blob: ada239253ec748d8dc5aa9ac89f612cc8a6cbea9 [file] [log] [blame]
Paul Menageddbcc7e2007-10-18 23:39:30 -07001#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>
Paul Menageddbcc7e2007-10-18 23:39:30 -070012#include <linux/cpumask.h>
13#include <linux/nodemask.h>
14#include <linux/rcupdate.h>
Tejun Heoeb6fd502012-11-09 09:12:29 -080015#include <linux/rculist.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070016#include <linux/cgroupstats.h>
Paul Menagecc31edc2008-10-18 20:28:04 -070017#include <linux/rwsem.h>
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070018#include <linux/idr.h>
Tejun Heo48ddbe12012-04-01 12:09:56 -070019#include <linux/workqueue.h>
Tejun Heo25a7e682013-04-14 20:15:25 -070020#include <linux/fs.h>
Tejun Heod3daf282013-06-13 19:39:16 -070021#include <linux/percpu-refcount.h>
Tejun Heo7da11272013-12-05 12:28:04 -050022#include <linux/seq_file.h>
Tejun Heo2bd59d42014-02-11 11:52:49 -050023#include <linux/kernfs.h>
Tejun Heof8f22e52014-04-23 11:13:16 -040024#include <linux/wait.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070025
26#ifdef CONFIG_CGROUPS
27
Tejun Heo3dd06ff2014-03-19 10:23:54 -040028struct cgroup_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -070029struct cgroup_subsys;
30struct inode;
Paul Menage84eea842008-07-25 01:47:00 -070031struct cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -070032
33extern int cgroup_init_early(void);
34extern int cgroup_init(void);
Paul Menageb4f48b62007-10-18 23:39:33 -070035extern void cgroup_fork(struct task_struct *p);
Paul Menage817929e2007-10-18 23:39:36 -070036extern void cgroup_post_fork(struct task_struct *p);
Li Zefan1ec41832014-03-28 15:22:19 +080037extern void cgroup_exit(struct task_struct *p);
Balbir Singh846c7bb2007-10-18 23:39:44 -070038extern int cgroupstats_build(struct cgroupstats *stats,
39 struct dentry *dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -070040
Al Viro8d8b97b2013-04-19 23:11:24 -040041extern int proc_cgroup_show(struct seq_file *, void *);
Paul Menagea4243162007-10-18 23:39:35 -070042
Tejun Heo3ed80a62014-02-08 10:36:58 -050043/* define the enumeration of all cgroup subsystems */
Tejun Heo073219e2014-02-08 10:36:58 -050044#define SUBSYS(_x) _x ## _cgrp_id,
Paul Menage817929e2007-10-18 23:39:36 -070045enum cgroup_subsys_id {
46#include <linux/cgroup_subsys.h>
Daniel Wagnera6f00292012-09-12 16:12:08 +020047 CGROUP_SUBSYS_COUNT,
Paul Menage817929e2007-10-18 23:39:36 -070048};
49#undef SUBSYS
50
Paul Menageddbcc7e2007-10-18 23:39:30 -070051/* Per-subsystem/per-cgroup state maintained by the system. */
52struct cgroup_subsys_state {
Tejun Heo72c97e52013-08-08 20:11:22 -040053 /* the cgroup that this css is attached to */
Paul Menageddbcc7e2007-10-18 23:39:30 -070054 struct cgroup *cgroup;
55
Tejun Heo72c97e52013-08-08 20:11:22 -040056 /* the cgroup subsystem that this css is attached to */
57 struct cgroup_subsys *ss;
58
Tejun Heod3daf282013-06-13 19:39:16 -070059 /* reference count - access via css_[try]get() and css_put() */
60 struct percpu_ref refcnt;
Paul Menageddbcc7e2007-10-18 23:39:30 -070061
Tejun Heo0ae78e02013-08-13 11:01:54 -040062 /* the parent css */
63 struct cgroup_subsys_state *parent;
64
Paul Menageddbcc7e2007-10-18 23:39:30 -070065 unsigned long flags;
Tejun Heo48ddbe12012-04-01 12:09:56 -070066
Tejun Heo0c21ead2013-08-13 20:22:51 -040067 /* percpu_ref killing and RCU release */
68 struct rcu_head rcu_head;
Tejun Heo35ef10d2013-08-13 11:01:54 -040069 struct work_struct destroy_work;
Paul Menageddbcc7e2007-10-18 23:39:30 -070070};
71
72/* bits in struct cgroup_subsys_state flags field */
73enum {
Tejun Heo38b53ab2012-11-19 08:13:36 -080074 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
Tejun Heo92fb9742012-11-19 08:13:38 -080075 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
Paul Menageddbcc7e2007-10-18 23:39:30 -070076};
77
Tejun Heo5de01072013-06-12 21:04:52 -070078/**
79 * css_get - obtain a reference on the specified css
80 * @css: target css
81 *
82 * The caller must already have a reference.
Paul Menageddbcc7e2007-10-18 23:39:30 -070083 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070084static inline void css_get(struct cgroup_subsys_state *css)
85{
86 /* We don't need to reference count the root state */
Tejun Heo38b53ab2012-11-19 08:13:36 -080087 if (!(css->flags & CSS_ROOT))
Tejun Heod3daf282013-06-13 19:39:16 -070088 percpu_ref_get(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -070089}
Paul Menagee7c5ec92009-01-07 18:08:38 -080090
Tejun Heo5de01072013-06-12 21:04:52 -070091/**
92 * css_tryget - try to obtain a reference on the specified css
93 * @css: target css
94 *
95 * Obtain a reference on @css if it's alive. The caller naturally needs to
96 * ensure that @css is accessible but doesn't have to be holding a
97 * reference on it - IOW, RCU protected access is good enough for this
98 * function. Returns %true if a reference count was successfully obtained;
99 * %false otherwise.
