1 /* SPDX-License-Identifier: GPL-2.0 */
3 * linux/cgroup-defs.h - basic definitions for cgroup
5 * This file provides basic type and interface. Include this file directly
6 * only if necessary to avoid cyclic dependencies.
8 #ifndef _LINUX_CGROUP_DEFS_H
9 #define _LINUX_CGROUP_DEFS_H
11 #include <linux/limits.h>
12 #include <linux/list.h>
13 #include <linux/idr.h>
14 #include <linux/wait.h>
15 #include <linux/mutex.h>
16 #include <linux/rcupdate.h>
17 #include <linux/refcount.h>
18 #include <linux/percpu-refcount.h>
19 #include <linux/percpu-rwsem.h>
20 #include <linux/u64_stats_sync.h>
21 #include <linux/workqueue.h>
22 #include <linux/bpf-cgroup-defs.h>
23 #include <linux/psi_types.h>
30 struct cgroup_taskset;
33 struct kernfs_open_file;
35 struct poll_table_struct;
37 #define MAX_CGROUP_TYPE_NAMELEN 32
38 #define MAX_CGROUP_ROOT_NAMELEN 64
39 #define MAX_CFTYPE_NAME 64
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>
49 /* bits in struct cgroup_subsys_state flags field */
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 */
58 /* bits in struct cgroup flags field */
60 /* Control Group requires release notifications to userspace */
61 CGRP_NOTIFY_ON_RELEASE,
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.
67 CGRP_CPUSET_CLONE_CHILDREN,
69 /* Control group has to be frozen. */
72 /* Cgroup is frozen. */
75 /* Control group has to be killed. */
79 /* cgroup_root->flags */
81 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
82 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
85 * Consider namespaces as delegation boundaries. If this flag is
86 * set, controller specific interface files in a namespace root
87 * aren't writeable from inside the namespace.
89 CGRP_ROOT_NS_DELEGATE = (1 << 3),
92 * Reduce latencies on dynamic cgroup modifications such as task
93 * migrations and controller on/offs by disabling percpu operation on
94 * cgroup_threadgroup_rwsem. This makes hot path operations such as
95 * forks and exits into the slow path and more expensive.
97 * The static usage pattern of creating a cgroup, enabling controllers,
98 * and then seeding it with CLONE_INTO_CGROUP doesn't require write
99 * locking cgroup_threadgroup_rwsem and thus doesn't benefit from
102 CGRP_ROOT_FAVOR_DYNMODS = (1 << 4),
105 * Enable cpuset controller in v1 cgroup to use v2 behavior.
107 CGRP_ROOT_CPUSET_V2_MODE = (1 << 16),
110 * Enable legacy local memory.events.
112 CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 17),
115 * Enable recursive subtree protection
117 CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 18),
120 * Enable hugetlb accounting for the memory controller.
122 CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING = (1 << 19),
125 * Enable legacy local pids.events.
127 CGRP_ROOT_PIDS_LOCAL_EVENTS = (1 << 20),
132 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
133 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
134 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
136 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
137 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
138 CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */
140 /* internal flags, do not use outside cgroup core proper */
141 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
142 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
143 __CFTYPE_ADDED = (1 << 18),
147 * cgroup_file is the handle for a file instance created in a cgroup which
148 * is used, for example, to generate file changed notifications. This can
149 * be obtained by setting cftype->file_offset.
152 /* do not access any fields from outside cgroup core */
153 struct kernfs_node *kn;
154 unsigned long notified_at;
155 struct timer_list notify_timer;
159 * Per-subsystem/per-cgroup state maintained by the system. This is the
160 * fundamental structural building block that controllers deal with.
162 * Fields marked with "PI:" are public and immutable and may be accessed
163 * directly without synchronization.
