]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - include/linux/cgroup.h
cgroup: no need to check css refs for release notification
[mirror_ubuntu-zesty-kernel.git] / include / linux / cgroup.h
CommitLineData
ddbcc7e8
PM
1#ifndef _LINUX_CGROUP_H
2#define _LINUX_CGROUP_H
3/*
4 * cgroup interface
5 *
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
8 *
9 */
10
11#include <linux/sched.h>
ddbcc7e8
PM
12#include <linux/cpumask.h>
13#include <linux/nodemask.h>
14#include <linux/rcupdate.h>
eb6fd504 15#include <linux/rculist.h>
846c7bb0 16#include <linux/cgroupstats.h>
31a7df01 17#include <linux/prio_heap.h>
cc31edce 18#include <linux/rwsem.h>
38460b48 19#include <linux/idr.h>
48ddbe19 20#include <linux/workqueue.h>
03b1cde6 21#include <linux/xattr.h>
ddbcc7e8
PM
22
23#ifdef CONFIG_CGROUPS
24
25struct cgroupfs_root;
26struct cgroup_subsys;
27struct inode;
84eea842 28struct cgroup;
38460b48 29struct css_id;
ddbcc7e8
PM
30
31extern int cgroup_init_early(void);
32extern int cgroup_init(void);
ddbcc7e8 33extern void cgroup_lock(void);
d11c563d 34extern int cgroup_lock_is_held(void);
84eea842 35extern bool cgroup_lock_live_group(struct cgroup *cgrp);
ddbcc7e8 36extern void cgroup_unlock(void);
b4f48b63 37extern void cgroup_fork(struct task_struct *p);
817929ec 38extern void cgroup_post_fork(struct task_struct *p);
b4f48b63 39extern void cgroup_exit(struct task_struct *p, int run_callbacks);
846c7bb0
BS
40extern int cgroupstats_build(struct cgroupstats *stats,
41 struct dentry *dentry);
e6a1105b 42extern int cgroup_load_subsys(struct cgroup_subsys *ss);
cf5d5941 43extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
ddbcc7e8 44
828c0950 45extern const struct file_operations proc_cgroup_operations;
a424316c 46
aae8aab4 47/* Define the enumeration of all builtin cgroup subsystems */
817929ec 48#define SUBSYS(_x) _x ## _subsys_id,
8a8e04df 49#define IS_SUBSYS_ENABLED(option) IS_ENABLED(option)
817929ec
PM
50enum cgroup_subsys_id {
51#include <linux/cgroup_subsys.h>
a6f00298 52 CGROUP_SUBSYS_COUNT,
817929ec 53};
5fc0b025 54#undef IS_SUBSYS_ENABLED
817929ec
PM
55#undef SUBSYS
56
ddbcc7e8
PM
57/* Per-subsystem/per-cgroup state maintained by the system. */
58struct cgroup_subsys_state {
d20a390a
PM
59 /*
60 * The cgroup that this subsystem is attached to. Useful
ddbcc7e8 61 * for subsystems that want to know about the cgroup
d20a390a
PM
62 * hierarchy structure
63 */
ddbcc7e8
PM
64 struct cgroup *cgroup;
65
d20a390a
PM
66 /*
67 * State maintained by the cgroup system to allow subsystems
e7c5ec91 68 * to be "busy". Should be accessed via css_get(),
d0b2fdd2 69 * css_tryget() and css_put().