100 */
Paul Menagee7c5ec92009-01-07 18:08:38 -0800101static inline bool css_tryget(struct cgroup_subsys_state *css)
102{
Tejun Heo38b53ab2012-11-19 08:13:36 -0800103 if (css->flags & CSS_ROOT)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800104 return true;
Tejun Heod3daf282013-06-13 19:39:16 -0700105 return percpu_ref_tryget(&css->refcnt);
Paul Menagee7c5ec92009-01-07 18:08:38 -0800106}
107
Tejun Heo5de01072013-06-12 21:04:52 -0700108/**
109 * css_put - put a css reference
110 * @css: target css
111 *
112 * Put a reference obtained via css_get() and css_tryget().
113 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700114static inline void css_put(struct cgroup_subsys_state *css)
115{
Tejun Heo38b53ab2012-11-19 08:13:36 -0800116 if (!(css->flags & CSS_ROOT))
Tejun Heod3daf282013-06-13 19:39:16 -0700117 percpu_ref_put(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700118}
119
Paul Menage3116f0e2008-04-29 01:00:04 -0700120/* bits in struct cgroup flags field */
121enum {
122 /* Control Group is dead */
Tejun Heo54766d42013-06-12 21:04:53 -0700123 CGRP_DEAD,
Paul Menaged20a3902009-04-02 16:57:22 -0700124 /*
125 * Control Group has previously had a child cgroup or a task,
126 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
127 */
Paul Menage3116f0e2008-04-29 01:00:04 -0700128 CGRP_RELEASABLE,
129 /* Control Group requires release notifications to userspace */
130 CGRP_NOTIFY_ON_RELEASE,
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700131 /*
Tejun Heo2260e7f2012-11-19 08:13:38 -0800132 * Clone the parent's configuration when creating a new child
133 * cpuset cgroup. For historical reasons, this option can be
134 * specified at mount time and thus is implemented here.
Daniel Lezcano97978e62010-10-27 15:33:35 -0700135 */
Tejun Heo2260e7f2012-11-19 08:13:38 -0800136 CGRP_CPUSET_CLONE_CHILDREN,
Tejun Heo873fe092013-04-14 20:15:26 -0700137 /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */
138 CGRP_SANE_BEHAVIOR,
Paul Menage3116f0e2008-04-29 01:00:04 -0700139};
140
Paul Menageddbcc7e2007-10-18 23:39:30 -0700141struct cgroup {
142 unsigned long flags; /* "unsigned long" so bitops work */
143
Li Zefanb414dc02013-07-31 09:51:06 +0800144 /*
145 * idr allocated in-hierarchy ID.
146 *
147 * The ID of the root cgroup is always 0, and a new cgroup
148 * will be assigned with a smallest available ID.
Li Zefan0ab02ca2014-02-11 16:05:46 +0800149 *
150 * Allocating/Removing ID must be protected by cgroup_mutex.
Li Zefanb414dc02013-07-31 09:51:06 +0800151 */
152 int id;
Tejun Heo0a950f62012-11-19 09:02:12 -0800153
Tejun Heof20104d2013-08-13 20:22:50 -0400154 /* the number of attached css's */
155 int nr_css;
156
Tejun Heo2bd59d42014-02-11 11:52:49 -0500157 atomic_t refcnt;
158
Paul Menageddbcc7e2007-10-18 23:39:30 -0700159 /*
160 * We link our 'sibling' struct into our parent's 'children'.
161 * Our children link their 'sibling' into our 'children'.
162 */
163 struct list_head sibling; /* my parent's children */
164 struct list_head children; /* my children */
165
Paul Menaged20a3902009-04-02 16:57:22 -0700166 struct cgroup *parent; /* my parent */
Tejun Heo2bd59d42014-02-11 11:52:49 -0500167 struct kernfs_node *kn; /* cgroup kernfs entry */
Tejun Heof8f22e52014-04-23 11:13:16 -0400168 struct kernfs_node *control_kn; /* kn for "cgroup.subtree_control" */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700169
Li Zefan65dff752013-03-01 15:01:56 +0800170 /*
Tejun Heo53fa5262013-05-24 10:55:38 +0900171 * Monotonically increasing unique serial number which defines a
172 * uniform order among all cgroups. It's guaranteed that all
173 * ->children lists are in the ascending order of ->serial_nr.
174 * It's used to allow interrupting and resuming iterations.
175 */
176 u64 serial_nr;
177
Tejun Heof392e512014-04-23 11:13:14 -0400178 /* the bitmask of subsystems enabled on the child cgroups */
179 unsigned long child_subsys_mask;
Tejun Heo94419622014-03-19 10:23:54 -0400180
Paul Menageddbcc7e2007-10-18 23:39:30 -0700181 /* Private pointers for each registered subsystem */
Tejun Heo73e80ed2013-08-13 11:01:55 -0400182 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700183
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400184 struct cgroup_root *root;
Paul Menage817929e2007-10-18 23:39:36 -0700185
186 /*
Tejun Heo69d02062013-06-12 21:04:50 -0700187 * List of cgrp_cset_links pointing at css_sets with tasks in this
188 * cgroup. Protected by css_set_lock.
Paul Menage817929e2007-10-18 23:39:36 -0700189 */
Tejun Heo69d02062013-06-12 21:04:50 -0700190 struct list_head cset_links;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700191
192 /*
Tejun Heo2d8f2432014-04-23 11:13:15 -0400193 * On the default hierarchy, a css_set for a cgroup with some
194 * susbsys disabled will point to css's which are associated with
195 * the closest ancestor which has the subsys enabled. The
196 * following lists all css_sets which point to this cgroup's css
197 * for the given subsystem.
198 */
199 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
200
201 /*
Paul Menage81a6a5c2007-10-18 23:39:38 -0700202 * Linked list running through all cgroups that can
203 * potentially be reaped by the release agent. Protected by
204 * release_list_lock
205 */
206 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700207
Ben Blum72a8cb32009-09-23 15:56:27 -0700208 /*
209 * list of pidlists, up to two for each namespace (one for procs, one
210 * for tasks); created on demand.
