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/percpu-refcount.h>
17 #include <linux/percpu-rwsem.h>
18 #include <linux/workqueue.h>
25 struct cgroup_taskset;
28 struct kernfs_open_file;
31 #define MAX_CGROUP_TYPE_NAMELEN 32
32 #define MAX_CGROUP_ROOT_NAMELEN 64
33 #define MAX_CFTYPE_NAME 64
35 /* define the enumeration of all cgroup subsystems */
36 #define SUBSYS(_x) _x ## _cgrp_id,
37 enum cgroup_subsys_id {
38 #include <linux/cgroup_subsys.h>
43 /* bits in struct cgroup_subsys_state flags field */
45 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
46 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
47 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
50 /* bits in struct cgroup flags field */
52 /* Control Group requires release notifications to userspace */
53 CGRP_NOTIFY_ON_RELEASE,
55 * Clone the parent's configuration when creating a new child
56 * cpuset cgroup. For historical reasons, this option can be
57 * specified at mount time and thus is implemented here.
59 CGRP_CPUSET_CLONE_CHILDREN,
62 /* cgroup_root->flags */
64 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
65 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
70 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
71 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
72 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
73 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
75 /* internal flags, do not use outside cgroup core proper */
76 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
77 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
81 * cgroup_file is the handle for a file instance created in a cgroup which
82 * is used, for example, to generate file changed notifications. This can
83 * be obtained by setting cftype->file_offset.
86 /* do not access any fields from outside cgroup core */
87 struct kernfs_node *kn;
91 * Per-subsystem/per-cgroup state maintained by the system. This is the
92 * fundamental structural building block that controllers deal with.
94 * Fields marked with "PI:" are public and immutable and may be accessed
95 * directly without synchronization.
97 struct cgroup_subsys_state {
98 /* PI: the cgroup that this css is attached to */
99 struct cgroup *cgroup;
101 /* PI: the cgroup subsystem that this css is attached to */
102 struct cgroup_subsys *ss;
104 /* reference count - access via css_[try]get() and css_put() */
105 struct percpu_ref refcnt;
107 /* PI: the parent css */
108 struct cgroup_subsys_state *parent;
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.
130 /* percpu_ref killing and RCU release */
131 struct rcu_head rcu_head;
132 struct work_struct destroy_work;
136 * A css_set is a structure holding pointers to a set of
137 * cgroup_subsys_state objects. This saves space in the task struct
138 * object and speeds up fork()/exit(), since a single inc/dec and a
139 * list_add()/del() can bump the reference count on the entire cgroup
143 /* Reference count */
147 * List running through all cgroup groups in the same hash
148 * slot. Protected by css_set_lock
150 struct hlist_node hlist;
153 * Lists running through all tasks using this cgroup group.
154 * mg_tasks lists tasks which belong to this cset but are in the
155 * process of being migrated out or in. Protected by
156 * css_set_rwsem, but, during migration, once tasks are moved to
157 * mg_tasks, it can be read safely while holding cgroup_mutex.
159 struct list_head tasks;
160 struct list_head mg_tasks;
163 * List of cgrp_cset_links pointing at cgroups referenced from this
164 * css_set. Protected by css_set_lock.
166 struct list_head cgrp_links;
168 /* the default cgroup associated with this css_set */
169 struct cgroup *dfl_cgrp;
172 * Set of subsystem states, one for each subsystem. This array is
173 * immutable after creation apart from the init_css_set during
174 * subsystem registration (at boot time).
176 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
179 * List of csets participating in the on-going migration either as
180 * source or destination. Protected by cgroup_mutex.