d20a390a 70 */
ddbcc7e8
PM
71
72 atomic_t refcnt;
73
74 unsigned long flags;
38460b48 75 /* ID for this css, if possible */
2c392b8c 76 struct css_id __rcu *id;
48ddbe19
TH
77
78 /* Used to put @cgroup->dentry on the last css_put() */
79 struct work_struct dput_work;
ddbcc7e8
PM
80};
81
82/* bits in struct cgroup_subsys_state flags field */
83enum {
38b53aba 84 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
92fb9748 85 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
ddbcc7e8
PM
86};
87
d7b9fff7
DN
88/* Caller must verify that the css is not for root cgroup */
89static inline void __css_get(struct cgroup_subsys_state *css, int count)
90{
91 atomic_add(count, &css->refcnt);
92}
93
ddbcc7e8 94/*
e7c5ec91
PM
95 * Call css_get() to hold a reference on the css; it can be used
96 * for a reference obtained via:
97 * - an existing ref-counted reference to the css
98 * - task->cgroups for a locked task
ddbcc7e8
PM
99 */
100
101static inline void css_get(struct cgroup_subsys_state *css)
102{
103 /* We don't need to reference count the root state */
38b53aba 104 if (!(css->flags & CSS_ROOT))
d7b9fff7 105 __css_get(css, 1);
ddbcc7e8 106}
e7c5ec91 107
e7c5ec91
PM
108/*
109 * Call css_tryget() to take a reference on a css if your existing
110 * (known-valid) reference isn't already ref-counted. Returns false if
111 * the css has been destroyed.
112 */
113
28b4c27b 114extern bool __css_tryget(struct cgroup_subsys_state *css);
e7c5ec91
PM
115static inline bool css_tryget(struct cgroup_subsys_state *css)
116{
38b53aba 117 if (css->flags & CSS_ROOT)
e7c5ec91 118 return true;
28b4c27b 119 return __css_tryget(css);
e7c5ec91
PM
120}
121
ddbcc7e8
PM
122/*
123 * css_put() should be called to release a reference taken by
e7c5ec91 124 * css_get() or css_tryget()
ddbcc7e8
PM
125 */
126
28b4c27b 127extern void __css_put(struct cgroup_subsys_state *css);
ddbcc7e8
PM
128static inline void css_put(struct cgroup_subsys_state *css)
129{
38b53aba 130 if (!(css->flags & CSS_ROOT))
28b4c27b 131 __css_put(css);
ddbcc7e8
PM
132}
133
3116f0e3
PM
134/* bits in struct cgroup flags field */
135enum {
136 /* Control Group is dead */
137 CGRP_REMOVED,
d20a390a
PM
138 /*
139 * Control Group has previously had a child cgroup or a task,
140 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
141 */
3116f0e3
PM
142 CGRP_RELEASABLE,
143 /* Control Group requires release notifications to userspace */
144 CGRP_NOTIFY_ON_RELEASE,
97978e6d 145 /*
2260e7fc
TH
146 * Clone the parent's configuration when creating a new child
147 * cpuset cgroup. For historical reasons, this option can be
148 * specified at mount time and thus is implemented here.
97978e6d 149 */
2260e7fc 150 CGRP_CPUSET_CLONE_CHILDREN,
3116f0e3
PM
151};
152
65dff759
LZ
153struct cgroup_name {
154 struct rcu_head rcu_head;
155 char name[];
156};
157
ddbcc7e8
PM
158struct cgroup {
159 unsigned long flags; /* "unsigned long" so bitops work */
160
d20a390a
PM
161 /*
162 * count users of this cgroup. >0 means busy, but doesn't
163 * necessarily indicate the number of tasks in the cgroup
164 */
ddbcc7e8
PM
165 atomic_t count;
166
0a950f65
TH
167 int id; /* ida allocated in-hierarchy ID */
168
ddbcc7e8
PM
169 /*
170 * We link our 'sibling' struct into our parent's 'children'.
171 * Our children link their 'sibling' into our 'children'.
172 */
173 struct list_head sibling; /* my parent's children */
174 struct list_head children; /* my children */
05ef1d7c 175 struct list_head files; /* my files */
ddbcc7e8 176
d20a390a 177 struct cgroup *parent; /* my parent */
febfcef6 178 struct dentry *dentry; /* cgroup fs entry, RCU protected */
ddbcc7e8 179
65dff759
LZ
180 /*
181 * This is a copy of dentry->d_name, and it's needed because
182 * we can't use dentry->d_name in cgroup_path().
183 *
184 * You must acquire rcu_read_lock() to access cgrp->name, and
185 * the only place that can change it is rename(), which is
186 * protected by parent dir's i_mutex.
187 *
188 * Normally you should use cgroup_name() wrapper rather than
189 * access it directly.
190 */
191 struct cgroup_name __rcu *name;
192
ddbcc7e8
PM
193 /* Private pointers for each registered subsystem */
194 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
195
196 struct cgroupfs_root *root;
197 struct cgroup *top_cgroup;
817929ec
PM
198
199 /*
200 * List of cg_cgroup_links pointing at css_sets with
201 * tasks in this cgroup. Protected by css_set_lock
202 */
203 struct list_head css_sets;
81a6a5cd 204
b0ca5a84 205 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
8e3f6541 206 struct list_head cft_q_node; /* used during cftype add/rm */
b0ca5a84 207
81a6a5cd
PM
208 /*
209 * Linked list running through all cgroups that can
210 * potentially be reaped by the release agent. Protected by
211 * release_list_lock
212 */
213 struct list_head release_list;
cc31edce 214
72a8cb30
BB
215 /*
216 * list of pidlists, up to two for each namespace (one for procs, one
217 * for tasks); created on demand.
218 */
219 struct list_head pidlists;
220 struct mutex pidlist_mutex;
a47295e6
PM
221
222 /* For RCU-protected deletion */
223 struct rcu_head rcu_head;
be445626 224 struct work_struct free_work;
0dea1168 225
25985edc 226 /* List of events which userspace want to receive */
0dea1168
KS
227 struct list_head event_list;
228 spinlock_t event_list_lock;
03b1cde6
AR
229
230 /* directory xattrs */
231 struct simple_xattrs xattrs;
817929ec
PM
232};
233
d20a390a
PM
234/*
235 * A css_set is a structure holding pointers to a set of
817929ec
PM
236 * cgroup_subsys_state objects. This saves space in the task struct
237 * object and speeds up fork()/exit(), since a single inc/dec and a
d20a390a
PM
238 * list_add()/del() can bump the reference count on the entire cgroup
239 * set for a task.
817929ec
PM
240 */
241
242struct css_set {
243
244 /* Reference count */
146aa1bd 245 atomic_t refcount;
817929ec 246
472b1053
LZ
247 /*
248 * List running through all cgroup groups in the same hash
249 * slot. Protected by css_set_lock
250 */
251 struct hlist_node hlist;
252
817929ec
PM
253 /*
254 * List running through all tasks using this cgroup
255 * group. Protected by css_set_lock
256 */
257 struct list_head tasks;
258
259 /*
260 * List of cg_cgroup_link objects on link chains from
261 * cgroups referenced from this css_set. Protected by
262 * css_set_lock
263 */
264 struct list_head cg_links;
265
266 /*
267 * Set of subsystem states, one for each subsystem. This array
268 * is immutable after creation apart from the init_css_set
cf5d5941
BB
269 * during subsystem registration (at boot time) and modular subsystem
270 * loading/unloading.
817929ec
PM
271 */
272 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
c378369d
BB
273
274 /* For RCU-protected deletion */
275 struct rcu_head rcu_head;
ddbcc7e8
PM
276};
277
91796569
PM
278/*
279 * cgroup_map_cb is an abstract callback API for reporting map-valued
280 * control files
281 */
282
283struct cgroup_map_cb {
284 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
285 void *state;
286};
287
d20a390a
PM
288/*
289 * struct cftype: handler definitions for cgroup control files
ddbcc7e8
PM
290 *
291 * When reading/writing to a file:
a043e3b2 292 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
ddbcc7e8
PM
293 * - the 'cftype' of the file is file->f_dentry->d_fsdata
294 */
295
8e3f6541
TH
296/* cftype->flags */
297#define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
d0b2fdd2 298#define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
8e3f6541
TH
299
300#define MAX_CFTYPE_NAME 64
301
ddbcc7e8 302struct cftype {
d20a390a
PM
303 /*
304 * By convention, the name should begin with the name of the
8e3f6541
TH
305 * subsystem, followed by a period. Zero length string indicates
306 * end of cftype array.