211 */
212 struct list_head pidlists;
213 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800214
Tejun Heo67f4c362013-08-08 20:11:24 -0400215 /* dummy css with NULL ->ss, points back to this cgroup */
216 struct cgroup_subsys_state dummy_css;
217
Tejun Heod3daf282013-06-13 19:39:16 -0700218 /* For css percpu_ref killing and RCU-protected deletion */
Paul Menagea47295e2009-01-07 18:07:44 -0800219 struct rcu_head rcu_head;
Tejun Heoea15f8c2013-06-13 19:27:42 -0700220 struct work_struct destroy_work;
Tejun Heof8f22e52014-04-23 11:13:16 -0400221
222 /* used to wait for offlining of csses */
223 wait_queue_head_t offline_waitq;
Paul Menage817929e2007-10-18 23:39:36 -0700224};
225
Tejun Heo25a7e682013-04-14 20:15:25 -0700226#define MAX_CGROUP_ROOT_NAMELEN 64
227
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400228/* cgroup_root->flags */
Tejun Heo25a7e682013-04-14 20:15:25 -0700229enum {
Tejun Heo873fe092013-04-14 20:15:26 -0700230 /*
231 * Unfortunately, cgroup core and various controllers are riddled
232 * with idiosyncrasies and pointless options. The following flag,
233 * when set, will force sane behavior - some options are forced on,
234 * others are disallowed, and some controllers will change their
235 * hierarchical or other behaviors.
236 *
237 * The set of behaviors affected by this flag are still being
238 * determined and developed and the mount option for this flag is
239 * prefixed with __DEVEL__. The prefix will be dropped once we
240 * reach the point where all behaviors are compatible with the
241 * planned unified hierarchy, which will automatically turn on this
242 * flag.
243 *
244 * The followings are the behaviors currently affected this flag.
245 *
Tejun Heod3ba07c2014-02-13 06:58:38 -0500246 * - Mount options "noprefix", "xattr", "clone_children",
247 * "release_agent" and "name" are disallowed.
Tejun Heo873fe092013-04-14 20:15:26 -0700248 *
249 * - When mounting an existing superblock, mount options should
250 * match.
251 *
252 * - Remount is disallowed.
253 *
Linus Torvalds0b0585c2013-07-02 20:04:25 -0700254 * - rename(2) is disallowed.
255 *
Tejun Heof63674f2013-06-13 19:58:38 -0700256 * - "tasks" is removed. Everything should be at process
257 * granularity. Use "cgroup.procs" instead.
258 *
Tejun Heoafb2bc12013-11-29 10:42:59 -0500259 * - "cgroup.procs" is not sorted. pids will be unique unless they
260 * got recycled inbetween reads.
261 *
Tejun Heof63674f2013-06-13 19:58:38 -0700262 * - "release_agent" and "notify_on_release" are removed.
263 * Replacement notification mechanism will be implemented.
264 *
Tejun Heod3ba07c2014-02-13 06:58:38 -0500265 * - "cgroup.clone_children" is removed.
Tejun Heo86bf4b62014-02-12 09:29:48 -0500266 *
Tejun Heoa2dd4242014-03-19 10:23:55 -0400267 * - If mount is requested with sane_behavior but without any
268 * subsystem, the default unified hierarchy is mounted.
269 *
Li Zefan5c5cc622013-06-09 17:16:29 +0800270 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
271 * and take masks of ancestors with non-empty cpus/mems, instead of
272 * being moved to an ancestor.
273 *
Li Zefan88fa5232013-06-09 17:16:46 +0800274 * - cpuset: a task can be moved into an empty cpuset, and again it
275 * takes masks of ancestors.
Tejun Heo6db8e852013-06-14 11:18:22 -0700276 *
Tejun Heof00baae2013-04-15 13:41:15 -0700277 * - memcg: use_hierarchy is on by default and the cgroup file for
278 * the flag is not created.
279 *
Tejun Heo91381252013-05-14 13:52:38 -0700280 * - blkcg: blk-throttle becomes properly hierarchical.
Tejun Heo873fe092013-04-14 20:15:26 -0700281 */
282 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
283
Tejun Heo25a7e682013-04-14 20:15:25 -0700284 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
285 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
Tejun Heo0ce6cba2013-06-27 19:37:26 -0700286
287 /* mount options live below bit 16 */
288 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
Tejun Heo25a7e682013-04-14 20:15:25 -0700289};
290
291/*
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400292 * A cgroup_root represents the root of a cgroup hierarchy, and may be
Tejun Heo2bd59d42014-02-11 11:52:49 -0500293 * associated with a kernfs_root to form an active hierarchy. This is
Tejun Heo25a7e682013-04-14 20:15:25 -0700294 * internal to cgroup core. Don't access directly from controllers.