182 struct list_head mg_preload_node;
183 struct list_head mg_node;
186 * If this cset is acting as the source of migration the following
187 * two fields are set. mg_src_cgrp is the source cgroup of the
188 * on-going migration and mg_dst_cset is the destination cset the
189 * target tasks on this cset should be migrated to. Protected by
192 struct cgroup *mg_src_cgrp;
193 struct css_set *mg_dst_cset;
196 * On the default hierarhcy, ->subsys[ssid] may point to a css
197 * attached to an ancestor instead of the cgroup this css_set is
198 * associated with. The following node is anchored at
199 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
200 * iterate through all css's attached to a given cgroup.
202 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
204 /* all css_task_iters currently walking this cset */
205 struct list_head task_iters;
207 /* For RCU-protected deletion */
208 struct rcu_head rcu_head;
212 /* self css with NULL ->ss, points back to this cgroup */
213 struct cgroup_subsys_state self;
215 unsigned long flags; /* "unsigned long" so bitops work */
218 * idr allocated in-hierarchy ID.
220 * ID 0 is not used, the ID of the root cgroup is always 1, and a
221 * new cgroup will be assigned with a smallest available ID.
223 * Allocating/Removing ID must be protected by cgroup_mutex.
228 * The depth this cgroup is at. The root is at depth zero and each
229 * step down the hierarchy increments the level. This along with
230 * ancestor_ids[] can determine whether a given cgroup is a
231 * descendant of another without traversing the hierarchy.
236 * Each non-empty css_set associated with this cgroup contributes
237 * one to populated_cnt. All children with non-zero popuplated_cnt
238 * of their own contribute one. The count is zero iff there's no
239 * task in this cgroup or its subtree.
243 struct kernfs_node *kn; /* cgroup kernfs entry */
244 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
245 struct cgroup_file events_file; /* handle for "cgroup.events" */
248 * The bitmask of subsystems enabled on the child cgroups.
249 * ->subtree_control is the one configured through
250 * "cgroup.subtree_control" while ->child_subsys_mask is the
251 * effective one which may have more subsystems enabled.
252 * Controller knobs are made available iff it's enabled in
255 unsigned int subtree_control;
256 unsigned int child_subsys_mask;
258 /* Private pointers for each registered subsystem */
259 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
261 struct cgroup_root *root;
264 * List of cgrp_cset_links pointing at css_sets with tasks in this
265 * cgroup. Protected by css_set_lock.
267 struct list_head cset_links;
270 * On the default hierarchy, a css_set for a cgroup with some
271 * susbsys disabled will point to css's which are associated with
272 * the closest ancestor which has the subsys enabled. The
273 * following lists all css_sets which point to this cgroup's css
274 * for the given subsystem.
276 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
279 * list of pidlists, up to two for each namespace (one for procs, one
280 * for tasks); created on demand.
282 struct list_head pidlists;
283 struct mutex pidlist_mutex;
285 /* used to wait for offlining of csses */
286 wait_queue_head_t offline_waitq;
288 /* used to schedule release agent */
289 struct work_struct release_agent_work;
291 /* ids of the ancestors at each level including self */
296 * A cgroup_root represents the root of a cgroup hierarchy, and may be
297 * associated with a kernfs_root to form an active hierarchy. This is
298 * internal to cgroup core. Don't access directly from controllers.
301 struct kernfs_root *kf_root;
303 /* The bitmask of subsystems attached to this hierarchy */
304 unsigned int subsys_mask;
306 /* Unique id for this hierarchy. */
309 /* The root cgroup. Root is destroyed on its release. */
312 /* for cgrp->ancestor_ids[0] */
313 int cgrp_ancestor_id_storage;
315 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
318 /* A list running through the active hierarchies */
319 struct list_head root_list;
321 /* Hierarchy-specific flags */
324 /* IDs for cgroups in this hierarchy */
325 struct idr cgroup_idr;
327 /* The path to use for release notifications. */
328 char release_agent_path[PATH_MAX];
330 /* The name for this hierarchy - may be empty */
331 char name[MAX_CGROUP_ROOT_NAMELEN];
335 * struct cftype: handler definitions for cgroup control files
337 * When reading/writing to a file:
338 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
339 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
343 * By convention, the name should begin with the name of the
344 * subsystem, followed by a period. Zero length string indicates
345 * end of cftype array.