d20a390a 307 */
ddbcc7e8
PM
308 char name[MAX_CFTYPE_NAME];
309 int private;
099fca32
LZ
310 /*
311 * If not 0, file mode is set to this value, otherwise it will
312 * be figured out automatically
313 */
a5e7ed32 314 umode_t mode;
db3b1497
PM
315
316 /*
317 * If non-zero, defines the maximum length of string that can
318 * be passed to write_string; defaults to 64
319 */
320 size_t max_write_len;
321
8e3f6541
TH
322 /* CFTYPE_* flags */
323 unsigned int flags;
324
03b1cde6
AR
325 /* file xattrs */
326 struct simple_xattrs xattrs;
327
ce16b49d
PM
328 int (*open)(struct inode *inode, struct file *file);
329 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
330 struct file *file,
331 char __user *buf, size_t nbytes, loff_t *ppos);
ddbcc7e8 332 /*
f4c753b7 333 * read_u64() is a shortcut for the common case of returning a
ddbcc7e8
PM
334 * single integer. Use it in place of read()
335 */
ce16b49d 336 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
e73d2c61
PM
337 /*
338 * read_s64() is a signed version of read_u64()
339 */
ce16b49d 340 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
91796569
PM
341 /*
342 * read_map() is used for defining a map of key/value
343 * pairs. It should call cb->fill(cb, key, value) for each
344 * entry. The key/value pairs (and their ordering) should not
345 * change between reboots.
346 */
ce16b49d
PM
347 int (*read_map)(struct cgroup *cont, struct cftype *cft,
348 struct cgroup_map_cb *cb);
29486df3
SH
349 /*
350 * read_seq_string() is used for outputting a simple sequence
351 * using seqfile.
352 */
ce16b49d
PM
353 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
354 struct seq_file *m);
91796569 355
ce16b49d
PM
356 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
357 struct file *file,
358 const char __user *buf, size_t nbytes, loff_t *ppos);
355e0c48
PM
359
360 /*
f4c753b7 361 * write_u64() is a shortcut for the common case of accepting
355e0c48
PM
362 * a single integer (as parsed by simple_strtoull) from
363 * userspace. Use in place of write(); return 0 or error.
364 */
ce16b49d 365 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
e73d2c61
PM
366 /*
367 * write_s64() is a signed version of write_u64()
368 */
ce16b49d 369 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
355e0c48 370
db3b1497
PM
371 /*
372 * write_string() is passed a nul-terminated kernelspace
373 * buffer of maximum length determined by max_write_len.
374 * Returns 0 or -ve error code.
375 */
376 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
377 const char *buffer);
d447ea2f
PE
378 /*
379 * trigger() callback can be used to get some kick from the
380 * userspace, when the actual string written is not important
381 * at all. The private field can be used to determine the
382 * kick type for multiplexing.
383 */
384 int (*trigger)(struct cgroup *cgrp, unsigned int event);
385
ce16b49d 386 int (*release)(struct inode *inode, struct file *file);
0dea1168
KS
387
388 /*
389 * register_event() callback will be used to add new userspace
390 * waiter for changes related to the cftype. Implement it if
391 * you want to provide this functionality. Use eventfd_signal()
392 * on eventfd to send notification to userspace.
393 */
394 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
395 struct eventfd_ctx *eventfd, const char *args);
396 /*
397 * unregister_event() callback will be called when userspace
398 * closes the eventfd or on cgroup removing.
399 * This callback must be implemented, if you want provide
400 * notification functionality.
0dea1168 401 */
907860ed 402 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
0dea1168 403 struct eventfd_ctx *eventfd);
ddbcc7e8
PM
404};
405
8e3f6541
TH
406/*
407 * cftype_sets describe cftypes belonging to a subsystem and are chained at
408 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
409 * terminated by zero length name.
410 */
411struct cftype_set {
412 struct list_head node; /* chained at subsys->cftsets */
03b1cde6 413 struct cftype *cfts;
8e3f6541
TH
414};
415
31a7df01
CW
416struct cgroup_scanner {
417 struct cgroup *cg;
418 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
419 void (*process_task)(struct task_struct *p,
420 struct cgroup_scanner *scan);
421 struct ptr_heap *heap;
bd1a8ab7 422 void *data;
31a7df01
CW
423};
424
65dff759
LZ
425/* Caller should hold rcu_read_lock() */
426static inline const char *cgroup_name(const struct cgroup *cgrp)
427{
428 return rcu_dereference(cgrp->name)->name;
429}
430
03b1cde6
AR
431int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
432int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
8e3f6541 433
ffd2d883 434int cgroup_is_removed(const struct cgroup *cgrp);
ddbcc7e8 435
ffd2d883 436int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
ddbcc7e8 437
ffd2d883 438int cgroup_task_count(const struct cgroup *cgrp);
bbcb81d0 439
313e924c
GN
440/* Return true if cgrp is a descendant of the task's cgroup */
441int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
ddbcc7e8 442
2f7ee569
TH
443/*
444 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
445 * methods.