295 */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400296struct cgroup_root {
Tejun Heo2bd59d42014-02-11 11:52:49 -0500297 struct kernfs_root *kf_root;
Tejun Heo25a7e682013-04-14 20:15:25 -0700298
Tejun Heof392e512014-04-23 11:13:14 -0400299 /* The bitmask of subsystems attached to this hierarchy */
300 unsigned long subsys_mask;
301
Tejun Heo25a7e682013-04-14 20:15:25 -0700302 /* Unique id for this hierarchy. */
303 int hierarchy_id;
304
Tejun Heo776f02f2014-02-12 09:29:50 -0500305 /* The root cgroup. Root is destroyed on its release. */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400306 struct cgroup cgrp;
Tejun Heo25a7e682013-04-14 20:15:25 -0700307
Tejun Heo3c9c8252014-02-12 09:29:50 -0500308 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
309 atomic_t nr_cgrps;
Tejun Heo25a7e682013-04-14 20:15:25 -0700310
311 /* A list running through the active hierarchies */
312 struct list_head root_list;
313
Tejun Heo25a7e682013-04-14 20:15:25 -0700314 /* Hierarchy-specific flags */
315 unsigned long flags;
316
317 /* IDs for cgroups in this hierarchy */
Li Zefan4e96ee8e2013-07-31 09:50:50 +0800318 struct idr cgroup_idr;
Tejun Heo25a7e682013-04-14 20:15:25 -0700319
320 /* The path to use for release notifications. */
321 char release_agent_path[PATH_MAX];
322
323 /* The name for this hierarchy - may be empty */
324 char name[MAX_CGROUP_ROOT_NAMELEN];
325};
326
Paul Menaged20a3902009-04-02 16:57:22 -0700327/*
328 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700329 * cgroup_subsys_state objects. This saves space in the task struct
330 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700331 * list_add()/del() can bump the reference count on the entire cgroup
332 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700333 */
334
335struct css_set {
336
337 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700338 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700339
340 /*
Li Zefan472b1052008-04-29 01:00:11 -0700341 * List running through all cgroup groups in the same hash
342 * slot. Protected by css_set_lock
343 */
344 struct hlist_node hlist;
345
346 /*
Tejun Heoc7561122014-02-25 10:04:01 -0500347 * Lists running through all tasks using this cgroup group.
348 * mg_tasks lists tasks which belong to this cset but are in the
349 * process of being migrated out or in. Protected by
350 * css_set_rwsem, but, during migration, once tasks are moved to
351 * mg_tasks, it can be read safely while holding cgroup_mutex.
Paul Menage817929e2007-10-18 23:39:36 -0700352 */
353 struct list_head tasks;
Tejun Heoc7561122014-02-25 10:04:01 -0500354 struct list_head mg_tasks;
Paul Menage817929e2007-10-18 23:39:36 -0700355
356 /*
Tejun Heo69d02062013-06-12 21:04:50 -0700357 * List of cgrp_cset_links pointing at cgroups referenced from this
358 * css_set. Protected by css_set_lock.
Paul Menage817929e2007-10-18 23:39:36 -0700359 */
Tejun Heo69d02062013-06-12 21:04:50 -0700360 struct list_head cgrp_links;
Paul Menage817929e2007-10-18 23:39:36 -0700361
Tejun Heo6803c002014-04-23 11:13:16 -0400362 /* the default cgroup associated with this css_set */
363 struct cgroup *dfl_cgrp;
364
Paul Menage817929e2007-10-18 23:39:36 -0700365 /*
Tejun Heo3ed80a62014-02-08 10:36:58 -0500366 * Set of subsystem states, one for each subsystem. This array is
367 * immutable after creation apart from the init_css_set during
368 * subsystem registration (at boot time).
Paul Menage817929e2007-10-18 23:39:36 -0700369 */
370 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700371
Tejun Heob3dc0942014-02-25 10:04:01 -0500372 /*
373 * List of csets participating in the on-going migration either as
374 * source or destination. Protected by cgroup_mutex.
375 */
Tejun Heo1958d2d2014-02-25 10:04:03 -0500376 struct list_head mg_preload_node;
Tejun Heob3dc0942014-02-25 10:04:01 -0500377 struct list_head mg_node;
378
379 /*
380 * If this cset is acting as the source of migration the following
381 * two fields are set. mg_src_cgrp is the source cgroup of the
382 * on-going migration and mg_dst_cset is the destination cset the
383 * target tasks on this cset should be migrated to. Protected by
384 * cgroup_mutex.
385 */
386 struct cgroup *mg_src_cgrp;
387 struct css_set *mg_dst_cset;
388
Tejun Heo2d8f2432014-04-23 11:13:15 -0400389 /*
390 * On the default hierarhcy, ->subsys[ssid] may point to a css
391 * attached to an ancestor instead of the cgroup this css_set is
392 * associated with. The following node is anchored at
393 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
394 * iterate through all css's attached to a given cgroup.
395 */
396 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
397
Ben Blumc3783692009-09-23 15:56:29 -0700398 /* For RCU-protected deletion */
399 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700400};
401
Paul Menage91796562008-04-29 01:00:01 -0700402/*
Paul Menaged20a3902009-04-02 16:57:22 -0700403 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700404 *
405 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800406 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700407 * - the 'cftype' of the file is file->f_dentry->d_fsdata
408 */
409
Tejun Heo8e3f6542012-04-01 12:09:55 -0700410/* cftype->flags */
Tejun Heo02c402d2013-06-24 15:21:47 -0700411enum {
Li Zefan6f4b7e62013-07-31 16:18:36 +0800412 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
413 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
Tejun Heo02c402d2013-06-24 15:21:47 -0700414 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
Tejun Heo9fa4db32013-08-26 18:40:56 -0400415 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
Tejun Heo8cbbf2c2014-03-19 10:23:55 -0400416 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
Tejun Heo02c402d2013-06-24 15:21:47 -0700417};
Tejun Heo8e3f6542012-04-01 12:09:55 -0700418
419#define MAX_CFTYPE_NAME 64
420
Paul Menageddbcc7e2007-10-18 23:39:30 -0700421struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700422 /*
423 * By convention, the name should begin with the name of the
Tejun Heo8e3f6542012-04-01 12:09:55 -0700424 * subsystem, followed by a period. Zero length string indicates
425 * end of cftype array.
Paul Menaged20a3902009-04-02 16:57:22 -0700426 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700427 char name[MAX_CFTYPE_NAME];
428 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700429 /*
430 * If not 0, file mode is set to this value, otherwise it will
431 * be figured out automatically
432 */
Al Viroa5e7ed32011-07-26 01:55:55 -0400433 umode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700434
435 /*
Tejun Heo5f469902014-02-11 11:52:48 -0500436 * The maximum length of string, excluding trailing nul, that can
437 * be passed to write_string. If < PAGE_SIZE-1, PAGE_SIZE-1 is
438 * assumed.