165 struct cgroup_subsys_state {
166 /* PI: the cgroup that this css is attached to */
167 struct cgroup *cgroup;
169 /* PI: the cgroup subsystem that this css is attached to */
170 struct cgroup_subsys *ss;
172 /* reference count - access via css_[try]get() and css_put() */
173 struct percpu_ref refcnt;
175 /* siblings list anchored at the parent's ->children */
176 struct list_head sibling;
177 struct list_head children;
179 /* flush target list anchored at cgrp->rstat_css_list */
180 struct list_head rstat_css_node;
183 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
184 * matching css can be looked up using css_from_id().
191 * Monotonically increasing unique serial number which defines a
192 * uniform order among all csses. It's guaranteed that all
193 * ->children lists are in the ascending order of ->serial_nr and
194 * used to allow interrupting and resuming iterations.
199 * Incremented by online self and children. Used to guarantee that
200 * parents are not offlined before their children.
204 /* percpu_ref killing and RCU release */
205 struct work_struct destroy_work;
206 struct rcu_work destroy_rwork;
209 * PI: the parent css. Placed here for cache proximity to following
210 * fields of the containing structure.
212 struct cgroup_subsys_state *parent;
215 * Keep track of total numbers of visible descendant CSSes.
216 * The total number of dying CSSes is tracked in
217 * css->cgroup->nr_dying_subsys[ssid].
218 * Protected by cgroup_mutex.
224 * A css_set is a structure holding pointers to a set of
225 * cgroup_subsys_state objects. This saves space in the task struct
226 * object and speeds up fork()/exit(), since a single inc/dec and a
227 * list_add()/del() can bump the reference count on the entire cgroup
232 * Set of subsystem states, one for each subsystem. This array is
233 * immutable after creation apart from the init_css_set during
234 * subsystem registration (at boot time).
236 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
238 /* reference count */
242 * For a domain cgroup, the following points to self. If threaded,
243 * to the matching cset of the nearest domain ancestor. The
244 * dom_cset provides access to the domain cgroup and its csses to
245 * which domain level resource consumptions should be charged.
247 struct css_set *dom_cset;
249 /* the default cgroup associated with this css_set */
250 struct cgroup *dfl_cgrp;
252 /* internal task count, protected by css_set_lock */
256 * Lists running through all tasks using this cgroup group.
257 * mg_tasks lists tasks which belong to this cset but are in the
258 * process of being migrated out or in. Protected by
259 * css_set_lock, but, during migration, once tasks are moved to
260 * mg_tasks, it can be read safely while holding cgroup_mutex.
262 struct list_head tasks;
263 struct list_head mg_tasks;
264 struct list_head dying_tasks;
266 /* all css_task_iters currently walking this cset */
267 struct list_head task_iters;
270 * On the default hierarchy, ->subsys[ssid] may point to a css
271 * attached to an ancestor instead of the cgroup this css_set is
272 * associated with. The following node is anchored at
273 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
274 * iterate through all css's attached to a given cgroup.
276 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
278 /* all threaded csets whose ->dom_cset points to this cset */
279 struct list_head threaded_csets;
280 struct list_head threaded_csets_node;
283 * List running through all cgroup groups in the same hash
284 * slot. Protected by css_set_lock
286 struct hlist_node hlist;
289 * List of cgrp_cset_links pointing at cgroups referenced from this
290 * css_set. Protected by css_set_lock.
292 struct list_head cgrp_links;
295 * List of csets participating in the on-going migration either as
296 * source or destination. Protected by cgroup_mutex.
298 struct list_head mg_src_preload_node;
299 struct list_head mg_dst_preload_node;
300 struct list_head mg_node;
303 * If this cset is acting as the source of migration the following
304 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
305 * respectively the source and destination cgroups of the on-going
306 * migration. mg_dst_cset is the destination cset the target tasks
307 * on this cset should be migrated to. Protected by cgroup_mutex.
309 struct cgroup *mg_src_cgrp;
310 struct cgroup *mg_dst_cgrp;
311 struct css_set *mg_dst_cset;
313 /* dead and being drained, ignore for migration */
316 /* For RCU-protected deletion */
317 struct rcu_head rcu_head;
320 struct cgroup_base_stat {
321 struct task_cputime cputime;
323 #ifdef CONFIG_SCHED_CORE
329 * rstat - cgroup scalable recursive statistics. Accounting is done
330 * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
331 * hierarchy on reads.
333 * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
334 * linked into the updated tree. On the following read, propagation only
335 * considers and consumes the updated tree. This makes reading O(the
336 * number of descendants which have been active since last read) instead of
337 * O(the total number of descendants).
339 * This is important because there can be a lot of (draining) cgroups which
340 * aren't active and stat may be read frequently. The combination can
341 * become very expensive. By propagating selectively, increasing reading
342 * frequency decreases the cost of each read.
344 * This struct hosts both the fields which implement the above -
345 * updated_children and updated_next - and the fields which track basic
346 * resource statistics on top of it - bsync, bstat and last_bstat.
348 struct cgroup_rstat_cpu {
350 * ->bsync protects ->bstat. These are the only fields which get
351 * updated in the hot path.
353 struct u64_stats_sync bsync;
354 struct cgroup_base_stat bstat;
357 * Snapshots at the last reading. These are used to calculate the
358 * deltas to propagate to the global counters.
360 struct cgroup_base_stat last_bstat;
363 * This field is used to record the cumulative per-cpu time of
364 * the cgroup and its descendants. Currently it can be read via
365 * eBPF/drgn etc, and we are still trying to determine how to
366 * expose it in the cgroupfs interface.
368 struct cgroup_base_stat subtree_bstat;
371 * Snapshots at the last reading. These are used to calculate the
372 * deltas to propagate to the per-cpu subtree_bstat.
374 struct cgroup_base_stat last_subtree_bstat;
377 * Child cgroups with stat updates on this cpu since the last read
378 * are linked on the parent's ->updated_children through
381 * In addition to being more compact, singly-linked list pointing
382 * to the cgroup makes it unnecessary for each per-cpu struct to
383 * point back to the associated cgroup.
385 * Protected by per-cpu cgroup_rstat_cpu_lock.
387 struct cgroup *updated_children; /* terminated by self cgroup */
388 struct cgroup *updated_next; /* NULL iff not on the list */
391 struct cgroup_freezer_state {
392 /* Should the cgroup and its descendants be frozen. */
395 /* Should the cgroup actually be frozen? */
398 /* Fields below are protected by css_set_lock */
400 /* Number of frozen descendant cgroups */
401 int nr_frozen_descendants;
404 * Number of tasks, which are counted as frozen:
405 * frozen, SIGSTOPped, and PTRACEd.
411 /* self css with NULL ->ss, points back to this cgroup */
412 struct cgroup_subsys_state self;
414 unsigned long flags; /* "unsigned long" so bitops work */
417 * The depth this cgroup is at. The root is at depth zero and each
418 * step down the hierarchy increments the level. This along with
419 * ancestors[] can determine whether a given cgroup is a
420 * descendant of another without traversing the hierarchy.
424 /* Maximum allowed descent tree depth */
428 * Keep track of total numbers of visible and dying descent cgroups.
429 * Dying cgroups are cgroups which were deleted by a user,
430 * but are still existing because someone else is holding a reference.
431 * max_descendants is a maximum allowed number of descent cgroups.
433 * nr_descendants and nr_dying_descendants are protected
434 * by cgroup_mutex and css_set_lock. It's fine to read them holding
435 * any of cgroup_mutex and css_set_lock; for writing both locks
439 int nr_dying_descendants;
443 * Each non-empty css_set associated with this cgroup contributes
444 * one to nr_populated_csets. The counter is zero iff this cgroup
445 * doesn't have any tasks.
447 * All children which have non-zero nr_populated_csets and/or
448 * nr_populated_children of their own contribute one to either
449 * nr_populated_domain_children or nr_populated_threaded_children
450 * depending on their type. Each counter is zero iff all cgroups
451 * of the type in the subtree proper don't have any tasks.