347 char name[MAX_CFTYPE_NAME];
348 unsigned long private;
351 * The maximum length of string, excluding trailing nul, that can
352 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
354 size_t max_write_len;
360 * If non-zero, should contain the offset from the start of css to
361 * a struct cgroup_file field. cgroup will record the handle of
362 * the created file into it. The recorded handle can be used as
363 * long as the containing css remains accessible.
365 unsigned int file_offset;
368 * Fields used for internal bookkeeping. Initialized automatically
369 * during registration.
371 struct cgroup_subsys *ss; /* NULL for cgroup core files */
372 struct list_head node; /* anchored at ss->cfts */
373 struct kernfs_ops *kf_ops;
376 * read_u64() is a shortcut for the common case of returning a
377 * single integer. Use it in place of read()
379 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
381 * read_s64() is a signed version of read_u64()
383 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
385 /* generic seq_file read interface */
386 int (*seq_show)(struct seq_file *sf, void *v);
388 /* optional ops, implement all or none */
389 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
390 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
391 void (*seq_stop)(struct seq_file *sf, void *v);
394 * write_u64() is a shortcut for the common case of accepting
395 * a single integer (as parsed by simple_strtoull) from
396 * userspace. Use in place of write(); return 0 or error.
398 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
401 * write_s64() is a signed version of write_u64()
403 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
407 * write() is the generic write callback which maps directly to
408 * kernfs write operation and overrides all other operations.
409 * Maximum write size is determined by ->max_write_len. Use
410 * of_css/cft() to access the associated css and cft.
412 ssize_t (*write)(struct kernfs_open_file *of,
413 char *buf, size_t nbytes, loff_t off);
415 #ifdef CONFIG_DEBUG_LOCK_ALLOC
416 struct lock_class_key lockdep_key;
421 * Control Group subsystem type.
422 * See Documentation/cgroups/cgroups.txt for details
424 struct cgroup_subsys {
425 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
426 int (*css_online)(struct cgroup_subsys_state *css);
427 void (*css_offline)(struct cgroup_subsys_state *css);
428 void (*css_released)(struct cgroup_subsys_state *css);
429 void (*css_free)(struct cgroup_subsys_state *css);
430 void (*css_reset)(struct cgroup_subsys_state *css);
431 void (*css_e_css_changed)(struct cgroup_subsys_state *css);
433 int (*can_attach)(struct cgroup_taskset *tset);
434 void (*cancel_attach)(struct cgroup_taskset *tset);
435 void (*attach)(struct cgroup_taskset *tset);
436 int (*can_fork)(struct task_struct *task);
437 void (*cancel_fork)(struct task_struct *task);
438 void (*fork)(struct task_struct *task);
439 void (*exit)(struct task_struct *task);
440 void (*free)(struct task_struct *task);
441 void (*bind)(struct cgroup_subsys_state *root_css);
446 * If %false, this subsystem is properly hierarchical -
447 * configuration, resource accounting and restriction on a parent
448 * cgroup cover those of its children. If %true, hierarchy support
449 * is broken in some ways - some subsystems ignore hierarchy
450 * completely while others are only implemented half-way.
452 * It's now disallowed to create nested cgroups if the subsystem is
453 * broken and cgroup core will emit a warning message on such
454 * cases. Eventually, all subsystems will be made properly
455 * hierarchical and this will go away.
457 bool broken_hierarchy;
458 bool warned_broken_hierarchy;
460 /* the following two fields are initialized automtically during boot */
464 /* optional, initialized automatically during boot if not set */
465 const char *legacy_name;
467 /* link to parent, protected by cgroup_lock() */
468 struct cgroup_root *root;
470 /* idr for css->id */
474 * List of cftypes. Each entry is the first entry of an array
475 * terminated by zero length name.
477 struct list_head cfts;
480 * Base cftypes which are automatically registered. The two can
481 * point to the same array.
483 struct cftype *dfl_cftypes; /* for the default hierarchy */
484 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
487 * A subsystem may depend on other subsystems. When such subsystem
488 * is enabled on a cgroup, the depended-upon subsystems are enabled
489 * together if available. Subsystems enabled due to dependency are
490 * not visible to userland until explicitly enabled. The following
491 * specifies the mask of subsystems that this one depends on.
493 unsigned int depends_on;
496 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
499 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
502 * Called from threadgroup_change_begin() and allows cgroup operations to
503 * synchronize against threadgroup changes using a percpu_rw_semaphore.
505 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
507 percpu_down_read(&cgroup_threadgroup_rwsem);
511 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
514 * Called from threadgroup_change_end(). Counterpart of
515 * cgroup_threadcgroup_change_begin().
517 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
519 percpu_up_read(&cgroup_threadgroup_rwsem);
522 #else /* CONFIG_CGROUPS */
524 #define CGROUP_SUBSYS_COUNT 0
526 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {}
527 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
529 #endif /* CONFIG_CGROUPS */
531 #ifdef CONFIG_SOCK_CGROUP_DATA
534 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
535 * per-socket cgroup information except for memcg association.
537 * On legacy hierarchies, net_prio and net_cls controllers directly set
538 * attributes on each sock which can then be tested by the network layer.
539 * On the default hierarchy, each sock is associated with the cgroup it was
540 * created in and the networking layer can match the cgroup directly.
542 * To avoid carrying all three cgroup related fields separately in sock,
543 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
544 * On boot, sock_cgroup_data records the cgroup that the sock was created
545 * in so that cgroup2 matches can be made; however, once either net_prio or
546 * net_cls starts being used, the area is overriden to carry prioidx and/or
547 * classid. The two modes are distinguished by whether the lowest bit is
548 * set. Clear bit indicates cgroup pointer while set bit prioidx and
551 * While userland may start using net_prio or net_cls at any time, once
552 * either is used, cgroup2 matching no longer works. There is no reason to
553 * mix the two and this is in line with how legacy and v2 compatibility is
554 * handled. On mode switch, cgroup references which are already being
555 * pointed to by socks may be leaked. While this can be remedied by adding
556 * synchronization around sock_cgroup_data, given that the number of leaked
557 * cgroups is bound and highly unlikely to be high, this seems to be the
560 struct sock_cgroup_data {
562 #ifdef __LITTLE_ENDIAN
582 * There's a theoretical window where the following accessors race with
583 * updaters and return part of the previous pointer as the prioidx or
584 * classid. Such races are short-lived and the result isn't critical.
586 static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd)
588 /* fallback to 1 which is always the ID of the root cgroup */
589 return (skcd->is_data & 1) ? skcd->prioidx : 1;
592 static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd)
594 /* fallback to 0 which is the unconfigured default classid */
595 return (skcd->is_data & 1) ? skcd->classid : 0;
599 * If invoked concurrently, the updaters may clobber each other. The
600 * caller is responsible for synchronization.
602 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
605 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
607 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
610 if (!(skcd_buf.is_data & 1)) {
612 skcd_buf.is_data = 1;
615 skcd_buf.prioidx = prioidx;
616 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
619 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
622 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
624 if (sock_cgroup_classid(&skcd_buf) == classid)
627 if (!(skcd_buf.is_data & 1)) {
629 skcd_buf.is_data = 1;
632 skcd_buf.classid = classid;
633 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
636 #else /* CONFIG_SOCK_CGROUP_DATA */
638 struct sock_cgroup_data {
641 #endif /* CONFIG_SOCK_CGROUP_DATA */
643 #endif /* _LINUX_CGROUP_DEFS_H */