blob: b76999954beb5ffbb835cd66dc917a051fb194e2 [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
Tejun Heo5c9d5352014-05-16 13:22:48 -040051/*
52 * Per-subsystem/per-cgroup state maintained by the system. This is the
53 * fundamental structural building block that controllers deal with.
54 *
55 * Fields marked with "PI:" are public and immutable and may be accessed
56 * directly without synchronization.
57 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070058struct cgroup_subsys_state {
Tejun Heo5c9d5352014-05-16 13:22:48 -040059 /* PI: the cgroup that this css is attached to */
Paul Menageddbcc7e2007-10-18 23:39:30 -070060 struct cgroup *cgroup;
61
Tejun Heo5c9d5352014-05-16 13:22:48 -040062 /* PI: the cgroup subsystem that this css is attached to */
Tejun Heo72c97e52013-08-08 20:11:22 -040063 struct cgroup_subsys *ss;
64
Tejun Heod3daf282013-06-13 19:39:16 -070065 /* reference count - access via css_[try]get() and css_put() */
66 struct percpu_ref refcnt;
Paul Menageddbcc7e2007-10-18 23:39:30 -070067
Tejun Heo5c9d5352014-05-16 13:22:48 -040068 /* PI: the parent css */
Tejun Heo0ae78e02013-08-13 11:01:54 -040069 struct cgroup_subsys_state *parent;
70
Tejun Heod5c419b2014-05-16 13:22:48 -040071 /* siblings list anchored at the parent's ->children */
72 struct list_head sibling;
73 struct list_head children;
74
Tejun Heo15a4c832014-05-04 15:09:14 -040075 /*
Tejun Heo5c9d5352014-05-16 13:22:48 -040076 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
Tejun Heo15a4c832014-05-04 15:09:14 -040077 * matching css can be looked up using css_from_id().
78 */
79 int id;
80
Tejun Heo69dfa002014-05-04 15:09:13 -040081 unsigned int flags;
Tejun Heo48ddbe12012-04-01 12:09:56 -070082
Tejun Heo0cb51d72014-05-16 13:22:49 -040083 /*
84 * Monotonically increasing unique serial number which defines a
85 * uniform order among all csses. It's guaranteed that all
86 * ->children lists are in the ascending order of ->serial_nr and
87 * used to allow interrupting and resuming iterations.
88 */
89 u64 serial_nr;
90
Tejun Heo0c21ead2013-08-13 20:22:51 -040091 /* percpu_ref killing and RCU release */
92 struct rcu_head rcu_head;
Tejun Heo35ef10d2013-08-13 11:01:54 -040093 struct work_struct destroy_work;
Paul Menageddbcc7e2007-10-18 23:39:30 -070094};
95
96/* bits in struct cgroup_subsys_state flags field */
97enum {
Tejun Heo3b514d22014-05-16 13:22:47 -040098 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
Tejun Heo92fb9742012-11-19 08:13:38 -080099 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
Tejun Heode3f0342014-05-16 13:22:49 -0400100 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700101};
102
Tejun Heo5de01072013-06-12 21:04:52 -0700103/**
104 * css_get - obtain a reference on the specified css
105 * @css: target css
106 *
107 * The caller must already have a reference.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700108 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700109static inline void css_get(struct cgroup_subsys_state *css)
110{
Tejun Heo3b514d22014-05-16 13:22:47 -0400111 if (!(css->flags & CSS_NO_REF))
112 percpu_ref_get(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700113}
Paul Menagee7c5ec92009-01-07 18:08:38 -0800114
Tejun Heo5de01072013-06-12 21:04:52 -0700115/**
Tejun Heoec903c02014-05-13 12:11:01 -0400116 * css_tryget_online - try to obtain a reference on the specified css if online
Tejun Heo5de01072013-06-12 21:04:52 -0700117 * @css: target css
118 *
Tejun Heoec903c02014-05-13 12:11:01 -0400119 * Obtain a reference on @css if it's online. The caller naturally needs
120 * to ensure that @css is accessible but doesn't have to be holding a
Tejun Heo5de01072013-06-12 21:04:52 -0700121 * reference on it - IOW, RCU protected access is good enough for this
122 * function. Returns %true if a reference count was successfully obtained;
123 * %false otherwise.
124 */
Tejun Heoec903c02014-05-13 12:11:01 -0400125static inline bool css_tryget_online(struct cgroup_subsys_state *css)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800126{
Tejun Heo3b514d22014-05-16 13:22:47 -0400127 if (!(css->flags & CSS_NO_REF))
128 return percpu_ref_tryget_live(&css->refcnt);
129 return true;
Paul Menagee7c5ec92009-01-07 18:08:38 -0800130}
131
Tejun Heo5de01072013-06-12 21:04:52 -0700132/**
133 * css_put - put a css reference
134 * @css: target css
135 *
Tejun Heoec903c02014-05-13 12:11:01 -0400136 * Put a reference obtained via css_get() and css_tryget_online().
Tejun Heo5de01072013-06-12 21:04:52 -0700137 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700138static inline void css_put(struct cgroup_subsys_state *css)
139{
Tejun Heo3b514d22014-05-16 13:22:47 -0400140 if (!(css->flags & CSS_NO_REF))
141 percpu_ref_put(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700142}
143
Paul Menage3116f0e2008-04-29 01:00:04 -0700144/* bits in struct cgroup flags field */
145enum {
Paul Menaged20a3902009-04-02 16:57:22 -0700146 /*
147 * Control Group has previously had a child cgroup or a task,
148 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
149 */
Paul Menage3116f0e2008-04-29 01:00:04 -0700150 CGRP_RELEASABLE,
151 /* Control Group requires release notifications to userspace */
152 CGRP_NOTIFY_ON_RELEASE,
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700153 /*
Tejun Heo2260e7f2012-11-19 08:13:38 -0800154 * Clone the parent's configuration when creating a new child
155 * cpuset cgroup. For historical reasons, this option can be
156 * specified at mount time and thus is implemented here.