446 */
447struct cgroup_taskset;
448struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
449struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
450struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
451int cgroup_taskset_size(struct cgroup_taskset *tset);
452
453/**
454 * cgroup_taskset_for_each - iterate cgroup_taskset
455 * @task: the loop cursor
456 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
457 * @tset: taskset to iterate
458 */
459#define cgroup_taskset_for_each(task, skip_cgrp, tset) \
460 for ((task) = cgroup_taskset_first((tset)); (task); \
461 (task) = cgroup_taskset_next((tset))) \
462 if (!(skip_cgrp) || \
463 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
464
21acb9ca
TLSC
465/*
466 * Control Group subsystem type.
467 * See Documentation/cgroups/cgroups.txt for details
468 */
ddbcc7e8
PM
469
470struct cgroup_subsys {
92fb9748
TH
471 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
472 int (*css_online)(struct cgroup *cgrp);
473 void (*css_offline)(struct cgroup *cgrp);
474 void (*css_free)(struct cgroup *cgrp);
475
761b3ef5
LZ
476 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
477 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
478 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
479 void (*fork)(struct task_struct *task);
480 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
481 struct task_struct *task);
761b3ef5 482 void (*bind)(struct cgroup *root);
e5991371 483
ddbcc7e8
PM
484 int subsys_id;
485 int active;
8bab8dde 486 int disabled;
ddbcc7e8 487 int early_init;
38460b48
KH
488 /*
489 * True if this subsys uses ID. ID is not available before cgroup_init()
490 * (not available in early_init time.)
491 */
492 bool use_id;
48ddbe19 493
8c7f6edb
TH
494 /*
495 * If %false, this subsystem is properly hierarchical -
496 * configuration, resource accounting and restriction on a parent
497 * cgroup cover those of its children. If %true, hierarchy support
498 * is broken in some ways - some subsystems ignore hierarchy
499 * completely while others are only implemented half-way.
500 *
501 * It's now disallowed to create nested cgroups if the subsystem is
502 * broken and cgroup core will emit a warning message on such
503 * cases. Eventually, all subsystems will be made properly
504 * hierarchical and this will go away.
505 */
506 bool broken_hierarchy;
507 bool warned_broken_hierarchy;
508
ddbcc7e8
PM
509#define MAX_CGROUP_TYPE_NAMELEN 32
510 const char *name;
511
999cd8a4
PM
512 /*
513 * Link to parent, and list entry in parent's children.
6be96a5c 514 * Protected by cgroup_lock()
999cd8a4
PM
515 */
516 struct cgroupfs_root *root;
ddbcc7e8 517 struct list_head sibling;
38460b48
KH
518 /* used when use_id == true */
519 struct idr idr;
42aee6c4 520 spinlock_t id_lock;
e6a1105b 521
8e3f6541
TH
522 /* list of cftype_sets */
523 struct list_head cftsets;
524
525 /* base cftypes, automatically [de]registered with subsys itself */
526 struct cftype *base_cftypes;
527 struct cftype_set base_cftset;
528
e6a1105b
BB
529 /* should be defined only by modular subsystems */
530 struct module *module;
ddbcc7e8
PM
531};
532
533#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
5fc0b025 534#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
ddbcc7e8 535#include <linux/cgroup_subsys.h>
5fc0b025 536#undef IS_SUBSYS_ENABLED
ddbcc7e8
PM
537#undef SUBSYS
538
539static inline struct cgroup_subsys_state *cgroup_subsys_state(
ffd2d883 540 struct cgroup *cgrp, int subsys_id)
ddbcc7e8 541{
ffd2d883 542 return cgrp->subsys[subsys_id];
ddbcc7e8
PM
543}
544
dc61b1d6
PZ
545/*
546 * function to get the cgroup_subsys_state which allows for extra
547 * rcu_dereference_check() conditions, such as locks used during the
548 * cgroup_subsys::attach() methods.
