blob: da0dae0600e67292849de21a5649d87f4f5e5873 [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;
Paul Menage84eea842008-07-25 01:47:00 -070030struct cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -070031
32extern int cgroup_init_early(void);
33extern int cgroup_init(void);
Paul Menageb4f48b62007-10-18 23:39:33 -070034extern void cgroup_fork(struct task_struct *p);
Paul Menage817929e2007-10-18 23:39:36 -070035extern void cgroup_post_fork(struct task_struct *p);
Li Zefan1ec41832014-03-28 15:22:19 +080036extern void cgroup_exit(struct task_struct *p);
Balbir Singh846c7bb2007-10-18 23:39:44 -070037extern int cgroupstats_build(struct cgroupstats *stats,
38 struct dentry *dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -070039
Zefan Li006f4ac2014-09-18 16:03:15 +080040extern int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
41 struct pid *pid, struct task_struct *tsk);
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/**
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800116 * css_get_many - obtain references on the specified css
117 * @css: target css
118 * @n: number of references to get
119 *
120 * The caller must already have a reference.
121 */
122static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
123{
124 if (!(css->flags & CSS_NO_REF))
125 percpu_ref_get_many(&css->refcnt, n);
126}
127
128/**
Tejun Heo6f4524d2014-05-16 13:22:52 -0400129 * css_tryget - try to obtain a reference on the specified css
130 * @css: target css
131 *
132 * Obtain a reference on @css unless it already has reached zero and is
133 * being released. This function doesn't care whether @css is on or
134 * offline. The caller naturally needs to ensure that @css is accessible
135 * but doesn't have to be holding a reference on it - IOW, RCU protected
136 * access is good enough for this function. Returns %true if a reference
137 * count was successfully obtained; %false otherwise.
138 */
139static inline bool css_tryget(struct cgroup_subsys_state *css)
140{
141 if (!(css->flags & CSS_NO_REF))
142 return percpu_ref_tryget(&css->refcnt);
143 return true;
144}
145
146/**
Tejun Heoec903c02014-05-13 12:11:01 -0400147 * css_tryget_online - try to obtain a reference on the specified css if online
Tejun Heo5de01072013-06-12 21:04:52 -0700148 * @css: target css
149 *
Tejun Heoec903c02014-05-13 12:11:01 -0400150 * Obtain a reference on @css if it's online. The caller naturally needs
151 * to ensure that @css is accessible but doesn't have to be holding a
Tejun Heo5de01072013-06-12 21:04:52 -0700152 * reference on it - IOW, RCU protected access is good enough for this
153 * function. Returns %true if a reference count was successfully obtained;
154 * %false otherwise.
155 */
Tejun Heoec903c02014-05-13 12:11:01 -0400156static inline bool css_tryget_online(struct cgroup_subsys_state *css)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800157{
Tejun Heo3b514d22014-05-16 13:22:47 -0400158 if (!(css->flags & CSS_NO_REF))
159 return percpu_ref_tryget_live(&css->refcnt);
160 return true;
Paul Menagee7c5ec92009-01-07 18:08:38 -0800161}
162
Tejun Heo5de01072013-06-12 21:04:52 -0700163/**
164 * css_put - put a css reference
165 * @css: target css
166 *
Tejun Heoec903c02014-05-13 12:11:01 -0400167 * Put a reference obtained via css_get() and css_tryget_online().
Tejun Heo5de01072013-06-12 21:04:52 -0700168 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700169static inline void css_put(struct cgroup_subsys_state *css)
170{
Tejun Heo3b514d22014-05-16 13:22:47 -0400171 if (!(css->flags & CSS_NO_REF))
172 percpu_ref_put(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700173}
174
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800175/**
176 * css_put_many - put css references
177 * @css: target css
178 * @n: number of references to put
179 *
180 * Put references obtained via css_get() and css_tryget_online().
181 */
182static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
183{
184 if (!(css->flags & CSS_NO_REF))
185 percpu_ref_put_many(&css->refcnt, n);
186}
187
Paul Menage3116f0e2008-04-29 01:00:04 -0700188/* bits in struct cgroup flags field */
189enum {
Paul Menage3116f0e2008-04-29 01:00:04 -0700190 /* Control Group requires release notifications to userspace */
191 CGRP_NOTIFY_ON_RELEASE,
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700192 /*
Tejun Heo2260e7f2012-11-19 08:13:38 -0800193 * Clone the parent's configuration when creating a new child
194 * cpuset cgroup. For historical reasons, this option can be
195 * specified at mount time and thus is implemented here.
Daniel Lezcano97978e62010-10-27 15:33:35 -0700196 */
Tejun Heo2260e7f2012-11-19 08:13:38 -0800197 CGRP_CPUSET_CLONE_CHILDREN,
Paul Menage3116f0e2008-04-29 01:00:04 -0700198};
199
Paul Menageddbcc7e2007-10-18 23:39:30 -0700200struct cgroup {
Tejun Heo9d800df2014-05-14 09:15:00 -0400201 /* self css with NULL ->ss, points back to this cgroup */
202 struct cgroup_subsys_state self;
203
Paul Menageddbcc7e2007-10-18 23:39:30 -0700204 unsigned long flags; /* "unsigned long" so bitops work */
205
Li Zefanb414dc02013-07-31 09:51:06 +0800206 /*
207 * idr allocated in-hierarchy ID.
208 *
Tejun Heo7d699dd2014-05-04 15:09:13 -0400209 * ID 0 is not used, the ID of the root cgroup is always 1, and a
210 * new cgroup will be assigned with a smallest available ID.
Li Zefan0ab02ca2014-02-11 16:05:46 +0800211 *
212 * Allocating/Removing ID must be protected by cgroup_mutex.
Li Zefanb414dc02013-07-31 09:51:06 +0800213 */
214 int id;
Tejun Heo0a950f62012-11-19 09:02:12 -0800215
Tejun Heo842b5972014-04-25 18:28:02 -0400216 /*
217 * If this cgroup contains any tasks, it contributes one to
218 * populated_cnt. All children with non-zero popuplated_cnt of
219 * their own contribute one. The count is zero iff there's no task
220 * in this cgroup or its subtree.
221 */
222 int populated_cnt;
223
Tejun Heo2bd59d42014-02-11 11:52:49 -0500224 struct kernfs_node *kn; /* cgroup kernfs entry */
Tejun Heo842b5972014-04-25 18:28:02 -0400225 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700226
Tejun Heo667c2492014-07-08 18:02:56 -0400227 /*
228 * The bitmask of subsystems enabled on the child cgroups.
229 * ->subtree_control is the one configured through
230 * "cgroup.subtree_control" while ->child_subsys_mask is the
231 * effective one which may have more subsystems enabled.
Tejun Heof63070d2014-07-08 18:02:57 -0400232 * Controller knobs are made available iff it's enabled in
233 * ->subtree_control.
Tejun Heo667c2492014-07-08 18:02:56 -0400234 */
235 unsigned int subtree_control;
Tejun Heo69dfa002014-05-04 15:09:13 -0400236 unsigned int child_subsys_mask;
Tejun Heo94419622014-03-19 10:23:54 -0400237
Paul Menageddbcc7e2007-10-18 23:39:30 -0700238 /* Private pointers for each registered subsystem */
Tejun Heo73e80ed2013-08-13 11:01:55 -0400239 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700240
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400241 struct cgroup_root *root;
Paul Menage817929e2007-10-18 23:39:36 -0700242
243 /*
Tejun Heo69d02062013-06-12 21:04:50 -0700244 * List of cgrp_cset_links pointing at css_sets with tasks in this
245 * cgroup. Protected by css_set_lock.
Paul Menage817929e2007-10-18 23:39:36 -0700246 */
Tejun Heo69d02062013-06-12 21:04:50 -0700247 struct list_head cset_links;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700248
249 /*
Tejun Heo2d8f2432014-04-23 11:13:15 -0400250 * On the default hierarchy, a css_set for a cgroup with some
251 * susbsys disabled will point to css's which are associated with
252 * the closest ancestor which has the subsys enabled. The
253 * following lists all css_sets which point to this cgroup's css
254 * for the given subsystem.
255 */
256 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
257
258 /*
Ben Blum72a8cb32009-09-23 15:56:27 -0700259 * list of pidlists, up to two for each namespace (one for procs, one
260 * for tasks); created on demand.
261 */
262 struct list_head pidlists;
263 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800264
Tejun Heof8f22e52014-04-23 11:13:16 -0400265 /* used to wait for offlining of csses */
266 wait_queue_head_t offline_waitq;
Zefan Li971ff492014-09-18 16:06:19 +0800267
268 /* used to schedule release agent */
269 struct work_struct release_agent_work;
Paul Menage817929e2007-10-18 23:39:36 -0700270};
271
Tejun Heo25a7e682013-04-14 20:15:25 -0700272#define MAX_CGROUP_ROOT_NAMELEN 64
273
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400274/* cgroup_root->flags */
Tejun Heo25a7e682013-04-14 20:15:25 -0700275enum {
Tejun Heoaa6ec292014-07-09 10:08:08 -0400276 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */
Tejun Heo25a7e682013-04-14 20:15:25 -0700277 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
278 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
279};
280
281/*
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400282 * A cgroup_root represents the root of a cgroup hierarchy, and may be
Tejun Heo2bd59d42014-02-11 11:52:49 -0500283 * associated with a kernfs_root to form an active hierarchy. This is
Tejun Heo25a7e682013-04-14 20:15:25 -0700284 * internal to cgroup core. Don't access directly from controllers.
285 */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400286struct cgroup_root {
Tejun Heo2bd59d42014-02-11 11:52:49 -0500287 struct kernfs_root *kf_root;
Tejun Heo25a7e682013-04-14 20:15:25 -0700288
Tejun Heof392e512014-04-23 11:13:14 -0400289 /* The bitmask of subsystems attached to this hierarchy */
Tejun Heo69dfa002014-05-04 15:09:13 -0400290 unsigned int subsys_mask;
Tejun Heof392e512014-04-23 11:13:14 -0400291
Tejun Heo25a7e682013-04-14 20:15:25 -0700292 /* Unique id for this hierarchy. */
293 int hierarchy_id;
294
Tejun Heo776f02f2014-02-12 09:29:50 -0500295 /* The root cgroup. Root is destroyed on its release. */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400296 struct cgroup cgrp;
Tejun Heo25a7e682013-04-14 20:15:25 -0700297
Tejun Heo3c9c8252014-02-12 09:29:50 -0500298 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
299 atomic_t nr_cgrps;
Tejun Heo25a7e682013-04-14 20:15:25 -0700300
301 /* A list running through the active hierarchies */
302 struct list_head root_list;
303
Tejun Heo25a7e682013-04-14 20:15:25 -0700304 /* Hierarchy-specific flags */
Tejun Heo69dfa002014-05-04 15:09:13 -0400305 unsigned int flags;
Tejun Heo25a7e682013-04-14 20:15:25 -0700306
307 /* IDs for cgroups in this hierarchy */
Li Zefan4e96ee8e2013-07-31 09:50:50 +0800308 struct idr cgroup_idr;
Tejun Heo25a7e682013-04-14 20:15:25 -0700309
310 /* The path to use for release notifications. */
311 char release_agent_path[PATH_MAX];
312
313 /* The name for this hierarchy - may be empty */
314 char name[MAX_CGROUP_ROOT_NAMELEN];
315};
316
Paul Menaged20a3902009-04-02 16:57:22 -0700317/*
318 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700319 * cgroup_subsys_state objects. This saves space in the task struct
320 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700321 * list_add()/del() can bump the reference count on the entire cgroup
322 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700323 */
324
325struct css_set {
326
327 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700328 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700329
330 /*
Li Zefan472b1052008-04-29 01:00:11 -0700331 * List running through all cgroup groups in the same hash
332 * slot. Protected by css_set_lock
333 */
334 struct hlist_node hlist;
335
336 /*
Tejun Heoc7561122014-02-25 10:04:01 -0500337 * Lists running through all tasks using this cgroup group.
338 * mg_tasks lists tasks which belong to this cset but are in the
339 * process of being migrated out or in. Protected by
340 * css_set_rwsem, but, during migration, once tasks are moved to
341 * mg_tasks, it can be read safely while holding cgroup_mutex.
Paul Menage817929e2007-10-18 23:39:36 -0700342 */
343 struct list_head tasks;
Tejun Heoc7561122014-02-25 10:04:01 -0500344 struct list_head mg_tasks;
Paul Menage817929e2007-10-18 23:39:36 -0700345
346 /*
Tejun Heo69d02062013-06-12 21:04:50 -0700347 * List of cgrp_cset_links pointing at cgroups referenced from this
348 * css_set. Protected by css_set_lock.
Paul Menage817929e2007-10-18 23:39:36 -0700349 */
Tejun Heo69d02062013-06-12 21:04:50 -0700350 struct list_head cgrp_links;
Paul Menage817929e2007-10-18 23:39:36 -0700351
Tejun Heo6803c002014-04-23 11:13:16 -0400352 /* the default cgroup associated with this css_set */
353 struct cgroup *dfl_cgrp;
354
Paul Menage817929e2007-10-18 23:39:36 -0700355 /*
Tejun Heo3ed80a62014-02-08 10:36:58 -0500356 * Set of subsystem states, one for each subsystem. This array is
357 * immutable after creation apart from the init_css_set during
358 * subsystem registration (at boot time).
Paul Menage817929e2007-10-18 23:39:36 -0700359 */
360 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700361
Tejun Heob3dc0942014-02-25 10:04:01 -0500362 /*
363 * List of csets participating in the on-going migration either as
364 * source or destination. Protected by cgroup_mutex.
365 */
Tejun Heo1958d2d2014-02-25 10:04:03 -0500366 struct list_head mg_preload_node;
Tejun Heob3dc0942014-02-25 10:04:01 -0500367 struct list_head mg_node;
368
369 /*
370 * If this cset is acting as the source of migration the following
371 * two fields are set. mg_src_cgrp is the source cgroup of the
372 * on-going migration and mg_dst_cset is the destination cset the
373 * target tasks on this cset should be migrated to. Protected by
374 * cgroup_mutex.
375 */
376 struct cgroup *mg_src_cgrp;
377 struct css_set *mg_dst_cset;
378
Tejun Heo2d8f2432014-04-23 11:13:15 -0400379 /*
380 * On the default hierarhcy, ->subsys[ssid] may point to a css
381 * attached to an ancestor instead of the cgroup this css_set is
382 * associated with. The following node is anchored at
383 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
384 * iterate through all css's attached to a given cgroup.
385 */
386 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
387
Ben Blumc3783692009-09-23 15:56:29 -0700388 /* For RCU-protected deletion */
389 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700390};
391
Paul Menage91796562008-04-29 01:00:01 -0700392/*
Paul Menaged20a3902009-04-02 16:57:22 -0700393 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700394 *
395 * When reading/writing to a file:
Al Virob5830432014-10-31 01:22:04 -0400396 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
397 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700398 */
399
Tejun Heo8e3f6542012-04-01 12:09:55 -0700400/* cftype->flags */
Tejun Heo02c402d2013-06-24 15:21:47 -0700401enum {
Li Zefan6f4b7e62013-07-31 16:18:36 +0800402 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
403 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
Tejun Heo9fa4db32013-08-26 18:40:56 -0400404 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
Tejun Heo05ebb6e2014-07-15 11:05:10 -0400405
406 /* internal flags, do not use outside cgroup core proper */
407 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
408 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
Tejun Heo02c402d2013-06-24 15:21:47 -0700409};
Tejun Heo8e3f6542012-04-01 12:09:55 -0700410
411#define MAX_CFTYPE_NAME 64
412
Paul Menageddbcc7e2007-10-18 23:39:30 -0700413struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700414 /*
415 * By convention, the name should begin with the name of the
Tejun Heo8e3f6542012-04-01 12:09:55 -0700416 * subsystem, followed by a period. Zero length string indicates
417 * end of cftype array.
Paul Menaged20a3902009-04-02 16:57:22 -0700418 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700419 char name[MAX_CFTYPE_NAME];
420 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700421 /*
422 * If not 0, file mode is set to this value, otherwise it will
423 * be figured out automatically
424 */
Al Viroa5e7ed32011-07-26 01:55:55 -0400425 umode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700426
427 /*
Tejun Heo5f469902014-02-11 11:52:48 -0500428 * The maximum length of string, excluding trailing nul, that can
Tejun Heo451af502014-05-13 12:16:21 -0400429 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
Paul Menagedb3b1492008-07-25 01:46:58 -0700430 */
431 size_t max_write_len;
432
Tejun Heo8e3f6542012-04-01 12:09:55 -0700433 /* CFTYPE_* flags */
434 unsigned int flags;
435
Tejun Heo2bb566c2013-08-08 20:11:23 -0400436 /*
Tejun Heo0adb0702014-02-12 09:29:48 -0500437 * Fields used for internal bookkeeping. Initialized automatically
438 * during registration.
Tejun Heo2bb566c2013-08-08 20:11:23 -0400439 */
Tejun Heo0adb0702014-02-12 09:29:48 -0500440 struct cgroup_subsys *ss; /* NULL for cgroup core files */
441 struct list_head node; /* anchored at ss->cfts */
Tejun Heo2bd59d42014-02-11 11:52:49 -0500442 struct kernfs_ops *kf_ops;
443
Paul Menageddbcc7e2007-10-18 23:39:30 -0700444 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700445 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700446 * single integer. Use it in place of read()
447 */
Tejun Heo182446d2013-08-08 20:11:24 -0400448 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700449 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700450 * read_s64() is a signed version of read_u64()
451 */
Tejun Heo182446d2013-08-08 20:11:24 -0400452 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
Tejun Heo2da8ca82013-12-05 12:28:04 -0500453
454 /* generic seq_file read interface */
455 int (*seq_show)(struct seq_file *sf, void *v);
Paul Menage91796562008-04-29 01:00:01 -0700456
Tejun Heo6612f052013-12-05 12:28:04 -0500457 /* optional ops, implement all or none */
458 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
459 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
460 void (*seq_stop)(struct seq_file *sf, void *v);
461
Paul Menage355e0c42007-10-18 23:39:33 -0700462 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700463 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700464 * a single integer (as parsed by simple_strtoull) from
465 * userspace. Use in place of write(); return 0 or error.
466 */
Tejun Heo182446d2013-08-08 20:11:24 -0400467 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
468 u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700469 /*
470 * write_s64() is a signed version of write_u64()
471 */
Tejun Heo182446d2013-08-08 20:11:24 -0400472 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
473 s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700474
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700475 /*
Tejun Heob4168642014-05-13 12:16:21 -0400476 * write() is the generic write callback which maps directly to
477 * kernfs write operation and overrides all other operations.
478 * Maximum write size is determined by ->max_write_len. Use
479 * of_css/cft() to access the associated css and cft.
480 */
481 ssize_t (*write)(struct kernfs_open_file *of,
482 char *buf, size_t nbytes, loff_t off);
483
Tejun Heo2bd59d42014-02-11 11:52:49 -0500484#ifdef CONFIG_DEBUG_LOCK_ALLOC
485 struct lock_class_key lockdep_key;
486#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700487};
488
Tejun Heoa2dd4242014-03-19 10:23:55 -0400489extern struct cgroup_root cgrp_dfl_root;
Tejun Heo5024ae22014-05-07 21:31:17 -0400490extern struct css_set init_css_set;
Tejun Heoa2dd4242014-03-19 10:23:55 -0400491
Tejun Heoaa6ec292014-07-09 10:08:08 -0400492/**
493 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
494 * @cgrp: the cgroup of interest
495 *
496 * The default hierarchy is the v2 interface of cgroup and this function
497 * can be used to test whether a cgroup is on the default hierarchy for
498 * cases where a subsystem should behave differnetly depending on the
499 * interface version.
500 *
501 * The set of behaviors which change on the default hierarchy are still
502 * being determined and the mount option is prefixed with __DEVEL__.
503 *
504 * List of changed behaviors:
505 *
506 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
507 * and "name" are disallowed.
508 *
509 * - When mounting an existing superblock, mount options should match.
510 *
511 * - Remount is disallowed.
512 *
513 * - rename(2) is disallowed.
514 *
515 * - "tasks" is removed. Everything should be at process granularity. Use
516 * "cgroup.procs" instead.
517 *
518 * - "cgroup.procs" is not sorted. pids will be unique unless they got
519 * recycled inbetween reads.
520 *
521 * - "release_agent" and "notify_on_release" are removed. Replacement
522 * notification mechanism will be implemented.
523 *
524 * - "cgroup.clone_children" is removed.
525 *
526 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
527 * and its descendants contain no task; otherwise, 1. The file also
528 * generates kernfs notification which can be monitored through poll and
529 * [di]notify when the value of the file changes.
530 *
531 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
532 * take masks of ancestors with non-empty cpus/mems, instead of being
533 * moved to an ancestor.
534 *
535 * - cpuset: a task can be moved into an empty cpuset, and again it takes
536 * masks of ancestors.
537 *
538 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
539 * is not created.
540 *
541 * - blkcg: blk-throttle becomes properly hierarchical.
542 *
543 * - debug: disallowed on the default hierarchy.
544 */
Tejun Heoa2dd4242014-03-19 10:23:55 -0400545static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
546{
547 return cgrp->root == &cgrp_dfl_root;
548}
549
Tejun Heo07bc3562014-02-13 06:58:39 -0500550/* no synchronization, the result can only be used as a hint */
551static inline bool cgroup_has_tasks(struct cgroup *cgrp)
552{
553 return !list_empty(&cgrp->cset_links);
554}
555
Zefan Lif29374b2014-09-19 16:29:31 +0800556/* returns ino associated with a cgroup */
Tejun Heob1664922014-02-11 11:52:49 -0500557static inline ino_t cgroup_ino(struct cgroup *cgrp)
558{
Zefan Lif29374b2014-09-19 16:29:31 +0800559 return cgrp->kn->ino;
Tejun Heob1664922014-02-11 11:52:49 -0500560}
561
Tejun Heob4168642014-05-13 12:16:21 -0400562/* cft/css accessors for cftype->write() operation */
563static inline struct cftype *of_cft(struct kernfs_open_file *of)
Tejun Heo7da11272013-12-05 12:28:04 -0500564{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500565 return of->kn->priv;
Tejun Heo7da11272013-12-05 12:28:04 -0500566}
567
Tejun Heob4168642014-05-13 12:16:21 -0400568struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
569
570/* cft/css accessors for cftype->seq_*() operations */
571static inline struct cftype *seq_cft(struct seq_file *seq)
572{
573 return of_cft(seq->private);
574}
575
576static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
577{
578 return of_css(seq->private);
579}
Tejun Heo59f52962014-02-11 11:52:49 -0500580
Tejun Heoe61734c2014-02-12 09:29:50 -0500581/*
582 * Name / path handling functions. All are thin wrappers around the kernfs
583 * counterparts and can be called under any context.
584 */
585
586static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
587{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400588 return kernfs_name(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500589}
590
591static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
592 size_t buflen)
593{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400594 return kernfs_path(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500595}
596
597static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
598{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400599 pr_cont_kernfs_name(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500600}
601
602static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
603{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400604 pr_cont_kernfs_path(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500605}
606
607char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
608
Tejun Heoa8ddc822014-07-15 11:05:10 -0400609int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
Tejun Heo2cf669a2014-07-15 11:05:09 -0400610int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
Tejun Heo2bb566c2013-08-08 20:11:23 -0400611int cgroup_rm_cftypes(struct cftype *cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -0700612
Li Zefan78574cf2013-04-08 19:00:38 -0700613bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700614
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100615/*
Tejun Heo2f7ee562011-12-12 18:12:21 -0800616 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
617 * methods.
618 */
619struct cgroup_taskset;
620struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
621struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
Tejun Heo2f7ee562011-12-12 18:12:21 -0800622
623/**
624 * cgroup_taskset_for_each - iterate cgroup_taskset
625 * @task: the loop cursor
Tejun Heo2f7ee562011-12-12 18:12:21 -0800626 * @tset: taskset to iterate
627 */
Tejun Heo924f0d92014-02-13 06:58:41 -0500628#define cgroup_taskset_for_each(task, tset) \
Tejun Heo2f7ee562011-12-12 18:12:21 -0800629 for ((task) = cgroup_taskset_first((tset)); (task); \
Tejun Heo924f0d92014-02-13 06:58:41 -0500630 (task) = cgroup_taskset_next((tset)))
Tejun Heo2f7ee562011-12-12 18:12:21 -0800631
632/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100633 * Control Group subsystem type.
634 * See Documentation/cgroups/cgroups.txt for details
635 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700636
637struct cgroup_subsys {
Tejun Heoeb954192013-08-08 20:11:23 -0400638 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
639 int (*css_online)(struct cgroup_subsys_state *css);
640 void (*css_offline)(struct cgroup_subsys_state *css);
Tejun Heo7d172cc2014-11-18 02:49:51 -0500641 void (*css_released)(struct cgroup_subsys_state *css);
Tejun Heoeb954192013-08-08 20:11:23 -0400642 void (*css_free)(struct cgroup_subsys_state *css);
Tejun Heob4536f0ca2014-07-08 18:02:57 -0400643 void (*css_reset)(struct cgroup_subsys_state *css);
Tejun Heo56c807b2014-11-18 02:49:51 -0500644 void (*css_e_css_changed)(struct cgroup_subsys_state *css);
Tejun Heo92fb9742012-11-19 08:13:38 -0800645
Tejun Heoeb954192013-08-08 20:11:23 -0400646 int (*can_attach)(struct cgroup_subsys_state *css,
647 struct cgroup_taskset *tset);
648 void (*cancel_attach)(struct cgroup_subsys_state *css,
649 struct cgroup_taskset *tset);
650 void (*attach)(struct cgroup_subsys_state *css,
651 struct cgroup_taskset *tset);
Li Zefan761b3ef2012-01-31 13:47:36 +0800652 void (*fork)(struct task_struct *task);
Tejun Heoeb954192013-08-08 20:11:23 -0400653 void (*exit)(struct cgroup_subsys_state *css,
654 struct cgroup_subsys_state *old_css,
Li Zefan761b3ef2012-01-31 13:47:36 +0800655 struct task_struct *task);
Tejun Heoeb954192013-08-08 20:11:23 -0400656 void (*bind)(struct cgroup_subsys_state *root_css);
Hugh Dickinse5991372009-01-06 14:39:22 -0800657
Paul Menage8bab8dd2008-04-04 14:29:57 -0700658 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700659 int early_init;
Tejun Heo48ddbe12012-04-01 12:09:56 -0700660
661 /*
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700662 * If %false, this subsystem is properly hierarchical -
663 * configuration, resource accounting and restriction on a parent
664 * cgroup cover those of its children. If %true, hierarchy support
665 * is broken in some ways - some subsystems ignore hierarchy
666 * completely while others are only implemented half-way.
667 *
668 * It's now disallowed to create nested cgroups if the subsystem is
669 * broken and cgroup core will emit a warning message on such
670 * cases. Eventually, all subsystems will be made properly
671 * hierarchical and this will go away.
672 */
673 bool broken_hierarchy;
674 bool warned_broken_hierarchy;
675
Tejun Heo073219e2014-02-08 10:36:58 -0500676 /* the following two fields are initialized automtically during boot */
Tejun Heoaec25022014-02-08 10:36:58 -0500677 int id;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700678#define MAX_CGROUP_TYPE_NAMELEN 32
679 const char *name;
680
Tejun Heob85d2042013-12-06 15:11:57 -0500681 /* link to parent, protected by cgroup_lock() */
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400682 struct cgroup_root *root;
Ben Blume6a11052010-03-10 15:22:09 -0800683
Tejun Heo15a4c832014-05-04 15:09:14 -0400684 /* idr for css->id */
685 struct idr css_idr;
686
Tejun Heo0adb0702014-02-12 09:29:48 -0500687 /*
688 * List of cftypes. Each entry is the first entry of an array
689 * terminated by zero length name.
690 */
691 struct list_head cfts;
Tejun Heo8e3f6542012-04-01 12:09:55 -0700692
Tejun Heoa8ddc822014-07-15 11:05:10 -0400693 /*
694 * Base cftypes which are automatically registered. The two can
695 * point to the same array.
696 */
697 struct cftype *dfl_cftypes; /* for the default hierarchy */
698 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
Tejun Heoaf0ba672014-07-08 18:02:57 -0400699
700 /*
701 * A subsystem may depend on other subsystems. When such subsystem
702 * is enabled on a cgroup, the depended-upon subsystems are enabled
703 * together if available. Subsystems enabled due to dependency are
704 * not visible to userland until explicitly enabled. The following
705 * specifies the mask of subsystems that this one depends on.
706 */
707 unsigned int depends_on;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700708};
709
Tejun Heo073219e2014-02-08 10:36:58 -0500710#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700711#include <linux/cgroup_subsys.h>
712#undef SUBSYS
713
Tejun Heo8af01f52013-08-08 20:11:22 -0400714/**
Tejun Heo14611e52013-06-25 11:48:32 -0700715 * task_css_set_check - obtain a task's css_set with extra access conditions
716 * @task: the task to obtain css_set for
717 * @__c: extra condition expression to be passed to rcu_dereference_check()
718 *
719 * A task's css_set is RCU protected, initialized and exited while holding
720 * task_lock(), and can only be modified while holding both cgroup_mutex
721 * and task_lock() while the task is alive. This macro verifies that the
722 * caller is inside proper critical section and returns @task's css_set.
723 *
724 * The caller can also specify additional allowed conditions via @__c, such
725 * as locks used during the cgroup_subsys::attach() methods.
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200726 */
Tejun Heo22194492013-04-07 09:29:51 -0700727#ifdef CONFIG_PROVE_RCU
728extern struct mutex cgroup_mutex;
Tejun Heo0e1d7682014-02-25 10:04:03 -0500729extern struct rw_semaphore css_set_rwsem;
Tejun Heo14611e52013-06-25 11:48:32 -0700730#define task_css_set_check(task, __c) \
731 rcu_dereference_check((task)->cgroups, \
Tejun Heo0e1d7682014-02-25 10:04:03 -0500732 lockdep_is_held(&cgroup_mutex) || \
733 lockdep_is_held(&css_set_rwsem) || \
734 ((task)->flags & PF_EXITING) || (__c))
Tejun Heo22194492013-04-07 09:29:51 -0700735#else
Tejun Heo14611e52013-06-25 11:48:32 -0700736#define task_css_set_check(task, __c) \
737 rcu_dereference((task)->cgroups)
Tejun Heo22194492013-04-07 09:29:51 -0700738#endif
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200739
Tejun Heo14611e52013-06-25 11:48:32 -0700740/**
Tejun Heo8af01f52013-08-08 20:11:22 -0400741 * task_css_check - obtain css for (task, subsys) w/ extra access conds
Tejun Heo14611e52013-06-25 11:48:32 -0700742 * @task: the target task
743 * @subsys_id: the target subsystem ID
744 * @__c: extra condition expression to be passed to rcu_dereference_check()
745 *
746 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
747 * synchronization rules are the same as task_css_set_check().
748 */
Tejun Heo8af01f52013-08-08 20:11:22 -0400749#define task_css_check(task, subsys_id, __c) \
Tejun Heo14611e52013-06-25 11:48:32 -0700750 task_css_set_check((task), (__c))->subsys[(subsys_id)]
751
752/**
753 * task_css_set - obtain a task's css_set
754 * @task: the task to obtain css_set for
755 *
756 * See task_css_set_check().
757 */
758static inline struct css_set *task_css_set(struct task_struct *task)
759{
760 return task_css_set_check(task, false);
761}
762
763/**
Tejun Heo8af01f52013-08-08 20:11:22 -0400764 * task_css - obtain css for (task, subsys)
Tejun Heo14611e52013-06-25 11:48:32 -0700765 * @task: the target task
766 * @subsys_id: the target subsystem ID
767 *
Tejun Heo8af01f52013-08-08 20:11:22 -0400768 * See task_css_check().
Tejun Heo14611e52013-06-25 11:48:32 -0700769 */
Tejun Heo8af01f52013-08-08 20:11:22 -0400770static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
771 int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700772{
Tejun Heo8af01f52013-08-08 20:11:22 -0400773 return task_css_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700774}
775
Tejun Heo5024ae22014-05-07 21:31:17 -0400776/**
777 * task_css_is_root - test whether a task belongs to the root css
778 * @task: the target task
779 * @subsys_id: the target subsystem ID
780 *
781 * Test whether @task belongs to the root css on the specified subsystem.
782 * May be invoked in any context.
783 */
784static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
785{
786 return task_css_check(task, subsys_id, true) ==
787 init_css_set.subsys[subsys_id];
788}
789
Tejun Heo8af01f52013-08-08 20:11:22 -0400790static inline struct cgroup *task_cgroup(struct task_struct *task,
791 int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700792{
Tejun Heo8af01f52013-08-08 20:11:22 -0400793 return task_css(task, subsys_id)->cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700794}
795
Tejun Heo492eb212013-08-08 20:11:25 -0400796struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
797 struct cgroup_subsys_state *parent);
Tejun Heo53fa5262013-05-24 10:55:38 +0900798
Li Zefan1cb650b2013-08-19 10:05:24 +0800799struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
800
Tejun Heo574bd9f2012-11-09 09:12:29 -0800801/**
Tejun Heo492eb212013-08-08 20:11:25 -0400802 * css_for_each_child - iterate through children of a css
803 * @pos: the css * to use as the loop cursor
804 * @parent: css whose children to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800805 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400806 * Walk @parent's children. Must be called under rcu_read_lock().
Tejun Heo574bd9f2012-11-09 09:12:29 -0800807 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400808 * If a subsystem synchronizes ->css_online() and the start of iteration, a
809 * css which finished ->css_online() is guaranteed to be visible in the
810 * future iterations and will stay visible until the last reference is put.
811 * A css which hasn't finished ->css_online() or already finished
812 * ->css_offline() may show up during traversal. It's each subsystem's
813 * responsibility to synchronize against on/offlining.
Tejun Heo75501a62013-05-24 10:55:38 +0900814 *
815 * It is allowed to temporarily drop RCU read lock during iteration. The
816 * caller is responsible for ensuring that @pos remains accessible until
817 * the start of the next iteration by, for example, bumping the css refcnt.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800818 */
Tejun Heo492eb212013-08-08 20:11:25 -0400819#define css_for_each_child(pos, parent) \
820 for ((pos) = css_next_child(NULL, (parent)); (pos); \
821 (pos) = css_next_child((pos), (parent)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800822
Tejun Heo492eb212013-08-08 20:11:25 -0400823struct cgroup_subsys_state *
824css_next_descendant_pre(struct cgroup_subsys_state *pos,
825 struct cgroup_subsys_state *css);
826
827struct cgroup_subsys_state *
828css_rightmost_descendant(struct cgroup_subsys_state *pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -0800829
830/**
Tejun Heo492eb212013-08-08 20:11:25 -0400831 * css_for_each_descendant_pre - pre-order walk of a css's descendants
832 * @pos: the css * to use as the loop cursor
833 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800834 *
Tejun Heobd8815a2013-08-08 20:11:27 -0400835 * Walk @root's descendants. @root is included in the iteration and the
Tejun Heoc2931b72014-05-16 13:22:51 -0400836 * first node to be visited. Must be called under rcu_read_lock().
Tejun Heo574bd9f2012-11-09 09:12:29 -0800837 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400838 * If a subsystem synchronizes ->css_online() and the start of iteration, a
839 * css which finished ->css_online() is guaranteed to be visible in the
840 * future iterations and will stay visible until the last reference is put.
841 * A css which hasn't finished ->css_online() or already finished
842 * ->css_offline() may show up during traversal. It's each subsystem's
843 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800844 *
Tejun Heoc2931b72014-05-16 13:22:51 -0400845 * For example, the following guarantees that a descendant can't escape
Tejun Heo574bd9f2012-11-09 09:12:29 -0800846 * state updates of its ancestors.
847 *
Tejun Heo492eb212013-08-08 20:11:25 -0400848 * my_online(@css)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800849 * {
Tejun Heo492eb212013-08-08 20:11:25 -0400850 * Lock @css's parent and @css;
851 * Inherit state from the parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -0800852 * Unlock both.
853 * }
854 *
Tejun Heo492eb212013-08-08 20:11:25 -0400855 * my_update_state(@css)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800856 * {
Tejun Heo492eb212013-08-08 20:11:25 -0400857 * css_for_each_descendant_pre(@pos, @css) {
Tejun Heo574bd9f2012-11-09 09:12:29 -0800858 * Lock @pos;
Tejun Heobd8815a2013-08-08 20:11:27 -0400859 * if (@pos == @css)
860 * Update @css's state;
861 * else
862 * