2 * linux/cgroup-defs.h - basic definitions for cgroup
4 * This file provides basic type and interface. Include this file directly
5 * only if necessary to avoid cyclic dependencies.
7 #ifndef _LINUX_CGROUP_DEFS_H
8 #define _LINUX_CGROUP_DEFS_H
10 #include <linux/limits.h>
11 #include <linux/list.h>
12 #include <linux/idr.h>
13 #include <linux/wait.h>
14 #include <linux/mutex.h>
15 #include <linux/rcupdate.h>
16 #include <linux/refcount.h>
17 #include <linux/percpu-refcount.h>
18 #include <linux/percpu-rwsem.h>
19 #include <linux/workqueue.h>
20 #include <linux/bpf-cgroup.h>
27 struct cgroup_taskset;
30 struct kernfs_open_file;
33 #define MAX_CGROUP_TYPE_NAMELEN 32
34 #define MAX_CGROUP_ROOT_NAMELEN 64
35 #define MAX_CFTYPE_NAME 64
37 /* define the enumeration of all cgroup subsystems */
38 #define SUBSYS(_x) _x ## _cgrp_id,
39 enum cgroup_subsys_id {
40 #include <linux/cgroup_subsys.h>
45 /* bits in struct cgroup_subsys_state flags field */
47 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
48 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
49 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
50 CSS_VISIBLE = (1 << 3), /* css is visible to userland */
53 /* bits in struct cgroup flags field */
55 /* Control Group requires release notifications to userspace */
56 CGRP_NOTIFY_ON_RELEASE,
58 * Clone the parent's configuration when creating a new child
59 * cpuset cgroup. For historical reasons, this option can be
60 * specified at mount time and thus is implemented here.
62 CGRP_CPUSET_CLONE_CHILDREN,
65 /* cgroup_root->flags */
67 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
68 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
73 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
74 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
75 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
76 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
78 /* internal flags, do not use outside cgroup core proper */
79 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
80 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
84 * cgroup_file is the handle for a file instance created in a cgroup which
85 * is used, for example, to generate file changed notifications. This can
86 * be obtained by setting cftype->file_offset.
89 /* do not access any fields from outside cgroup core */
90 struct kernfs_node *kn;
94 * Per-subsystem/per-cgroup state maintained by the system. This is the
95 * fundamental structural building block that controllers deal with.
97 * Fields marked with "PI:" are public and immutable and may be accessed
98 * directly without synchronization.
100 struct cgroup_subsys_state {
101 /* PI: the cgroup that this css is attached to */
102 struct cgroup *cgroup;
104 /* PI: the cgroup subsystem that this css is attached to */
105 struct cgroup_subsys *ss;
107 /* reference count - access via css_[try]get() and css_put() */
108 struct percpu_ref refcnt;
110 /* siblings list anchored at the parent's ->children */
111 struct list_head sibling;
112 struct list_head children;
115 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
116 * matching css can be looked up using css_from_id().
123 * Monotonically increasing unique serial number which defines a
124 * uniform order among all csses. It's guaranteed that all
125 * ->children lists are in the ascending order of ->serial_nr and
126 * used to allow interrupting and resuming iterations.
131 * Incremented by online self and children. Used to guarantee that
132 * parents are not offlined before their children.
136 /* percpu_ref killing and RCU release */
137 struct rcu_head rcu_head;
138 struct work_struct destroy_work;
141 * PI: the parent css. Placed here for cache proximity to following
142 * fields of the containing structure.
144 struct cgroup_subsys_state *parent;
148 * A css_set is a structure holding pointers to a set of
149 * cgroup_subsys_state objects. This saves space in the task struct
150 * object and speeds up fork()/exit(), since a single inc/dec and a
151 * list_add()/del() can bump the reference count on the entire cgroup
156 * Set of subsystem states, one for each subsystem. This array is
157 * immutable after creation apart from the init_css_set during
158 * subsystem registration (at boot time).
160 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
162 /* reference count */
165 /* the default cgroup associated with this css_set */
166 struct cgroup *dfl_cgrp;
169 * Lists running through all tasks using this cgroup group.
170 * mg_tasks lists tasks which belong to this cset but are in the
171 * process of being migrated out or in. Protected by
172 * css_set_rwsem, but, during migration, once tasks are moved to
173 * mg_tasks, it can be read safely while holding cgroup_mutex.