Paul Menagedb3b1492008-07-25 01:46:58 -0700439 */
440 size_t max_write_len;
441
Tejun Heo8e3f6542012-04-01 12:09:55 -0700442 /* CFTYPE_* flags */
443 unsigned int flags;
444
Tejun Heo2bb566c2013-08-08 20:11:23 -0400445 /*
Tejun Heo0adb0702014-02-12 09:29:48 -0500446 * Fields used for internal bookkeeping. Initialized automatically
447 * during registration.
Tejun Heo2bb566c2013-08-08 20:11:23 -0400448 */
Tejun Heo0adb0702014-02-12 09:29:48 -0500449 struct cgroup_subsys *ss; /* NULL for cgroup core files */
450 struct list_head node; /* anchored at ss->cfts */
Tejun Heo2bd59d42014-02-11 11:52:49 -0500451 struct kernfs_ops *kf_ops;
452
Paul Menageddbcc7e2007-10-18 23:39:30 -0700453 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700454 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700455 * single integer. Use it in place of read()
456 */
Tejun Heo182446d2013-08-08 20:11:24 -0400457 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700458 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700459 * read_s64() is a signed version of read_u64()
460 */
Tejun Heo182446d2013-08-08 20:11:24 -0400461 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
Tejun Heo2da8ca82013-12-05 12:28:04 -0500462
463 /* generic seq_file read interface */
464 int (*seq_show)(struct seq_file *sf, void *v);
Paul Menage91796562008-04-29 01:00:01 -0700465
Tejun Heo6612f052013-12-05 12:28:04 -0500466 /* optional ops, implement all or none */
467 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
468 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
469 void (*seq_stop)(struct seq_file *sf, void *v);
470
Paul Menage355e0c42007-10-18 23:39:33 -0700471 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700472 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700473 * a single integer (as parsed by simple_strtoull) from
474 * userspace. Use in place of write(); return 0 or error.
475 */
Tejun Heo182446d2013-08-08 20:11:24 -0400476 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
477 u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700478 /*
479 * write_s64() is a signed version of write_u64()
480 */
Tejun Heo182446d2013-08-08 20:11:24 -0400481 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
482 s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700483
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700484 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700485 * write_string() is passed a nul-terminated kernelspace
486 * buffer of maximum length determined by max_write_len.
487 * Returns 0 or -ve error code.
488 */
Tejun Heo182446d2013-08-08 20:11:24 -0400489 int (*write_string)(struct cgroup_subsys_state *css, struct cftype *cft,
Tejun Heo4d3bb512014-03-19 10:23:54 -0400490 char *buffer);
Paul Menagedb3b1492008-07-25 01:46:58 -0700491 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700492 * trigger() callback can be used to get some kick from the
493 * userspace, when the actual string written is not important
494 * at all. The private field can be used to determine the
495 * kick type for multiplexing.
496 */
Tejun Heo182446d2013-08-08 20:11:24 -0400497 int (*trigger)(struct cgroup_subsys_state *css, unsigned int event);
Tejun Heo2bd59d42014-02-11 11:52:49 -0500498
499#ifdef CONFIG_DEBUG_LOCK_ALLOC
500 struct lock_class_key lockdep_key;
501#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700502};
503
Tejun Heoa2dd4242014-03-19 10:23:55 -0400504extern struct cgroup_root cgrp_dfl_root;
505
506static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
507{
508 return cgrp->root == &cgrp_dfl_root;
509}
510
Tejun Heo8e3f6542012-04-01 12:09:55 -0700511/*
Tejun Heo873fe092013-04-14 20:15:26 -0700512 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
513 * function can be called as long as @cgrp is accessible.
514 */
515static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
516{
517 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
518}
519
Tejun Heo07bc3562014-02-13 06:58:39 -0500520/* no synchronization, the result can only be used as a hint */
521static inline bool cgroup_has_tasks(struct cgroup *cgrp)
522{
523 return !list_empty(&cgrp->cset_links);
524}
525
Tejun Heob1664922014-02-11 11:52:49 -0500526/* returns ino associated with a cgroup, 0 indicates unmounted root */
527static inline ino_t cgroup_ino(struct cgroup *cgrp)
528{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500529 if (cgrp->kn)
530 return cgrp->kn->ino;
Tejun Heob1664922014-02-11 11:52:49 -0500531 else
532 return 0;
533}
534
Tejun Heo7da11272013-12-05 12:28:04 -0500535static inline struct cftype *seq_cft(struct seq_file *seq)
536{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500537 struct kernfs_open_file *of = seq->private;
538
539 return of->kn->priv;
Tejun Heo7da11272013-12-05 12:28:04 -0500540}
541
Tejun Heo59f52962014-02-11 11:52:49 -0500542struct cgroup_subsys_state *seq_css(struct seq_file *seq);
543
Tejun Heoe61734c2014-02-12 09:29:50 -0500544/*
545 * Name / path handling functions. All are thin wrappers around the kernfs
546 * counterparts and can be called under any context.
547 */
548
549static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
550{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400551 return kernfs_name(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500552}
553
554static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
555 size_t buflen)
556{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400557 return kernfs_path(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500558}
559
560static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
561{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400562 pr_cont_kernfs_name(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500563}
564
565static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
566{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400567 pr_cont_kernfs_path(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500568}
569
570char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
571
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400572int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
Tejun Heo2bb566c2013-08-08 20:11:23 -0400573int cgroup_rm_cftypes(struct cftype *cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -0700574
Li Zefan78574cf2013-04-08 19:00:38 -0700575bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700576
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100577/*
Tejun Heo2f7ee562011-12-12 18:12:21 -0800578 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
579 * methods.
