css);
591
592 int (*can_attach)(struct cgroup_taskset *tset);
593 void (*cancel_attach)(struct cgroup_taskset *tset);
594 void (*attach)(struct cgroup_taskset *tset);
595 void (*post_attach)(void);
596 int (*can_fork)(struct task_struct *task);
597 void (*cancel_fork)(struct task_struct *task);
598 void (*fork)(struct task_struct *task);
599 void (*exit)(struct task_struct *task);
600 void (*free)(struct task_struct *task);
601 void (*bind)(struct cgroup_subsys_state *root_css);
602
603 bool early_init:1;
604
605 /*
606 * If %true, the controller, on the default hierarchy, doesn't show
607 * up in "cgroup.controllers" or "cgroup.subtree_control", is
608 * implicitly enabled on all cgroups on the default hierarchy, and
609 * bypasses the "no internal process" constraint. This is for
610 * utility type controllers which is transparent to userland.
611 *
612 * An implicit controller can be stolen from the default hierarchy
613 * anytime and thus must be okay with offline csses from previous
614 * hierarchies coexisting with csses for the current one.
615 */
616 bool implicit_on_dfl:1;
617
618 /*
619 * If %true, the controller, supports threaded mode on the default
620 * hierarchy. In a threaded subtree, both process granularity and
621 * no-internal-process constraint are ignored and a threaded
622 * controllers should be able to handle that.
623 *
624 * Note that as an implicit controller is automatically enabled on
625 * all cgroups on the default hierarchy, it should also be
626 * threaded. implicit && !threaded is not supported.
627 */
628 bool threaded:1;
629
630 /*
631 * If %false, this subsystem is properly hierarchical -
632 * configuration, resource accounting and restriction on a parent
633 * cgroup cover those of its children. If %true, hierarchy support
634 * is broken in some ways - some subsystems ignore hierarchy
635 * completely while others are only implemented half-way.
636 *
637 * It's now disallowed to create nested cgroups if the subsystem is
638 * broken and cgroup core will emit a warning message on such
639 * cases. Eventually, all subsystems will be made properly
640 * hierarchical and this will go away.
641 */
642 bool broken_hierarchy:1;
643 bool warned_broken_hierarchy:1;
644
645 /* the following two fields are initialized automtically during boot */
646 int id;
647 const char *name;
648
649 /* optional, initialized automatically during boot if not set */
650 const char *legacy_name;
651
652 /* link to parent, protected by cgroup_lock() */
653 struct cgroup_root *root;
654
655 /* idr for css->id */
656 struct idr css_idr;
657
658 /*
659 * List of cftypes. Each entry is the first entry of an array
660 * terminated by zero length name.
661 */
662 struct list_head cfts;
663
664 /*
665 * Base cftypes which are automatically registered. The two can
666 * point to the same array.
667 */
668 struct cftype *dfl_cftypes; /* for the default hierarchy */
669 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
670
671 /*
672 * A subsystem may depend on other subsystems. When such subsystem
673 * is enabled on a cgroup, the depended-upon subsystems are enabled
674 * together if available. Subsystems enabled due to dependency are
675 * not visible to userland until explicitly enabled. The following
676 * specifies the mask of subsystems that this one depends on.
677 */
678 unsigned int depends_on;
679 };
680
681 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
682
683 /**
684 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
685 * @tsk: target task
686 *
687 * Allows cgroup operations to synchronize against threadgroup changes
688 * using a percpu_rw_semaphore.
689 */
690 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
691 {
692 percpu_down_read(&cgroup_threadgroup_rwsem);
693 }
694
695 /**
696 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
697 * @tsk: target task
698 *
699 * Counterpart of cgroup_threadcgroup_change_begin().
700 */
701 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
702 {
703 percpu_up_read(&cgroup_threadgroup_rwsem);
704 }
705
706 #else /* CONFIG_CGROUPS */
707
708 #define CGROUP_SUBSYS_COUNT 0
709
710 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
711 {
712 might_sleep();
713 }
714
715 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
716
717 #endif /* CONFIG_CGROUPS */
718
719 #ifdef CONFIG_SOCK_CGROUP_DATA
720
721 /*
722 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
723 * per-socket cgroup information except for memcg association.
724 *
725 * On legacy hierarchies, net_prio and net_cls controllers directly set
726 * attributes on each sock which can then be tested by the network layer.
727 * On the default hierarchy, each sock is associated with the cgroup it was
728 * created in and the networking layer can match the cgroup directly.
729 *
730 * To avoid carrying all three cgroup related fields separately in sock,
731 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
732 * On boot, sock_cgroup_data records the cgroup that the sock was created
733 * in so that cgroup2 matches can be made; however, once either net_prio or
734 * net_cls starts being used, the area is overriden to carry prioidx and/or
735 * classid. The two modes are distinguished by whether the lowest bit is
736 * set. Clear bit indicates cgroup pointer while set bit prioidx and
737 * classid.
738 *
739 * While userland may start using net_prio or net_cls at any time, once
740 * either is used, cgroup2 matching no longer works. There is no reason to
741 * mix the two and this is in line with how legacy and v2 compatibility is
742 * handled. On mode switch, cgroup references which are already being
743 * pointed to by socks may be leaked. While this can be remedied by adding
744 * synchronization around sock_cgroup_data, given that the number of leaked
745 * cgroups is bound and highly unlikely to be high, this seems to be the
746 * better trade-off.
747 */
748 struct sock_cgroup_data {
749 union {
750 #ifdef __LITTLE_ENDIAN
751 struct {
752 u8 is_data;
753 u8 padding;
754 u16 prioidx;
755 u32 classid;
756 } __packed;
757 #else
758 struct {
759 u32 classid;
760 u16 prioidx;
761 u8 padding;
762 u8 is_data;
763 } __packed;
764 #endif
765 u64 val;
766 };
767 };
768
769 /*
770 * There's a theoretical window where the following accessors race with
771 * updaters and return part of the previous pointer as the prioidx or
772 * classid. Such races are short-lived and the result isn't critical.
773 */
774 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
775 {
776 /* fallback to 1 which is always the ID of the root cgroup */
777 return (skcd->is_data & 1) ? skcd->prioidx : 1;
778 }
779
780 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
781 {
782 /* fallback to 0 which is the unconfigured default classid */
783 return (skcd->is_data & 1) ? skcd->classid : 0;
784 }
785
786 /*
787 * If invoked concurrently, the updaters may clobber each other. The
788 * caller is responsible for synchronization.
789 */
790 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
791 u16 prioidx)
792 {
793 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
794
795 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
796 return;
797
798 if (!(skcd_buf.is_data & 1)) {
799 skcd_buf.val = 0;
800 skcd_buf.is_data = 1;
801 }
802
803 skcd_buf.prioidx = prioidx;
804 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
805 }
806
807 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
808 u32 classid)
809 {
810 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
811
812 if (sock_cgroup_classid(&skcd_buf) == classid)
813 return;
814
815 if (!(skcd_buf.is_data & 1)) {
816 skcd_buf.val = 0;
817 skcd_buf.is_data = 1;
818 }
819
820 skcd_buf.classid = classid;
821 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
822 }
823
824 #else /* CONFIG_SOCK_CGROUP_DATA */
825
826 struct sock_cgroup_data {
827 };
828
829 #endif /* CONFIG_SOCK_CGROUP_DATA */
830
831 #endif /* _LINUX_CGROUP_DEFS_H */