175 struct list_head tasks;
176 struct list_head mg_tasks;
178 /* all css_task_iters currently walking this cset */
179 struct list_head task_iters;
182 * On the default hierarhcy, ->subsys[ssid] may point to a css
183 * attached to an ancestor instead of the cgroup this css_set is
184 * associated with. The following node is anchored at
185 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
186 * iterate through all css's attached to a given cgroup.
188 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
191 * List running through all cgroup groups in the same hash
192 * slot. Protected by css_set_lock
194 struct hlist_node hlist;
197 * List of cgrp_cset_links pointing at cgroups referenced from this
198 * css_set. Protected by css_set_lock.
200 struct list_head cgrp_links;
203 * List of csets participating in the on-going migration either as
204 * source or destination. Protected by cgroup_mutex.
206 struct list_head mg_preload_node;
207 struct list_head mg_node;
210 * If this cset is acting as the source of migration the following
211 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
212 * respectively the source and destination cgroups of the on-going
213 * migration. mg_dst_cset is the destination cset the target tasks
214 * on this cset should be migrated to. Protected by cgroup_mutex.
216 struct cgroup *mg_src_cgrp;
217 struct cgroup *mg_dst_cgrp;
218 struct css_set *mg_dst_cset;
220 /* dead and being drained, ignore for migration */
223 /* For RCU-protected deletion */
224 struct rcu_head rcu_head;
228 /* self css with NULL ->ss, points back to this cgroup */
229 struct cgroup_subsys_state self;
231 unsigned long flags; /* "unsigned long" so bitops work */
234 * idr allocated in-hierarchy ID.
236 * ID 0 is not used, the ID of the root cgroup is always 1, and a
237 * new cgroup will be assigned with a smallest available ID.
239 * Allocating/Removing ID must be protected by cgroup_mutex.
244 * The depth this cgroup is at. The root is at depth zero and each
245 * step down the hierarchy increments the level. This along with
246 * ancestor_ids[] can determine whether a given cgroup is a
247 * descendant of another without traversing the hierarchy.
252 * Each non-empty css_set associated with this cgroup contributes
253 * one to populated_cnt. All children with non-zero popuplated_cnt
254 * of their own contribute one. The count is zero iff there's no
255 * task in this cgroup or its subtree.
259 struct kernfs_node *kn; /* cgroup kernfs entry */
260 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
261 struct cgroup_file events_file; /* handle for "cgroup.events" */
264 * The bitmask of subsystems enabled on the child cgroups.
265 * ->subtree_control is the one configured through
266 * "cgroup.subtree_control" while ->child_ss_mask is the effective
267 * one which may have more subsystems enabled. Controller knobs
268 * are made available iff it's enabled in ->subtree_control.
272 u16 old_subtree_control;
273 u16 old_subtree_ss_mask;
275 /* Private pointers for each registered subsystem */
276 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
278 struct cgroup_root *root;
281 * List of cgrp_cset_links pointing at css_sets with tasks in this
282 * cgroup. Protected by css_set_lock.
284 struct list_head cset_links;
287 * On the default hierarchy, a css_set for a cgroup with some
288 * susbsys disabled will point to css's which are associated with
289 * the closest ancestor which has the subsys enabled. The
290 * following lists all css_sets which point to this cgroup's css
291 * for the given subsystem.
293 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
296 * list of pidlists, up to two for each namespace (one for procs, one
297 * for tasks); created on demand.
299 struct list_head pidlists;
300 struct mutex pidlist_mutex;
302 /* used to wait for offlining of csses */
303 wait_queue_head_t offline_waitq;
305 /* used to schedule release agent */
306 struct work_struct release_agent_work;
308 /* used to store eBPF programs */
309 struct cgroup_bpf bpf;
311 /* ids of the ancestors at each level including self */
316 * A cgroup_root represents the root of a cgroup hierarchy, and may be
317 * associated with a kernfs_root to form an active hierarchy. This is
318 * internal to cgroup core. Don't access directly from controllers.