Daniel Lezcano97978e62010-10-27 15:33:35 -0700157 */
Tejun Heo2260e7f2012-11-19 08:13:38 -0800158 CGRP_CPUSET_CLONE_CHILDREN,
Paul Menage3116f0e2008-04-29 01:00:04 -0700159};
160
Paul Menageddbcc7e2007-10-18 23:39:30 -0700161struct cgroup {
Tejun Heo9d800df2014-05-14 09:15:00 -0400162 /* self css with NULL ->ss, points back to this cgroup */
163 struct cgroup_subsys_state self;
164
Paul Menageddbcc7e2007-10-18 23:39:30 -0700165 unsigned long flags; /* "unsigned long" so bitops work */
166
Li Zefanb414dc02013-07-31 09:51:06 +0800167 /*
168 * idr allocated in-hierarchy ID.
169 *
Tejun Heo7d699dd2014-05-04 15:09:13 -0400170 * ID 0 is not used, the ID of the root cgroup is always 1, and a
171 * new cgroup will be assigned with a smallest available ID.
Li Zefan0ab02ca2014-02-11 16:05:46 +0800172 *
173 * Allocating/Removing ID must be protected by cgroup_mutex.
Li Zefanb414dc02013-07-31 09:51:06 +0800174 */
175 int id;
Tejun Heo0a950f62012-11-19 09:02:12 -0800176
Tejun Heo842b5972014-04-25 18:28:02 -0400177 /*
178 * If this cgroup contains any tasks, it contributes one to
179 * populated_cnt. All children with non-zero popuplated_cnt of
180 * their own contribute one. The count is zero iff there's no task
181 * in this cgroup or its subtree.
182 */
183 int populated_cnt;
184
Tejun Heo2bd59d42014-02-11 11:52:49 -0500185 struct kernfs_node *kn; /* cgroup kernfs entry */
Tejun Heo842b5972014-04-25 18:28:02 -0400186 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700187
Tejun Heof392e512014-04-23 11:13:14 -0400188 /* the bitmask of subsystems enabled on the child cgroups */
Tejun Heo69dfa002014-05-04 15:09:13 -0400189 unsigned int child_subsys_mask;
Tejun Heo94419622014-03-19 10:23:54 -0400190
Paul Menageddbcc7e2007-10-18 23:39:30 -0700191 /* Private pointers for each registered subsystem */
Tejun Heo73e80ed2013-08-13 11:01:55 -0400192 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700193
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400194 struct cgroup_root *root;
Paul Menage817929e2007-10-18 23:39:36 -0700195
196 /*
Tejun Heo69d02062013-06-12 21:04:50 -0700197 * List of cgrp_cset_links pointing at css_sets with tasks in this
198 * cgroup. Protected by css_set_lock.
Paul Menage817929e2007-10-18 23:39:36 -0700199 */
Tejun Heo69d02062013-06-12 21:04:50 -0700200 struct list_head cset_links;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700201
202 /*
Tejun Heo2d8f2432014-04-23 11:13:15 -0400203 * On the default hierarchy, a css_set for a cgroup with some
204 * susbsys disabled will point to css's which are associated with
205 * the closest ancestor which has the subsys enabled. The
206 * following lists all css_sets which point to this cgroup's css
207 * for the given subsystem.
208 */
209 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
210
211 /*
Paul Menage81a6a5c2007-10-18 23:39:38 -0700212 * Linked list running through all cgroups that can
213 * potentially be reaped by the release agent. Protected by
214 * release_list_lock
215 */
216 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700217
Ben Blum72a8cb32009-09-23 15:56:27 -0700218 /*
219 * list of pidlists, up to two for each namespace (one for procs, one
220 * for tasks); created on demand.
221 */
222 struct list_head pidlists;
223 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800224
Tejun Heof8f22e52014-04-23 11:13:16 -0400225 /* used to wait for offlining of csses */
226 wait_queue_head_t offline_waitq;
Paul Menage817929e2007-10-18 23:39:36 -0700227};
228
Tejun Heo25a7e682013-04-14 20:15:25 -0700229#define MAX_CGROUP_ROOT_NAMELEN 64
230
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400231/* cgroup_root->flags */
Tejun Heo25a7e682013-04-14 20:15:25 -0700232enum {
Tejun Heo873fe092013-04-14 20:15:26 -0700233 /*
234 * Unfortunately, cgroup core and various controllers are riddled
235 * with idiosyncrasies and pointless options. The following flag,
236 * when set, will force sane behavior - some options are forced on,
237 * others are disallowed, and some controllers will change their
238 * hierarchical or other behaviors.
239 *
240 * The set of behaviors affected by this flag are still being
241 * determined and developed and the mount option for this flag is
242 * prefixed with __DEVEL__. The prefix will be dropped once we
243 * reach the point where all behaviors are compatible with the
244 * planned unified hierarchy, which will automatically turn on this
245 * flag.
246 *
247 * The followings are the behaviors currently affected this flag.
248 *
Tejun Heod3ba07c2014-02-13 06:58:38 -0500249 * - Mount options "noprefix", "xattr", "clone_children",
250 * "release_agent" and "name" are disallowed.
Tejun Heo873fe092013-04-14 20:15:26 -0700251 *
252 * - When mounting an existing superblock, mount options should
253 * match.