580 */
581struct cgroup_taskset;
582struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
583struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
Tejun Heo2f7ee562011-12-12 18:12:21 -0800584
585/**
586 * cgroup_taskset_for_each - iterate cgroup_taskset
587 * @task: the loop cursor
Tejun Heo2f7ee562011-12-12 18:12:21 -0800588 * @tset: taskset to iterate
589 */
Tejun Heo924f0d92014-02-13 06:58:41 -0500590#define cgroup_taskset_for_each(task, tset) \
Tejun Heo2f7ee562011-12-12 18:12:21 -0800591 for ((task) = cgroup_taskset_first((tset)); (task); \
Tejun Heo924f0d92014-02-13 06:58:41 -0500592 (task) = cgroup_taskset_next((tset)))
Tejun Heo2f7ee562011-12-12 18:12:21 -0800593
594/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100595 * Control Group subsystem type.
596 * See Documentation/cgroups/cgroups.txt for details
597 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700598
599struct cgroup_subsys {
Tejun Heoeb954192013-08-08 20:11:23 -0400600 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
601 int (*css_online)(struct cgroup_subsys_state *css);
602 void (*css_offline)(struct cgroup_subsys_state *css);
603 void (*css_free)(struct cgroup_subsys_state *css);
Tejun Heo92fb9742012-11-19 08:13:38 -0800604
Tejun Heoeb954192013-08-08 20:11:23 -0400605 int (*can_attach)(struct cgroup_subsys_state *css,
606 struct cgroup_taskset *tset);
607 void (*cancel_attach)(struct cgroup_subsys_state *css,
608 struct cgroup_taskset *tset);
609 void (*attach)(struct cgroup_subsys_state *css,
610 struct cgroup_taskset *tset);
Li Zefan761b3ef2012-01-31 13:47:36 +0800611 void (*fork)(struct task_struct *task);
Tejun Heoeb954192013-08-08 20:11:23 -0400612 void (*exit)(struct cgroup_subsys_state *css,
613 struct cgroup_subsys_state *old_css,
Li Zefan761b3ef2012-01-31 13:47:36 +0800614 struct task_struct *task);
Tejun Heoeb954192013-08-08 20:11:23 -0400615 void (*bind)(struct cgroup_subsys_state *root_css);
Hugh Dickinse5991372009-01-06 14:39:22 -0800616
Paul Menage8bab8dd2008-04-04 14:29:57 -0700617 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700618 int early_init;
Tejun Heo48ddbe12012-04-01 12:09:56 -0700619
620 /*
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700621 * If %false, this subsystem is properly hierarchical -
622 * configuration, resource accounting and restriction on a parent
623 * cgroup cover those of its children. If %true, hierarchy support
624 * is broken in some ways - some subsystems ignore hierarchy
625 * completely while others are only implemented half-way.
626 *
627 * It's now disallowed to create nested cgroups if the subsystem is
628 * broken and cgroup core will emit a warning message on such
629 * cases. Eventually, all subsystems will be made properly
630 * hierarchical and this will go away.
631 */
632 bool broken_hierarchy;
633 bool warned_broken_hierarchy;
634
Tejun Heo073219e2014-02-08 10:36:58 -0500635 /* the following two fields are initialized automtically during boot */
Tejun Heoaec25022014-02-08 10:36:58 -0500636 int id;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700637#define MAX_CGROUP_TYPE_NAMELEN 32
638 const char *name;
639
Tejun Heob85d2042013-12-06 15:11:57 -0500640 /* link to parent, protected by cgroup_lock() */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400641 struct cgroup_root *root;
Ben Blume6a11052010-03-10 15:22:09 -0800642
Tejun Heo0adb0702014-02-12 09:29:48 -0500643 /*
644 * List of cftypes. Each entry is the first entry of an array
645 * terminated by zero length name.
646 */
647 struct list_head cfts;
Tejun Heo8e3f6542012-04-01 12:09:55 -0700648
Tejun Heode00ffa2014-02-11 11:52:48 -0500649 /* base cftypes, automatically registered with subsys itself */
Tejun Heo8e3f6542012-04-01 12:09:55 -0700650 struct cftype *base_cftypes;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700651};
652
Tejun Heo073219e2014-02-08 10:36:58 -0500653#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700654#include <linux/cgroup_subsys.h>
655#undef SUBSYS
656
Tejun Heo8af01f52013-08-08 20:11:22 -0400657/**
Tejun Heo63876982013-08-08 20:11:23 -0400658 * css_parent - find the parent css
659 * @css: the target cgroup_subsys_state
660 *
661 * Return the parent css of @css. This function is guaranteed to return
662 * non-NULL parent as long as @css isn't the root.
663 */
664static inline
665struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css)
666{
Tejun Heo0ae78e02013-08-13 11:01:54 -0400667 return css->parent;
Tejun Heo63876982013-08-08 20:11:23 -0400668}
669
670/**
Tejun Heo14611e52013-06-25 11:48:32 -0700671 * task_css_set_check - obtain a task's css_set with extra access conditions
672 * @task: the task to obtain css_set for
673 * @__c: extra condition expression to be passed to rcu_dereference_check()
674 *
675 * A task's css_set is RCU protected, initialized and exited while holding
676 * task_lock(), and can only be modified while holding both cgroup_mutex
677 * and task_lock() while the task is alive. This macro verifies that the
678 * caller is inside proper critical section and returns @task's css_set.
679 *
680 * The caller can also specify additional allowed conditions via @__c, such
681 * as locks used during the cgroup_subsys::attach() methods.