321 struct kernfs_root *kf_root;
323 /* The bitmask of subsystems attached to this hierarchy */
324 unsigned int subsys_mask;
326 /* Unique id for this hierarchy. */
329 /* The root cgroup. Root is destroyed on its release. */
332 /* for cgrp->ancestor_ids[0] */
333 int cgrp_ancestor_id_storage;
335 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
338 /* A list running through the active hierarchies */
339 struct list_head root_list;
341 /* Hierarchy-specific flags */
344 /* IDs for cgroups in this hierarchy */
345 struct idr cgroup_idr;
347 /* The path to use for release notifications. */
348 char release_agent_path[PATH_MAX];
350 /* The name for this hierarchy - may be empty */
351 char name[MAX_CGROUP_ROOT_NAMELEN];
355 * struct cftype: handler definitions for cgroup control files
357 * When reading/writing to a file:
358 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
359 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
363 * By convention, the name should begin with the name of the
364 * subsystem, followed by a period. Zero length string indicates
365 * end of cftype array.
367 char name[MAX_CFTYPE_NAME];
368 unsigned long private;
371 * The maximum length of string, excluding trailing nul, that can
372 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
374 size_t max_write_len;
380 * If non-zero, should contain the offset from the start of css to
381 * a struct cgroup_file field. cgroup will record the handle of
382 * the created file into it. The recorded handle can be used as
383 * long as the containing css remains accessible.
385 unsigned int file_offset;
388 * Fields used for internal bookkeeping. Initialized automatically
389 * during registration.
391 struct cgroup_subsys *ss; /* NULL for cgroup core files */
392 struct list_head node; /* anchored at ss->cfts */
393 struct kernfs_ops *kf_ops;
395 int (*open)(struct kernfs_open_file *of);
396 void (*release)(struct kernfs_open_file *of);
399 * read_u64() is a shortcut for the common case of returning a
400 * single integer. Use it in place of read()
402 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
404 * read_s64() is a signed version of read_u64()
406 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
408 /* generic seq_file read interface */
409 int (*seq_show)(struct seq_file *sf, void *v);
411 /* optional ops, implement all or none */
412 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
413 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
414 void (*seq_stop)(struct seq_file *sf, void *v);
417 * write_u64() is a shortcut for the common case of accepting
418 * a single integer (as parsed by simple_strtoull) from
419 * userspace. Use in place of write(); return 0 or error.
421 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
424 * write_s64() is a signed version of write_u64()
426 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
430 * write() is the generic write callback which maps directly to
431 * kernfs write operation and overrides all other operations.
432 * Maximum write size is determined by ->max_write_len. Use
433 * of_css/cft() to access the associated css and cft.
435 ssize_t (*write)(struct kernfs_open_file *of,
436 char *buf, size_t nbytes, loff_t off);
438 #ifdef CONFIG_DEBUG_LOCK_ALLOC
439 struct lock_class_key lockdep_key;
444 * Control Group subsystem type.
445 * See Documentation/cgroups/cgroups.txt for details
447 struct cgroup_subsys {
448 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
449 int (*css_online)(struct cgroup_subsys_state *css);
450 void (*css_offline)(struct cgroup_subsys_state *css);
451 void (*css_released)(struct cgroup_subsys_state *css);
452 void (*css_free)(struct cgroup_subsys_state *css);
453 void (*css_reset)(struct cgroup_subsys_state *css);
455 int (*can_attach)(struct cgroup_taskset *tset);
456 void (*cancel_attach)(struct cgroup_taskset *tset);
457 void (*attach)(struct cgroup_taskset *tset);
458 void (*post_attach)(void);
459 int (*can_fork)(struct task_struct *task);
460 void (*cancel_fork)(struct task_struct *task);
461 void (*fork)(struct task_struct *task);
462 void (*exit)(struct task_struct *task);
463 void (*free)(struct task_struct *task);
464 void (*bind)(struct cgroup_subsys_state *root_css);
469 * If %true, the controller, on the default hierarchy, doesn't show
470 * up in "cgroup.controllers" or "cgroup.subtree_control", is
471 * implicitly enabled on all cgroups on the default hierarchy, and
472 * bypasses the "no internal process" constraint. This is for
473 * utility type controllers which is transparent to userland.
475 * An implicit controller can be stolen from the default hierarchy
476 * anytime and thus must be okay with offline csses from previous
477 * hierarchies coexisting with csses for the current one.
479 bool implicit_on_dfl:1;
482 * If %false, this subsystem is properly hierarchical -
483 * configuration, resource accounting and restriction on a parent
484 * cgroup cover those of its children. If %true, hierarchy support
485 * is broken in some ways - some subsystems ignore hierarchy
486 * completely while others are only implemented half-way.
488 * It's now disallowed to create nested cgroups if the subsystem is
489 * broken and cgroup core will emit a warning message on such
490 * cases. Eventually, all subsystems will be made properly
491 * hierarchical and this will go away.
493 bool broken_hierarchy:1;
494 bool warned_broken_hierarchy:1;
496 /* the following two fields are initialized automtically during boot */
500 /* optional, initialized automatically during boot if not set */
501 const char *legacy_name;
503 /* link to parent, protected by cgroup_lock() */
504 struct cgroup_root *root;
506 /* idr for css->id */
510 * List of cftypes. Each entry is the first entry of an array
511 * terminated by zero length name.
513 struct list_head cfts;
516 * Base cftypes which are automatically registered. The two can
517 * point to the same array.
519 struct cftype *dfl_cftypes; /* for the default hierarchy */
520 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
523 * A subsystem may depend on other subsystems. When such subsystem
524 * is enabled on a cgroup, the depended-upon subsystems are enabled
525 * together if available. Subsystems enabled due to dependency are
526 * not visible to userland until explicitly enabled. The following
527 * specifies the mask of subsystems that this one depends on.
529 unsigned int depends_on;
532 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
535 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
538 * Allows cgroup operations to synchronize against threadgroup changes
539 * using a percpu_rw_semaphore.
541 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
543 percpu_down_read(&cgroup_threadgroup_rwsem);
547 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
550 * Counterpart of cgroup_threadcgroup_change_begin().
552 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
554 percpu_up_read(&cgroup_threadgroup_rwsem);
557 #else /* CONFIG_CGROUPS */
559 #define CGROUP_SUBSYS_COUNT 0
561 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
566 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
568 #endif /* CONFIG_CGROUPS */
570 #ifdef CONFIG_SOCK_CGROUP_DATA
573 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
574 * per-socket cgroup information except for memcg association.
576 * On legacy hierarchies, net_prio and net_cls controllers directly set
577 * attributes on each sock which can then be tested by the network layer.
578 * On the default hierarchy, each sock is associated with the cgroup it was
579 * created in and the networking layer can match the cgroup directly.
581 * To avoid carrying all three cgroup related fields separately in sock,
582 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
583 * On boot, sock_cgroup_data records the cgroup that the sock was created
584 * in so that cgroup2 matches can be made; however, once either net_prio or
585 * net_cls starts being used, the area is overriden to carry prioidx and/or
586 * classid. The two modes are distinguished by whether the lowest bit is
587 * set. Clear bit indicates cgroup pointer while set bit prioidx and
590 * While userland may start using net_prio or net_cls at any time, once
591 * either is used, cgroup2 matching no longer works. There is no reason to
592 * mix the two and this is in line with how legacy and v2 compatibility is
593 * handled. On mode switch, cgroup references which are already being
594 * pointed to by socks may be leaked. While this can be remedied by adding
595 * synchronization around sock_cgroup_data, given that the number of leaked
596 * cgroups is bound and highly unlikely to be high, this seems to be the
599 struct sock_cgroup_data {
601 #ifdef __LITTLE_ENDIAN
621 * There's a theoretical window where the following accessors race with
622 * updaters and return part of the previous pointer as the prioidx or
623 * classid. Such races are short-lived and the result isn't critical.
625 static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd)
627 /* fallback to 1 which is always the ID of the root cgroup */
628 return (skcd->is_data & 1) ? skcd->prioidx : 1;
631 static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd)
633 /* fallback to 0 which is the unconfigured default classid */
634 return (skcd->is_data & 1) ? skcd->classid : 0;
638 * If invoked concurrently, the updaters may clobber each other. The
639 * caller is responsible for synchronization.
641 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
644 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
646 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
649 if (!(skcd_buf.is_data & 1)) {
651 skcd_buf.is_data = 1;
654 skcd_buf.prioidx = prioidx;
655 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
658 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
661 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
663 if (sock_cgroup_classid(&skcd_buf) == classid)
666 if (!(skcd_buf.is_data & 1)) {
668 skcd_buf.is_data = 1;
671 skcd_buf.classid = classid;
672 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
675 #else /* CONFIG_SOCK_CGROUP_DATA */
677 struct sock_cgroup_data {
680 #endif /* CONFIG_SOCK_CGROUP_DATA */
682 #endif /* _LINUX_CGROUP_DEFS_H */