254 *
255 * - Remount is disallowed.
256 *
Linus Torvalds0b0585c2013-07-02 20:04:25 -0700257 * - rename(2) is disallowed.
258 *
Tejun Heof63674f2013-06-13 19:58:38 -0700259 * - "tasks" is removed. Everything should be at process
260 * granularity. Use "cgroup.procs" instead.
261 *
Tejun Heoafb2bc12013-11-29 10:42:59 -0500262 * - "cgroup.procs" is not sorted. pids will be unique unless they
263 * got recycled inbetween reads.
264 *
Tejun Heof63674f2013-06-13 19:58:38 -0700265 * - "release_agent" and "notify_on_release" are removed.
266 * Replacement notification mechanism will be implemented.
267 *
Tejun Heod3ba07c2014-02-13 06:58:38 -0500268 * - "cgroup.clone_children" is removed.
Tejun Heo86bf4b62014-02-12 09:29:48 -0500269 *
Tejun Heo842b5972014-04-25 18:28:02 -0400270 * - "cgroup.subtree_populated" is available. Its value is 0 if
271 * the cgroup and its descendants contain no task; otherwise, 1.
272 * The file also generates kernfs notification which can be
273 * monitored through poll and [di]notify when the value of the
274 * file changes.
275 *
Tejun Heoa2dd4242014-03-19 10:23:55 -0400276 * - If mount is requested with sane_behavior but without any
277 * subsystem, the default unified hierarchy is mounted.
278 *
Li Zefan5c5cc622013-06-09 17:16:29 +0800279 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
280 * and take masks of ancestors with non-empty cpus/mems, instead of
281 * being moved to an ancestor.
282 *
Li Zefan88fa5232013-06-09 17:16:46 +0800283 * - cpuset: a task can be moved into an empty cpuset, and again it
284 * takes masks of ancestors.
Tejun Heo6db8e852013-06-14 11:18:22 -0700285 *
Tejun Heof00baae2013-04-15 13:41:15 -0700286 * - memcg: use_hierarchy is on by default and the cgroup file for
287 * the flag is not created.
288 *
Tejun Heo91381252013-05-14 13:52:38 -0700289 * - blkcg: blk-throttle becomes properly hierarchical.
Tejun Heo873fe092013-04-14 20:15:26 -0700290 */
291 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
292
Tejun Heo25a7e682013-04-14 20:15:25 -0700293 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
294 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
Tejun Heo0ce6cba2013-06-27 19:37:26 -0700295
296 /* mount options live below bit 16 */
297 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
Tejun Heo25a7e682013-04-14 20:15:25 -0700298};
299
300/*
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400301 * A cgroup_root represents the root of a cgroup hierarchy, and may be
Tejun Heo2bd59d42014-02-11 11:52:49 -0500302 * associated with a kernfs_root to form an active hierarchy. This is
Tejun Heo25a7e682013-04-14 20:15:25 -0700303 * internal to cgroup core. Don't access directly from controllers.
304 */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400305struct cgroup_root {
Tejun Heo2bd59d42014-02-11 11:52:49 -0500306 struct kernfs_root *kf_root;
Tejun Heo25a7e682013-04-14 20:15:25 -0700307
Tejun Heof392e512014-04-23 11:13:14 -0400308 /* The bitmask of subsystems attached to this hierarchy */
Tejun Heo69dfa002014-05-04 15:09:13 -0400309 unsigned int subsys_mask;
Tejun Heof392e512014-04-23 11:13:14 -0400310
Tejun Heo25a7e682013-04-14 20:15:25 -0700311 /* Unique id for this hierarchy. */
312 int hierarchy_id;
313
Tejun Heo776f02f2014-02-12 09:29:50 -0500314 /* The root cgroup. Root is destroyed on its release. */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400315 struct cgroup cgrp;
Tejun Heo25a7e682013-04-14 20:15:25 -0700316
Tejun Heo3c9c8252014-02-12 09:29:50 -0500317 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
318 atomic_t nr_cgrps;
Tejun Heo25a7e682013-04-14 20:15:25 -0700319
320 /* A list running through the active hierarchies */
321 struct list_head root_list;
322
Tejun Heo25a7e682013-04-14 20:15:25 -0700323 /* Hierarchy-specific flags */
Tejun Heo69dfa002014-05-04 15:09:13 -0400324 unsigned int flags;
Tejun Heo25a7e682013-04-14 20:15:25 -0700325
326 /* IDs for cgroups in this hierarchy */
Li Zefan4e96ee8e2013-07-31 09:50:50 +0800327 struct idr cgroup_idr;
Tejun Heo25a7e682013-04-14 20:15:25 -0700328
329 /* The path to use for release notifications. */
330 char release_agent_path[PATH_MAX];
331
332 /* The name for this hierarchy - may be empty */
333 char name[MAX_CGROUP_ROOT_NAMELEN];
334};
335
Paul Menaged20a3902009-04-02 16:57:22 -0700336/*
337 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700338 * cgroup_subsys_state objects. This saves space in the task struct
339 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700340 * list_add()/del() can bump the reference count on the entire cgroup
341 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700342 */
343
344struct css_set {
345
346 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700347 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700348
349 /*
Li Zefan472b1052008-04-29 01:00:11 -0700350 * List running through all cgroup groups in the same hash
351 * slot. Protected by css_set_lock
352 */
353 struct hlist_node hlist;
354
355 /*
Tejun Heoc7561122014-02-25 10:04:01 -0500356 * Lists running through all tasks using this cgroup group.
357 * mg_tasks lists tasks which belong to this cset but are in the
358 * process of being migrated out or in. Protected by
359 * css_set_rwsem, but, during migration, once tasks are moved to
360 * mg_tasks, it can be read safely while holding cgroup_mutex.