549 */
550#define task_subsys_state_check(task, subsys_id, __c) \
551 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
dc61b1d6
PZ
552 lockdep_is_held(&task->alloc_lock) || \
553 cgroup_lock_is_held() || (__c))
554
555static inline struct cgroup_subsys_state *
556task_subsys_state(struct task_struct *task, int subsys_id)
ddbcc7e8 557{
dc61b1d6 558 return task_subsys_state_check(task, subsys_id, false);
ddbcc7e8
PM
559}
560
561static inline struct cgroup* task_cgroup(struct task_struct *task,
562 int subsys_id)
563{
564 return task_subsys_state(task, subsys_id)->cgroup;
565}
566
574bd9f7
TH
567/**
568 * cgroup_for_each_child - iterate through children of a cgroup
569 * @pos: the cgroup * to use as the loop cursor
570 * @cgroup: cgroup whose children to walk
571 *
572 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
92fb9748
TH
573 * cgroup which hasn't finished ->css_online() or already has finished
574 * ->css_offline() may show up during traversal and it's each subsystem's
574bd9f7
TH
575 * responsibility to verify that each @pos is alive.
576 *
92fb9748
TH
577 * If a subsystem synchronizes against the parent in its ->css_online() and
578 * before starting iterating, a cgroup which finished ->css_online() is
579 * guaranteed to be visible in the future iterations.
574bd9f7
TH
580 */
581#define cgroup_for_each_child(pos, cgroup) \
582 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
583
584struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
585 struct cgroup *cgroup);
12a9d2fe 586struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
574bd9f7
TH
587
588/**
589 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
590 * @pos: the cgroup * to use as the loop cursor
591 * @cgroup: cgroup whose descendants to walk
592 *
593 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
92fb9748
TH
594 * descendant cgroup which hasn't finished ->css_online() or already has
595 * finished ->css_offline() may show up during traversal and it's each
574bd9f7
TH
596 * subsystem's responsibility to verify that each @pos is alive.
597 *
92fb9748
TH
598 * If a subsystem synchronizes against the parent in its ->css_online() and
599 * before starting iterating, and synchronizes against @pos on each
600 * iteration, any descendant cgroup which finished ->css_offline() is
574bd9f7
TH
601 * guaranteed to be visible in the future iterations.
602 *
603 * In other words, the following guarantees that a descendant can't escape
604 * state updates of its ancestors.
605 *
92fb9748 606 * my_online(@cgrp)
574bd9f7
TH
607 * {
608 * Lock @cgrp->parent and @cgrp;
609 * Inherit state from @cgrp->parent;
610 * Unlock both.
611 * }
612 *
613 * my_update_state(@cgrp)
614 * {
615 * Lock @cgrp;
616 * Update @cgrp's state;
617 * Unlock @cgrp;
618 *
619 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
620 * Lock @pos;
621 * Verify @pos is alive and inherit state from @pos->parent;
622 * Unlock @pos;
623 * }
624 * }
625 *
626 * As long as the inheriting step, including checking the parent state, is
627 * enclosed inside @pos locking, double-locking the parent isn't necessary
628 * while inheriting. The state update to the parent is guaranteed to be
629 * visible by walking order and, as long as inheriting operations to the
630 * same @pos are atomic to each other, multiple updates racing each other
631 * still result in the correct state. It's guaranateed that at least one
632 * inheritance happens for any cgroup after the latest update to its
633 * parent.
634 *
635 * If checking parent's state requires locking the parent, each inheriting
636 * iteration should lock and unlock both @pos->parent and @pos.
637 *
638 * Alternatively, a subsystem may choose to use a single global lock to
92fb9748 639 * synchronize ->css_online() and ->css_offline() against tree-walking
574bd9f7
TH
640 * operations.
641 */
642#define cgroup_for_each_descendant_pre(pos, cgroup) \
643 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
644 pos = cgroup_next_descendant_pre((pos), (cgroup)))
645
646struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
647 struct cgroup *cgroup);
648
649/**
650 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
651 * @pos: the cgroup * to use as the loop cursor
652 * @cgroup: cgroup whose descendants to walk
653 *
654 * Similar to cgroup_for_each_descendant_pre() but performs post-order
655 * traversal instead. Note that the walk visibility guarantee described in
656 * pre-order walk doesn't apply the same to post-order walks.
657 */
658#define cgroup_for_each_descendant_post(pos, cgroup) \
659 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
660 pos = cgroup_next_descendant_post((pos), (cgroup)))
661
817929ec
PM
662/* A cgroup_iter should be treated as an opaque object */
663struct cgroup_iter {
664 struct list_head *cg_link;
665 struct list_head *task;
666};
667
d20a390a
PM
668/*
669 * To iterate across the tasks in a cgroup:
817929ec 670 *
b595076a 671 * 1) call cgroup_iter_start to initialize an iterator
817929ec
PM
672 *
673 * 2) call cgroup_iter_next() to retrieve member tasks until it
674 * returns NULL or until you want to end the iteration
675 *
676 * 3) call cgroup_iter_end() to destroy the iterator.
31a7df01 677 *
d20a390a
PM
678 * Or, call cgroup_scan_tasks() to iterate through every task in a
679 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
680 * the test_task() callback, but not while calling the process_task()
681 * callback.
817929ec 682 */
ffd2d883
LZ
683void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
684struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
817929ec 685 struct cgroup_iter *it);
ffd2d883 686void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
31a7df01 687int cgroup_scan_tasks(struct cgroup_scanner *scan);
956db3ca 688int cgroup_attach_task(struct cgroup *, struct task_struct *);
31583bb0
MT
689int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
690
38460b48
KH
691/*
692 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
693 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
694 * CSS ID is assigned at cgroup allocation (create) automatically
695 * and removed when subsys calls free_css_id() function. This is because
696 * the lifetime of cgroup_subsys_state is subsys's matter.
697 *
698 * Looking up and scanning function should be called under rcu_read_lock().
6be96a5c 699 * Taking cgroup_mutex is not necessary for following calls.
38460b48
KH
700 * But the css returned by this routine can be "not populated yet" or "being
701 * destroyed". The caller should check css and cgroup's status.
702 */
703
704/*
705 * Typically Called at ->destroy(), or somewhere the subsys frees
706 * cgroup_subsys_state.
707 */
708void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
709
710/* Find a cgroup_subsys_state which has given ID */
711
712struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
713
714/*
715 * Get a cgroup whose id is greater than or equal to id under tree of root.
716 * Returning a cgroup_subsys_state or NULL.
717 */
718struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
719 struct cgroup_subsys_state *root, int *foundid);
720
721/* Returns true if root is ancestor of cg */
722bool css_is_ancestor(struct cgroup_subsys_state *cg,
0b7f569e 723 const struct cgroup_subsys_state *root);
38460b48
KH
724
725/* Get id and depth of css */
726unsigned short css_id(struct cgroup_subsys_state *css);
727unsigned short css_depth(struct cgroup_subsys_state *css);
e5d1367f 728struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
38460b48 729
ddbcc7e8
PM
730#else /* !CONFIG_CGROUPS */
731
732static inline int cgroup_init_early(void) { return 0; }
733static inline int cgroup_init(void) { return 0; }
b4f48b63 734static inline void cgroup_fork(struct task_struct *p) {}
817929ec 735static inline void cgroup_post_fork(struct task_struct *p) {}
b4f48b63 736static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
ddbcc7e8
PM
737
738static inline void cgroup_lock(void) {}
739static inline void cgroup_unlock(void) {}
846c7bb0
BS
740static inline int cgroupstats_build(struct cgroupstats *stats,
741 struct dentry *dentry)
742{
743 return -EINVAL;
744}
ddbcc7e8 745
d7926ee3 746/* No cgroups - nothing to do */
31583bb0
MT
747static inline int cgroup_attach_task_all(struct task_struct *from,
748 struct task_struct *t)
749{
750 return 0;
751}
d7926ee3 752
ddbcc7e8
PM
753#endif /* !CONFIG_CGROUPS */
754
755#endif /* _LINUX_CGROUP_H */