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200682 */
Tejun Heo22194492013-04-07 09:29:51 -0700683#ifdef CONFIG_PROVE_RCU
684extern struct mutex cgroup_mutex;
Tejun Heo0e1d7682014-02-25 10:04:03 -0500685extern struct rw_semaphore css_set_rwsem;
Tejun Heo14611e52013-06-25 11:48:32 -0700686#define task_css_set_check(task, __c) \
687 rcu_dereference_check((task)->cgroups, \
Tejun Heo0e1d7682014-02-25 10:04:03 -0500688 lockdep_is_held(&cgroup_mutex) || \
689 lockdep_is_held(&css_set_rwsem) || \
690 ((task)->flags & PF_EXITING) || (__c))
Tejun Heo22194492013-04-07 09:29:51 -0700691#else
Tejun Heo14611e52013-06-25 11:48:32 -0700692#define task_css_set_check(task, __c) \
693 rcu_dereference((task)->cgroups)
Tejun Heo22194492013-04-07 09:29:51 -0700694#endif
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200695
Tejun Heo14611e52013-06-25 11:48:32 -0700696/**
Tejun Heo8af01f52013-08-08 20:11:22 -0400697 * task_css_check - obtain css for (task, subsys) w/ extra access conds
Tejun Heo14611e52013-06-25 11:48:32 -0700698 * @task: the target task
699 * @subsys_id: the target subsystem ID
700 * @__c: extra condition expression to be passed to rcu_dereference_check()
701 *
702 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
703 * synchronization rules are the same as task_css_set_check().
704 */
Tejun Heo8af01f52013-08-08 20:11:22 -0400705#define task_css_check(task, subsys_id, __c) \
Tejun Heo14611e52013-06-25 11:48:32 -0700706 task_css_set_check((task), (__c))->subsys[(subsys_id)]
707
708/**
709 * task_css_set - obtain a task's css_set
710 * @task: the task to obtain css_set for
711 *
712 * See task_css_set_check().
713 */
714static inline struct css_set *task_css_set(struct task_struct *task)
715{
716 return task_css_set_check(task, false);
717}
718
719/**
Tejun Heo8af01f52013-08-08 20:11:22 -0400720 * task_css - obtain css for (task, subsys)
Tejun Heo14611e52013-06-25 11:48:32 -0700721 * @task: the target task
722 * @subsys_id: the target subsystem ID
723 *
Tejun Heo8af01f52013-08-08 20:11:22 -0400724 * See task_css_check().
Tejun Heo14611e52013-06-25 11:48:32 -0700725 */
Tejun Heo8af01f52013-08-08 20:11:22 -0400726static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
727 int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700728{
Tejun Heo8af01f52013-08-08 20:11:22 -0400729 return task_css_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700730}
731
Tejun Heo8af01f52013-08-08 20:11:22 -0400732static inline struct cgroup *task_cgroup(struct task_struct *task,
733 int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700734{
Tejun Heo8af01f52013-08-08 20:11:22 -0400735 return task_css(task, subsys_id)->cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700736}
737
Tejun Heo492eb212013-08-08 20:11:25 -0400738struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
739 struct cgroup_subsys_state *parent);
Tejun Heo53fa5262013-05-24 10:55:38 +0900740
Li Zefan1cb650b2013-08-19 10:05:24 +0800741struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
742
Tejun Heo574bd9f2012-11-09 09:12:29 -0800743/**
Tejun Heo492eb212013-08-08 20:11:25 -0400744 * css_for_each_child - iterate through children of a css
745 * @pos: the css * to use as the loop cursor
746 * @parent: css whose children to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800747 *
Tejun Heo492eb212013-08-08 20:11:25 -0400748 * Walk @parent's children. Must be called under rcu_read_lock(). A child
749 * css which hasn't finished ->css_online() or already has finished
Tejun Heo92fb9742012-11-19 08:13:38 -0800750 * ->css_offline() may show up during traversal and it's each subsystem's
Tejun Heo574bd9f2012-11-09 09:12:29 -0800751 * responsibility to verify that each @pos is alive.
752 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800753 * If a subsystem synchronizes against the parent in its ->css_online() and
Tejun Heo492eb212013-08-08 20:11:25 -0400754 * before starting iterating, a css which finished ->css_online() is
Tejun Heo92fb9742012-11-19 08:13:38 -0800755 * guaranteed to be visible in the future iterations.
Tejun Heo75501a62013-05-24 10:55:38 +0900756 *
757 * It is allowed to temporarily drop RCU read lock during iteration. The
758 * caller is responsible for ensuring that @pos remains accessible until
759 * the start of the next iteration by, for example, bumping the css refcnt.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800760 */
Tejun Heo492eb212013-08-08 20:11:25 -0400761#define css_for_each_child(pos, parent) \
762 for ((pos) = css_next_child(NULL, (parent)); (pos); \
763 (pos) = css_next_child((pos), (parent)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800764
Tejun Heo492eb212013-08-08 20:11:25 -0400765struct cgroup_subsys_state *
766css_next_descendant_pre(struct cgroup_subsys_state *pos,
767 struct cgroup_subsys_state *css);
768
769struct cgroup_subsys_state *
770css_rightmost_descendant(struct cgroup_subsys_state *pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -0800771
772/**
Tejun Heo492eb212013-08-08 20:11:25 -0400773 * css_for_each_descendant_pre - pre-order walk of a css's descendants
774 * @pos: the css * to use as the loop cursor
775 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800776 *
Tejun Heobd8815a2013-08-08 20:11:27 -0400777 * Walk @root's descendants. @root is included in the iteration and the
778 * first node to be visited. Must be called under rcu_read_lock(). A
Tejun Heo492eb212013-08-08 20:11:25 -0400779 * descendant css which hasn't finished ->css_online() or already has
Tejun Heo92fb9742012-11-19 08:13:38 -0800780 * finished ->css_offline() may show up during traversal and it's each
Tejun Heo574bd9f2012-11-09 09:12:29 -0800781 * subsystem's responsibility to verify that each @pos is alive.
782 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800783 * If a subsystem synchronizes against the parent in its ->css_online() and
784 * before starting iterating, and synchronizes against @pos on each
Tejun Heo492eb212013-08-08 20:11:25 -0400785 * iteration, any descendant css which finished ->css_online() is
Tejun Heo574bd9f2012-11-09 09:12:29 -0800786 * guaranteed to be visible in the future iterations.