Paul Menage817929e2007-10-18 23:39:36 -0700361 */
362 struct list_head tasks;
Tejun Heoc7561122014-02-25 10:04:01 -0500363 struct list_head mg_tasks;
Paul Menage817929e2007-10-18 23:39:36 -0700364
365 /*
Tejun Heo69d02062013-06-12 21:04:50 -0700366 * List of cgrp_cset_links pointing at cgroups referenced from this
367 * css_set. Protected by css_set_lock.
Paul Menage817929e2007-10-18 23:39:36 -0700368 */
Tejun Heo69d02062013-06-12 21:04:50 -0700369 struct list_head cgrp_links;
Paul Menage817929e2007-10-18 23:39:36 -0700370
Tejun Heo6803c002014-04-23 11:13:16 -0400371 /* the default cgroup associated with this css_set */
372 struct cgroup *dfl_cgrp;
373
Paul Menage817929e2007-10-18 23:39:36 -0700374 /*
Tejun Heo3ed80a62014-02-08 10:36:58 -0500375 * Set of subsystem states, one for each subsystem. This array is
376 * immutable after creation apart from the init_css_set during
377 * subsystem registration (at boot time).
Paul Menage817929e2007-10-18 23:39:36 -0700378 */
379 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700380
Tejun Heob3dc0942014-02-25 10:04:01 -0500381 /*
382 * List of csets participating in the on-going migration either as
383 * source or destination. Protected by cgroup_mutex.
384 */
Tejun Heo1958d2d2014-02-25 10:04:03 -0500385 struct list_head mg_preload_node;
Tejun Heob3dc0942014-02-25 10:04:01 -0500386 struct list_head mg_node;
387
388 /*
389 * If this cset is acting as the source of migration the following
390 * two fields are set. mg_src_cgrp is the source cgroup of the
391 * on-going migration and mg_dst_cset is the destination cset the
392 * target tasks on this cset should be migrated to. Protected by
393 * cgroup_mutex.
394 */
395 struct cgroup *mg_src_cgrp;
396 struct css_set *mg_dst_cset;
397
Tejun Heo2d8f2432014-04-23 11:13:15 -0400398 /*
399 * On the default hierarhcy, ->subsys[ssid] may point to a css
400 * attached to an ancestor instead of the cgroup this css_set is
401 * associated with. The following node is anchored at
402 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
403 * iterate through all css's attached to a given cgroup.
404 */
405 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
406
Ben Blumc3783692009-09-23 15:56:29 -0700407 /* For RCU-protected deletion */
408 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700409};
410
Paul Menage91796562008-04-29 01:00:01 -0700411/*
Paul Menaged20a3902009-04-02 16:57:22 -0700412 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700413 *
414 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800415 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700416 * - the 'cftype' of the file is file->f_dentry->d_fsdata
417 */
418
Tejun Heo8e3f6542012-04-01 12:09:55 -0700419/* cftype->flags */
Tejun Heo02c402d2013-06-24 15:21:47 -0700420enum {
Li Zefan6f4b7e62013-07-31 16:18:36 +0800421 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
422 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
Tejun Heo02c402d2013-06-24 15:21:47 -0700423 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
Tejun Heo9fa4db32013-08-26 18:40:56 -0400424 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
Tejun Heo8cbbf2c2014-03-19 10:23:55 -0400425 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
Tejun Heo02c402d2013-06-24 15:21:47 -0700426};
Tejun Heo8e3f6542012-04-01 12:09:55 -0700427
428#define MAX_CFTYPE_NAME 64
429
Paul Menageddbcc7e2007-10-18 23:39:30 -0700430struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700431 /*
432 * By convention, the name should begin with the name of the
Tejun Heo8e3f6542012-04-01 12:09:55 -0700433 * subsystem, followed by a period. Zero length string indicates
434 * end of cftype array.
Paul Menaged20a3902009-04-02 16:57:22 -0700435 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700436 char name[MAX_CFTYPE_NAME];
437 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700438 /*
439 * If not 0, file mode is set to this value, otherwise it will
440 * be figured out automatically
441 */
Al Viroa5e7ed32011-07-26 01:55:55 -0400442 umode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700443
444 /*
Tejun Heo5f469902014-02-11 11:52:48 -0500445 * The maximum length of string, excluding trailing nul, that can
Tejun Heo451af502014-05-13 12:16:21 -0400446 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
Paul Menagedb3b1492008-07-25 01:46:58 -0700447 */
448 size_t max_write_len;
449
Tejun Heo8e3f6542012-04-01 12:09:55 -0700450 /* CFTYPE_* flags */
451 unsigned int flags;
452
Tejun Heo2bb566c2013-08-08 20:11:23 -0400453 /*
Tejun Heo0adb0702014-02-12 09:29:48 -0500454 * Fields used for internal bookkeeping. Initialized automatically
455 * during registration.
Tejun Heo2bb566c2013-08-08 20:11:23 -0400456 */
Tejun Heo0adb0702014-02-12 09:29:48 -0500457 struct cgroup_subsys *ss; /* NULL for cgroup core files */
458 struct list_head node; /* anchored at ss->cfts */
Tejun Heo2bd59d42014-02-11 11:52:49 -0500459 struct kernfs_ops *kf_ops;
460
Paul Menageddbcc7e2007-10-18 23:39:30 -0700461 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700462 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700463 * single integer. Use it in place of read()
464 */
Tejun Heo182446d2013-08-08 20:11:24 -0400465 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700466 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700467 * read_s64() is a signed version of read_u64()
468 */
Tejun Heo182446d2013-08-08 20:11:24 -0400469 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
Tejun Heo2da8ca82013-12-05 12:28:04 -0500470
471 /* generic seq_file read interface */
472 int (*seq_show)(struct seq_file *sf, void *v);
Paul Menage91796562008-04-29 01:00:01 -0700473
Tejun Heo6612f052013-12-05 12:28:04 -0500474 /* optional ops, implement all or none */
475 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
476 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
477 void (*seq_stop)(struct seq_file *sf, void *v);
478
Paul Menage355e0c42007-10-18 23:39:33 -0700479 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700480 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700481 * a single integer (as parsed by simple_strtoull) from
482 * userspace. Use in place of write(); return 0 or error.
483 */
Tejun Heo182446d2013-08-08 20:11:24 -0400484 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
485 u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700486 /*
487 * write_s64() is a signed version of write_u64()
488 */
Tejun Heo182446d2013-08-08 20:11:24 -0400489 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
490 s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700491
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700492 /*
Tejun Heob4168642014-05-13 12:16:21 -0400493 * write() is the generic write callback which maps directly to
494 * kernfs write operation and overrides all other operations.
495 * Maximum write size is determined by ->max_write_len. Use
496 * of_css/cft() to access the associated css and cft.
497 */
498 ssize_t (*write)(struct kernfs_open_file *of,
499 char *buf, size_t nbytes, loff_t off);
500
Tejun Heo2bd59d42014-02-11 11:52:49 -0500501#ifdef CONFIG_DEBUG_LOCK_ALLOC
502 struct lock_class_key lockdep_key;
503#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700504};
505
Tejun Heoa2dd4242014-03-19 10:23:55 -0400506extern struct cgroup_root cgrp_dfl_root;
Tejun Heo5024ae22014-05-07 21:31:17 -0400507extern struct css_set init_css_set;
Tejun Heoa2dd4242014-03-19 10:23:55 -0400508
509static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
510{
511 return cgrp->root == &cgrp_dfl_root;
512}
513
Tejun Heo8e3f6542012-04-01 12:09:55 -0700514/*
Tejun Heo873fe092013-04-14 20:15:26 -0700515 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
516 * function can be called as long as @cgrp is accessible.
517 */
518static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
519{
520 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
521}
522
Tejun Heo07bc3562014-02-13 06:58:39 -0500523/* no synchronization, the result can only be used as a hint */
524static inline bool cgroup_has_tasks(struct cgroup *cgrp)
525{
526 return !list_empty(&cgrp->cset_links);
527}
528
Tejun Heob1664922014-02-11 11:52:49 -0500529/* returns ino associated with a cgroup, 0 indicates unmounted root */
530static inline ino_t cgroup_ino(struct cgroup *cgrp)
531{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500532 if (cgrp->kn)
533 return cgrp->kn->ino;
Tejun Heob1664922014-02-11 11:52:49 -0500534 else
535 return 0;
536}
537
Tejun Heob4168642014-05-13 12:16:21 -0400538/* cft/css accessors for cftype->write() operation */
539static inline struct cftype *of_cft(struct kernfs_open_file *of)
Tejun Heo7da11272013-12-05 12:28:04 -0500540{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500541 return of->kn->priv;
Tejun Heo7da11272013-12-05 12:28:04 -0500542}
543
Tejun Heob4168642014-05-13 12:16:21 -0400544struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
545
546/* cft/css accessors for cftype->seq_*() operations */
547static inline struct cftype *seq_cft(struct seq_file *seq)
548{
549 return of_cft(seq->private);
550}
551
552static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
553{
554 return of_css(seq->private);
555}
Tejun Heo59f52962014-02-11 11:52:49 -0500556
Tejun Heoe61734c2014-02-12 09:29:50 -0500557/*
558 * Name / path handling functions. All are thin wrappers around the kernfs
559 * counterparts and can be called under any context.
560 */
561
562static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
563{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400564 return kernfs_name(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500565}
566
567static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
568 size_t buflen)
569{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400570 return kernfs_path(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500571}
572
573static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
574{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400575 pr_cont_kernfs_name(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500576}
577
578static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
579{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400580 pr_cont_kernfs_path(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500581}
582
583char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
584
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400585int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
Tejun Heo2bb566c2013-08-08 20:11:23 -0400586int cgroup_rm_cftypes(struct cftype *cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -0700587
Li Zefan78574cf2013-04-08 19:00:38 -0700588bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700589
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100590/*
Tejun Heo2f7ee562011-12-12 18:12:21 -0800591 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
592 * methods.
593 */
594struct cgroup_taskset;
595struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
596struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
Tejun Heo2f7ee562011-12-12 18:12:21 -0800597
598/**
599 * cgroup_taskset_for_each - iterate cgroup_taskset
600 * @task: the loop cursor
Tejun Heo2f7ee562011-12-12 18:12:21 -0800601 * @tset: taskset to iterate
602 */
Tejun Heo924f0d92014-02-13 06:58:41 -0500603#define cgroup_taskset_for_each(task, tset) \
Tejun Heo2f7ee562011-12-12 18:12:21 -0800604 for ((task) = cgroup_taskset_first((tset)); (task); \
Tejun Heo924f0d92014-02-13 06:58:41 -0500605 (task) = cgroup_taskset_next((tset)))
Tejun Heo2f7ee562011-12-12 18:12:21 -0800606
607/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100608 * Control Group subsystem type.
609 * See Documentation/cgroups/cgroups.txt for details
610 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700611
612struct cgroup_subsys {
Tejun Heoeb954192013-08-08 20:11:23 -0400613 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
614 int (*css_online)(struct cgroup_subsys_state *css);
615 void (*css_offline)(struct cgroup_subsys_state *css);
616 void (*css_free)(struct cgroup_subsys_state *css);
Tejun Heo92fb9742012-11-19 08:13:38 -0800617
Tejun Heoeb954192013-08-08 20:11:23 -0400618 int (*can_attach)(struct cgroup_subsys_state *css,
619 struct cgroup_taskset *tset);
620 void (*cancel_attach)(struct cgroup_subsys_state *css,
621 struct cgroup_taskset *tset);
622 void (*attach)(struct cgroup_subsys_state *css,
623 struct cgroup_taskset *tset);
Li Zefan761b3ef2012-01-31 13:47:36 +0800624 void (*fork)(struct task_struct *task);
Tejun Heoeb954192013-08-08 20:11:23 -0400625 void (*exit)(struct cgroup_subsys_state *css,
626 struct cgroup_subsys_state *old_css,
Li Zefan761b3ef2012-01-31 13:47:36 +0800627 struct task_struct *task);
Tejun Heoeb954192013-08-08 20:11:23 -0400628 void (*bind)(struct cgroup_subsys_state *root_css);
Hugh Dickinse5991372009-01-06 14:39:22 -0800629
Paul Menage8bab8dd2008-04-04 14:29:57 -0700630 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700631 int early_init;
Tejun Heo48ddbe12012-04-01 12:09:56 -0700632
633 /*
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700634 * If %false, this subsystem is properly hierarchical -
635 * configuration, resource accounting and restriction on a parent
636 * cgroup cover those of its children. If %true, hierarchy support
637 * is broken in some ways - some subsystems ignore hierarchy
638 * completely while others are only implemented half-way.
639 *
640 * It's now disallowed to create nested cgroups if the subsystem is
641 * broken and cgroup core will emit a warning message on such
642 * cases. Eventually, all subsystems will be made properly
643 * hierarchical and this will go away.
644 */
645 bool broken_hierarchy;
646 bool warned_broken_hierarchy;
647
Tejun Heo073219e2014-02-08 10:36:58 -0500648 /* the following two fields are initialized automtically during boot */
Tejun Heoaec25022014-02-08 10:36:58 -0500649 int id;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700650#define MAX_CGROUP_TYPE_NAMELEN 32
651 const char *name;
652
Tejun Heob85d2042013-12-06 15:11:57 -0500653 /* link to parent, protected by cgroup_lock() */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400654 struct cgroup_root *root;
Ben Blume6a11052010-03-10 15:22:09 -0800655
Tejun Heo15a4c832014-05-04 15:09:14 -0400656 /* idr for css->id */
657 struct idr css_idr;
658
Tejun Heo0adb0702014-02-12 09:29:48 -0500659 /*
660 * List of cftypes. Each entry is the first entry of an array
661 * terminated by zero length name.
662 */
663 struct list_head cfts;
Tejun Heo8e3f6542012-04-01 12:09:55 -0700664
Tejun Heode00ffa2014-02-11 11:52:48 -0500665 /* base cftypes, automatically registered with subsys itself */
Tejun Heo8e3f6542012-04-01 12:09:55 -0700666 struct cftype *base_cftypes;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700667};
668
Tejun Heo073219e2014-02-08 10:36:58 -0500669#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700670#include <linux/cgroup_subsys.h>
671#undef SUBSYS
672
Tejun Heo8af01f52013-08-08 20:11:22 -0400673/**
Tejun Heo14611e52013-06-25 11:48:32 -0700674 * task_css_set_check - obtain a task's css_set with extra access conditions
675 * @task: the task to obtain css_set for
676 * @__c: extra condition expression to be passed to rcu_dereference_check()
677 *
678 * A task's css_set is RCU protected, initialized and exited while holding
679 * task_lock(), and can only be modified while holding both cgroup_mutex
680 * and task_lock() while the task is alive. This macro verifies that the
681 * caller is inside proper critical section and returns @task's css_set.
682 *
683 * The caller can also specify additional allowed conditions via @__c, such
684 * as locks used during the cgroup_subsys::attach() methods.
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200685 */
Tejun Heo22194492013-04-07 09:29:51 -0700686#ifdef CONFIG_PROVE_RCU
687extern struct mutex cgroup_mutex;
Tejun Heo0e1d7682014-02-25 10:04:03 -0500688extern struct rw_semaphore css_set_rwsem;
Tejun Heo14611e52013-06-25 11:48:32 -0700689#define task_css_set_check(task, __c) \
690 rcu_dereference_check((task)->cgroups, \
Tejun Heo0e1d7682014-02-25 10:04:03 -0500691 lockdep_is_held(&cgroup_mutex) || \
692 lockdep_is_held(&css_set_rwsem) || \
693 ((task)->flags & PF_EXITING) || (__c))
Tejun Heo22194492013-04-07 09:29:51 -0700694#else
Tejun Heo14611e52013-06-25 11:48:32 -0700695#define task_css_set_check(task, __c) \
696 rcu_dereference((task)->cgroups)
Tejun Heo22194492013-04-07 09:29:51 -0700697#endif
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200698
Tejun Heo14611e52013-06-25 11:48:32 -0700699/**
Tejun Heo8af01f52013-08-08 20:11:22 -0400700 * task_css_check - obtain css for (task, subsys) w/ extra access conds
Tejun Heo14611e52013-06-25 11:48:32 -0700701 * @task: the target task
702 * @subsys_id: the target subsystem ID
703 * @__c: extra condition expression to be passed to rcu_dereference_check()
704 *
705 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
706 * synchronization rules are the same as task_css_set_check().