453 int nr_populated_csets;
454 int nr_populated_domain_children;
455 int nr_populated_threaded_children;
457 int nr_threaded_children; /* # of live threaded child cgroups */
459 struct kernfs_node *kn; /* cgroup kernfs entry */
460 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
461 struct cgroup_file events_file; /* handle for "cgroup.events" */
463 /* handles for "{cpu,memory,io,irq}.pressure" */
464 struct cgroup_file psi_files[NR_PSI_RESOURCES];
467 * The bitmask of subsystems enabled on the child cgroups.
468 * ->subtree_control is the one configured through
469 * "cgroup.subtree_control" while ->subtree_ss_mask is the effective
470 * one which may have more subsystems enabled. Controller knobs
471 * are made available iff it's enabled in ->subtree_control.
475 u16 old_subtree_control;
476 u16 old_subtree_ss_mask;
478 /* Private pointers for each registered subsystem */
479 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
482 * Keep track of total number of dying CSSes at and below this cgroup.
483 * Protected by cgroup_mutex.
485 int nr_dying_subsys[CGROUP_SUBSYS_COUNT];
487 struct cgroup_root *root;
490 * List of cgrp_cset_links pointing at css_sets with tasks in this
491 * cgroup. Protected by css_set_lock.
493 struct list_head cset_links;
496 * On the default hierarchy, a css_set for a cgroup with some
497 * susbsys disabled will point to css's which are associated with
498 * the closest ancestor which has the subsys enabled. The
499 * following lists all css_sets which point to this cgroup's css
500 * for the given subsystem.
502 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
505 * If !threaded, self. If threaded, it points to the nearest
506 * domain ancestor. Inside a threaded subtree, cgroups are exempt
507 * from process granularity and no-internal-task constraint.
508 * Domain level resource consumptions which aren't tied to a
509 * specific task are charged to the dom_cgrp.
511 struct cgroup *dom_cgrp;
512 struct cgroup *old_dom_cgrp; /* used while enabling threaded */
514 /* per-cpu recursive resource statistics */
515 struct cgroup_rstat_cpu __percpu *rstat_cpu;
516 struct list_head rstat_css_list;
519 * Add padding to separate the read mostly rstat_cpu and
520 * rstat_css_list into a different cacheline from the following
521 * rstat_flush_next and *bstat fields which can have frequent updates.
523 CACHELINE_PADDING(_pad_);
526 * A singly-linked list of cgroup structures to be rstat flushed.
527 * This is a scratch field to be used exclusively by
528 * cgroup_rstat_flush_locked() and protected by cgroup_rstat_lock.
530 struct cgroup *rstat_flush_next;
532 /* cgroup basic resource statistics */
533 struct cgroup_base_stat last_bstat;
534 struct cgroup_base_stat bstat;
535 struct prev_cputime prev_cputime; /* for printing out cputime */
538 * list of pidlists, up to two for each namespace (one for procs, one
539 * for tasks); created on demand.
541 struct list_head pidlists;
542 struct mutex pidlist_mutex;
544 /* used to wait for offlining of csses */
545 wait_queue_head_t offline_waitq;
547 /* used to schedule release agent */
548 struct work_struct release_agent_work;
550 /* used to track pressure stalls */
551 struct psi_group *psi;
553 /* used to store eBPF programs */
554 struct cgroup_bpf bpf;
556 /* Used to store internal freezer state */
557 struct cgroup_freezer_state freezer;
559 #ifdef CONFIG_BPF_SYSCALL
560 struct bpf_local_storage __rcu *bpf_cgrp_storage;
563 /* All ancestors including self */
564 struct cgroup *ancestors[];
568 * A cgroup_root represents the root of a cgroup hierarchy, and may be
569 * associated with a kernfs_root to form an active hierarchy. This is
570 * internal to cgroup core. Don't access directly from controllers.
573 struct kernfs_root *kf_root;
575 /* The bitmask of subsystems attached to this hierarchy */
576 unsigned int subsys_mask;
578 /* Unique id for this hierarchy. */
581 /* A list running through the active hierarchies */
582 struct list_head root_list;
583 struct rcu_head rcu; /* Must be near the top */
586 * The root cgroup. The containing cgroup_root will be destroyed on its
587 * release. cgrp->ancestors[0] will be used overflowing into the
588 * following field. cgrp_ancestor_storage must immediately follow.
592 /* must follow cgrp for cgrp->ancestors[0], see above */
593 struct cgroup *cgrp_ancestor_storage;
595 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
598 /* Hierarchy-specific flags */
601 /* The path to use for release notifications. */
602 char release_agent_path[PATH_MAX];
604 /* The name for this hierarchy - may be empty */
605 char name[MAX_CGROUP_ROOT_NAMELEN];
609 * struct cftype: handler definitions for cgroup control files
611 * When reading/writing to a file:
612 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
613 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
617 * By convention, the name should begin with the name of the
618 * subsystem, followed by a period. Zero length string indicates
619 * end of cftype array.
621 char name[MAX_CFTYPE_NAME];
622 unsigned long private;
625 * The maximum length of string, excluding trailing nul, that can
626 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
628 size_t max_write_len;
634 * If non-zero, should contain the offset from the start of css to
635 * a struct cgroup_file field. cgroup will record the handle of
636 * the created file into it. The recorded handle can be used as
637 * long as the containing css remains accessible.
639 unsigned int file_offset;
642 * Fields used for internal bookkeeping. Initialized automatically
643 * during registration.
645 struct cgroup_subsys *ss; /* NULL for cgroup core files */
646 struct list_head node; /* anchored at ss->cfts */
647 struct kernfs_ops *kf_ops;
649 int (*open)(struct kernfs_open_file *of);
650 void (*release)(struct kernfs_open_file *of);
653 * read_u64() is a shortcut for the common case of returning a
654 * single integer. Use it in place of read()
656 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
658 * read_s64() is a signed version of read_u64()
660 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
662 /* generic seq_file read interface */
663 int (*seq_show)(struct seq_file *sf, void *v);
665 /* optional ops, implement all or none */
666 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
667 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
668 void (*seq_stop)(struct seq_file *sf, void *v);
671 * write_u64() is a shortcut for the common case of accepting
672 * a single integer (as parsed by simple_strtoull) from
673 * userspace. Use in place of write(); return 0 or error.
675 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
678 * write_s64() is a signed version of write_u64()
680 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
684 * write() is the generic write callback which maps directly to
685 * kernfs write operation and overrides all other operations.
686 * Maximum write size is determined by ->max_write_len. Use
687 * of_css/cft() to access the associated css and cft.
689 ssize_t (*write)(struct kernfs_open_file *of,
690 char *buf, size_t nbytes, loff_t off);
692 __poll_t (*poll)(struct kernfs_open_file *of,
693 struct poll_table_struct *pt);
695 struct lock_class_key lockdep_key;
699 * Control Group subsystem type.
700 * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
702 struct cgroup_subsys {
703 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
704 int (*css_online)(struct cgroup_subsys_state *css);
705 void (*css_offline)(struct cgroup_subsys_state *css);
706 void (*css_released)(struct cgroup_subsys_state *css);
707 void (*css_free)(struct cgroup_subsys_state *css);
708 void (*css_reset)(struct cgroup_subsys_state *css);
709 void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
710 int (*css_extra_stat_show)(struct seq_file *seq,
711 struct cgroup_subsys_state *css);
712 int (*css_local_stat_show)(struct seq_file *seq,
713 struct cgroup_subsys_state *css);
715 int (*can_attach)(struct cgroup_taskset *tset);
716 void (*cancel_attach)(struct cgroup_taskset *tset);
717 void (*attach)(struct cgroup_taskset *tset);
718 void (*post_attach)(void);
719 int (*can_fork)(struct task_struct *task,
720 struct css_set *cset);
721 void (*cancel_fork)(struct task_struct *task, struct css_set *cset);
722 void (*fork)(struct task_struct *task);
723 void (*exit)(struct task_struct *task);
724 void (*release)(struct task_struct *task);
725 void (*bind)(struct cgroup_subsys_state *root_css);
730 * If %true, the controller, on the default hierarchy, doesn't show
731 * up in "cgroup.controllers" or "cgroup.subtree_control", is
732 * implicitly enabled on all cgroups on the default hierarchy, and
733 * bypasses the "no internal process" constraint. This is for
734 * utility type controllers which is transparent to userland.
736 * An implicit controller can be stolen from the default hierarchy
737 * anytime and thus must be okay with offline csses from previous
738 * hierarchies coexisting with csses for the current one.
740 bool implicit_on_dfl:1;
743 * If %true, the controller, supports threaded mode on the default
744 * hierarchy. In a threaded subtree, both process granularity and
745 * no-internal-process constraint are ignored and a threaded
746 * controllers should be able to handle that.
748 * Note that as an implicit controller is automatically enabled on
749 * all cgroups on the default hierarchy, it should also be
750 * threaded. implicit && !threaded is not supported.
754 /* the following two fields are initialized automatically during boot */
758 /* optional, initialized automatically during boot if not set */
759 const char *legacy_name;
761 /* link to parent, protected by cgroup_lock() */
762 struct cgroup_root *root;
764 /* idr for css->id */
768 * List of cftypes. Each entry is the first entry of an array
769 * terminated by zero length name.
771 struct list_head cfts;
774 * Base cftypes which are automatically registered. The two can
775 * point to the same array.
777 struct cftype *dfl_cftypes; /* for the default hierarchy */
778 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
781 * A subsystem may depend on other subsystems. When such subsystem
782 * is enabled on a cgroup, the depended-upon subsystems are enabled
783 * together if available. Subsystems enabled due to dependency are
784 * not visible to userland until explicitly enabled. The following
785 * specifies the mask of subsystems that this one depends on.
787 unsigned int depends_on;
790 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
793 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
796 * Allows cgroup operations to synchronize against threadgroup changes
797 * using a percpu_rw_semaphore.
799 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
801 percpu_down_read(&cgroup_threadgroup_rwsem);
805 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
808 * Counterpart of cgroup_threadcgroup_change_begin().
810 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
812 percpu_up_read(&cgroup_threadgroup_rwsem);
815 #else /* CONFIG_CGROUPS */
817 #define CGROUP_SUBSYS_COUNT 0
819 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
824 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
826 #endif /* CONFIG_CGROUPS */
828 #ifdef CONFIG_SOCK_CGROUP_DATA
831 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
832 * per-socket cgroup information except for memcg association.
834 * On legacy hierarchies, net_prio and net_cls controllers directly
835 * set attributes on each sock which can then be tested by the network
836 * layer. On the default hierarchy, each sock is associated with the
837 * cgroup it was created in and the networking layer can match the
840 struct sock_cgroup_data {
841 struct cgroup *cgroup; /* v2 */
842 #ifdef CONFIG_CGROUP_NET_CLASSID
843 u32 classid; /* v1 */
845 #ifdef CONFIG_CGROUP_NET_PRIO
846 u16 prioidx; /* v1 */
850 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
852 #ifdef CONFIG_CGROUP_NET_PRIO
853 return READ_ONCE(skcd->prioidx);
859 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
861 #ifdef CONFIG_CGROUP_NET_CLASSID
862 return READ_ONCE(skcd->classid);
868 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
871 #ifdef CONFIG_CGROUP_NET_PRIO
872 WRITE_ONCE(skcd->prioidx, prioidx);
876 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
879 #ifdef CONFIG_CGROUP_NET_CLASSID
880 WRITE_ONCE(skcd->classid, classid);
884 #else /* CONFIG_SOCK_CGROUP_DATA */
886 struct sock_cgroup_data {
889 #endif /* CONFIG_SOCK_CGROUP_DATA */
891 #endif /* _LINUX_CGROUP_DEFS_H */