787 *
788 * In other words, the following guarantees that a descendant can't escape
789 * state updates of its ancestors.
790 *
Tejun Heo492eb212013-08-08 20:11:25 -0400791 * my_online(@css)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800792 * {
Tejun Heo492eb212013-08-08 20:11:25 -0400793 * Lock @css's parent and @css;
794 * Inherit state from the parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -0800795 * Unlock both.
796 * }
797 *
Tejun Heo492eb212013-08-08 20:11:25 -0400798 * my_update_state(@css)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800799 * {
Tejun Heo492eb212013-08-08 20:11:25 -0400800 * css_for_each_descendant_pre(@pos, @css) {
Tejun Heo574bd9f2012-11-09 09:12:29 -0800801 * Lock @pos;
Tejun Heobd8815a2013-08-08 20:11:27 -0400802 * if (@pos == @css)
803 * Update @css's state;
804 * else
805 * Verify @pos is alive and inherit state from its parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -0800806 * Unlock @pos;
807 * }
808 * }
809 *
810 * As long as the inheriting step, including checking the parent state, is
811 * enclosed inside @pos locking, double-locking the parent isn't necessary
812 * while inheriting. The state update to the parent is guaranteed to be
813 * visible by walking order and, as long as inheriting operations to the
814 * same @pos are atomic to each other, multiple updates racing each other
815 * still result in the correct state. It's guaranateed that at least one
Tejun Heo492eb212013-08-08 20:11:25 -0400816 * inheritance happens for any css after the latest update to its parent.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800817 *
818 * If checking parent's state requires locking the parent, each inheriting
819 * iteration should lock and unlock both @pos->parent and @pos.
820 *
821 * Alternatively, a subsystem may choose to use a single global lock to
Tejun Heo92fb9742012-11-19 08:13:38 -0800822 * synchronize ->css_online() and ->css_offline() against tree-walking
Tejun Heo574bd9f2012-11-09 09:12:29 -0800823 * operations.
Tejun Heo75501a62013-05-24 10:55:38 +0900824 *
825 * It is allowed to temporarily drop RCU read lock during iteration. The
826 * caller is responsible for ensuring that @pos remains accessible until
827 * the start of the next iteration by, for example, bumping the css refcnt.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800828 */
Tejun Heo492eb212013-08-08 20:11:25 -0400829#define css_for_each_descendant_pre(pos, css) \
830 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
831 (pos) = css_next_descendant_pre((pos), (css)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800832
Tejun Heo492eb212013-08-08 20:11:25 -0400833struct cgroup_subsys_state *
834css_next_descendant_post(struct cgroup_subsys_state *pos,
835 struct cgroup_subsys_state *css);
Tejun Heo574bd9f2012-11-09 09:12:29 -0800836
837/**
Tejun Heo492eb212013-08-08 20:11:25 -0400838 * css_for_each_descendant_post - post-order walk of a css's descendants
839 * @pos: the css * to use as the loop cursor
840 * @css: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800841 *
Tejun Heo492eb212013-08-08 20:11:25 -0400842 * Similar to css_for_each_descendant_pre() but performs post-order
Tejun Heobd8815a2013-08-08 20:11:27 -0400843 * traversal instead. @root is included in the iteration and the last
844 * node to be visited. Note that the walk visibility guarantee described
845 * in pre-order walk doesn't apply the same to post-order walks.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800846 */
Tejun Heo492eb212013-08-08 20:11:25 -0400847#define css_for_each_descendant_post(pos, css) \
848 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
849 (pos) = css_next_descendant_post((pos), (css)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800850
Tejun Heo72ec7022013-08-08 20:11:26 -0400851/* A css_task_iter should be treated as an opaque object */
852struct css_task_iter {
Tejun Heo3ebb2b62014-04-23 11:13:15 -0400853 struct cgroup_subsys *ss;
854
Tejun Heo0f0a2b42014-04-23 11:13:15 -0400855 struct list_head *cset_pos;
856 struct list_head *cset_head;
857
858 struct list_head *task_pos;
859 struct list_head *tasks_head;
860 struct list_head *mg_tasks_head;
Paul Menage817929e2007-10-18 23:39:36 -0700861};
862
Tejun Heo72ec7022013-08-08 20:11:26 -0400863void css_task_iter_start(struct cgroup_subsys_state *css,
864 struct css_task_iter *it);
865struct task_struct *css_task_iter_next(struct css_task_iter *it);
866void css_task_iter_end(struct css_task_iter *it);
Tejun Heoe5358372013-08-08 20:11:26 -0400867
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700868int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
Tejun Heo8cc99342013-04-07 09:29:50 -0700869int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700870
Tejun Heo5a17f542014-02-11 11:52:47 -0500871struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry,
872 struct cgroup_subsys *ss);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700873
Paul Menageddbcc7e2007-10-18 23:39:30 -0700874#else /* !CONFIG_CGROUPS */
875
876static inline int cgroup_init_early(void) { return 0; }
877static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700878static inline void cgroup_fork(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700879static inline void cgroup_post_fork(struct task_struct *p) {}
Li Zefan1ec41832014-03-28 15:22:19 +0800880static inline void cgroup_exit(struct task_struct *p) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700881
Balbir Singh846c7bb2007-10-18 23:39:44 -0700882static inline int cgroupstats_build(struct cgroupstats *stats,
883 struct dentry *dentry)
884{
885 return -EINVAL;
886}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700887
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200888/* No cgroups - nothing to do */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700889static inline int cgroup_attach_task_all(struct task_struct *from,
890 struct task_struct *t)
891{
892 return 0;
893}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200894
Paul Menageddbcc7e2007-10-18 23:39:30 -0700895#endif /* !CONFIG_CGROUPS */
896
Paul Menageddbcc7e2007-10-18 23:39:30 -0700897#endif /* _LINUX_CGROUP_H */