707 */
Tejun Heo8af01f52013-08-08 20:11:22 -0400708#define task_css_check(task, subsys_id, __c) \
Tejun Heo14611e52013-06-25 11:48:32 -0700709 task_css_set_check((task), (__c))->subsys[(subsys_id)]
710
711/**
712 * task_css_set - obtain a task's css_set
713 * @task: the task to obtain css_set for
714 *
715 * See task_css_set_check().
716 */
717static inline struct css_set *task_css_set(struct task_struct *task)
718{
719 return task_css_set_check(task, false);
720}
721
722/**
Tejun Heo8af01f52013-08-08 20:11:22 -0400723 * task_css - obtain css for (task, subsys)
Tejun Heo14611e52013-06-25 11:48:32 -0700724 * @task: the target task
725 * @subsys_id: the target subsystem ID
726 *
Tejun Heo8af01f52013-08-08 20:11:22 -0400727 * See task_css_check().
Tejun Heo14611e52013-06-25 11:48:32 -0700728 */
Tejun Heo8af01f52013-08-08 20:11:22 -0400729static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
730 int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700731{
Tejun Heo8af01f52013-08-08 20:11:22 -0400732 return task_css_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700733}
734
Tejun Heo5024ae22014-05-07 21:31:17 -0400735/**
736 * task_css_is_root - test whether a task belongs to the root css
737 * @task: the target task
738 * @subsys_id: the target subsystem ID
739 *
740 * Test whether @task belongs to the root css on the specified subsystem.
741 * May be invoked in any context.
742 */
743static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
744{
745 return task_css_check(task, subsys_id, true) ==
746 init_css_set.subsys[subsys_id];
747}
748
Tejun Heo8af01f52013-08-08 20:11:22 -0400749static inline struct cgroup *task_cgroup(struct task_struct *task,
750 int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700751{
Tejun Heo8af01f52013-08-08 20:11:22 -0400752 return task_css(task, subsys_id)->cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700753}
754
Tejun Heo492eb212013-08-08 20:11:25 -0400755struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
756 struct cgroup_subsys_state *parent);
Tejun Heo53fa5262013-05-24 10:55:38 +0900757
Li Zefan1cb650b2013-08-19 10:05:24 +0800758struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
759
Tejun Heo574bd9f2012-11-09 09:12:29 -0800760/**
Tejun Heo492eb212013-08-08 20:11:25 -0400761 * css_for_each_child - iterate through children of a css
762 * @pos: the css * to use as the loop cursor
763 * @parent: css whose children to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800764 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400765 * Walk @parent's children. Must be called under rcu_read_lock().
Tejun Heo574bd9f2012-11-09 09:12:29 -0800766 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400767 * If a subsystem synchronizes ->css_online() and the start of iteration, a
768 * css which finished ->css_online() is guaranteed to be visible in the
769 * future iterations and will stay visible until the last reference is put.
770 * A css which hasn't finished ->css_online() or already finished
771 * ->css_offline() may show up during traversal. It's each subsystem's
772 * responsibility to synchronize against on/offlining.
Tejun Heo75501a62013-05-24 10:55:38 +0900773 *
774 * It is allowed to temporarily drop RCU read lock during iteration. The
775 * caller is responsible for ensuring that @pos remains accessible until
776 * the start of the next iteration by, for example, bumping the css refcnt.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800777 */
Tejun Heo492eb212013-08-08 20:11:25 -0400778#define css_for_each_child(pos, parent) \
779 for ((pos) = css_next_child(NULL, (parent)); (pos); \
780 (pos) = css_next_child((pos), (parent)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800781
Tejun Heo492eb212013-08-08 20:11:25 -0400782struct cgroup_subsys_state *
783css_next_descendant_pre(struct cgroup_subsys_state *pos,
784 struct cgroup_subsys_state *css);
785
786struct cgroup_subsys_state *
787css_rightmost_descendant(struct cgroup_subsys_state *pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -0800788
789/**
Tejun Heo492eb212013-08-08 20:11:25 -0400790 * css_for_each_descendant_pre - pre-order walk of a css's descendants
791 * @pos: the css * to use as the loop cursor
792 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800793 *
Tejun Heobd8815a2013-08-08 20:11:27 -0400794 * Walk @root's descendants. @root is included in the iteration and the
Tejun Heoc2931b72014-05-16 13:22:51 -0400795 * first node to be visited. Must be called under rcu_read_lock().
Tejun Heo574bd9f2012-11-09 09:12:29 -0800796 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400797 * If a subsystem synchronizes ->css_online() and the start of iteration, a
798 * css which finished ->css_online() is guaranteed to be visible in the
799 * future iterations and will stay visible until the last reference is put.
800 * A css which hasn't finished ->css_online() or already finished
801 * ->css_offline() may show up during traversal. It's each subsystem's
802 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800803 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400804 * For example, the following guarantees that a descendant can't escape
Tejun Heo574bd9f2012-11-09 09:12:29 -0800805 * state updates of its ancestors.
806 *
Tejun Heo492eb212013-08-08 20:11:25 -0400807 * my_online(@css)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800808 * {
Tejun Heo492eb212013-08-08 20:11:25 -0400809 * Lock @css's parent and @css;
810 * Inherit state from the parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -0800811 * Unlock both.
812 * }
813 *
Tejun Heo492eb212013-08-08 20:11:25 -0400814 * my_update_state(@css)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800815 * {
Tejun Heo492eb212013-08-08 20:11:25 -0400816 * css_for_each_descendant_pre(@pos, @css) {
Tejun Heo574bd9f2012-11-09 09:12:29 -0800817 * Lock @pos;
Tejun Heobd8815a2013-08-08 20:11:27 -0400818 * if (@pos == @css)
819 * Update @css's state;
820 * else
821 * Verify @pos is alive and inherit state from its parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -0800822 * Unlock @pos;
823 * }
824 * }
825 *
826 * As long as the inheriting step, including checking the parent state, is
827 * enclosed inside @pos locking, double-locking the parent isn't necessary
828 * while inheriting. The state update to the parent is guaranteed to be
829 * visible by walking order and, as long as inheriting operations to the
830 * same @pos are atomic to each other, multiple updates racing each other
831 * still result in the correct state. It's guaranateed that at least one
Tejun Heo492eb212013-08-08 20:11:25 -0400832 * inheritance happens for any css after the latest update to its parent.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800833 *
834 * If checking parent's state requires locking the parent, each inheriting
835 * iteration should lock and unlock both @pos->parent and @pos.
836 *
837 * Alternatively, a subsystem may choose to use a single global lock to
Tejun Heo92fb9742012-11-19 08:13:38 -0800838 * synchronize ->css_online() and ->css_offline() against tree-walking
Tejun Heo574bd9f2012-11-09 09:12:29 -0800839 * operations.
Tejun Heo75501a62013-05-24 10:55:38 +0900840 *
841 * It is allowed to temporarily drop RCU read lock during iteration. The
842 * caller is responsible for ensuring that @pos remains accessible until
843 * the start of the next iteration by, for example, bumping the css refcnt.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800844 */
Tejun Heo492eb212013-08-08 20:11:25 -0400845#define css_for_each_descendant_pre(pos, css) \
846 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
847 (pos) = css_next_descendant_pre((pos), (css)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800848
Tejun Heo492eb212013-08-08 20:11:25 -0400849struct cgroup_subsys_state *
850css_next_descendant_post(struct cgroup_subsys_state *pos,
851 struct cgroup_subsys_state *css);
Tejun Heo574bd9f2012-11-09 09:12:29 -0800852
853/**
Tejun Heo492eb212013-08-08 20:11:25 -0400854 * css_for_each_descendant_post - post-order walk of a css's descendants
855 * @pos: the css * to use as the loop cursor
856 * @css: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800857 *
Tejun Heo492eb212013-08-08 20:11:25 -0400858 * Similar to css_for_each_descendant_pre() but performs post-order
Tejun Heobd8815a2013-08-08 20:11:27 -0400859 * traversal instead. @root is included in the iteration and the last
Tejun Heoc2931b72014-05-16 13:22:51 -0400860 * node to be visited.
861 *
862 * If a subsystem synchronizes ->css_online() and the start of iteration, a
863 * css which finished ->css_online() is guaranteed to be visible in the
864 * future iterations and will stay visible until the last reference is put.
865 * A css which hasn't finished ->css_online() or already finished
866 * ->css_offline() may show up during traversal. It's each subsystem's
867 * responsibility to synchronize against on/offlining.
868 *
869 * Note that the walk visibility guarantee example described in pre-order
870 * walk doesn't apply the same to post-order walks.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800871 */
Tejun Heo492eb212013-08-08 20:11:25 -0400872#define css_for_each_descendant_post(pos, css) \
873 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
874 (pos) = css_next_descendant_post((pos), (css)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800875
Tejun Heof3d46502014-05-16 13:22:52 -0400876bool css_has_online_children(struct cgroup_subsys_state *css);
877
Tejun Heo72ec7022013-08-08 20:11:26 -0400878/* A css_task_iter should be treated as an opaque object */
879struct css_task_iter {
Tejun Heo3ebb2b62014-04-23 11:13:15 -0400880 struct cgroup_subsys *ss;
881
Tejun Heo0f0a2b42014-04-23 11:13:15 -0400882 struct list_head *cset_pos;
883 struct list_head *cset_head;
884
885 struct list_head *task_pos;
886 struct list_head *tasks_head;
887 struct list_head *mg_tasks_head;
Paul Menage817929e2007-10-18 23:39:36 -0700888};
889
Tejun Heo72ec7022013-08-08 20:11:26 -0400890void css_task_iter_start(struct cgroup_subsys_state *css,
891 struct css_task_iter *it);
892struct task_struct *css_task_iter_next(struct css_task_iter *it);
893void css_task_iter_end(struct css_task_iter *it);
Tejun Heoe5358372013-08-08 20:11:26 -0400894
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700895int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
Tejun Heo8cc99342013-04-07 09:29:50 -0700896int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700897
Tejun Heoec903c02014-05-13 12:11:01 -0400898struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
899 struct cgroup_subsys *ss);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700900
Paul Menageddbcc7e2007-10-18 23:39:30 -0700901#else /* !CONFIG_CGROUPS */
902
903static inline int cgroup_init_early(void) { return 0; }
904static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700905static inline void cgroup_fork(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700906static inline void cgroup_post_fork(struct task_struct *p) {}
Li Zefan1ec41832014-03-28 15:22:19 +0800907static inline void cgroup_exit(struct task_struct *p) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700908
Balbir Singh846c7bb2007-10-18 23:39:44 -0700909static inline int cgroupstats_build(struct cgroupstats *stats,
910 struct dentry *dentry)
911{
912 return -EINVAL;
913}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700914
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200915/* No cgroups - nothing to do */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700916static inline int cgroup_attach_task_all(struct task_struct *from,
917 struct task_struct *t)
918{
919 return 0;
920}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200921
Paul Menageddbcc7e2007-10-18 23:39:30 -0700922#endif /* !CONFIG_CGROUPS */
923
Paul Menageddbcc7e2007-10-18 23:39:30 -0700924#endif /* _LINUX_CGROUP_H */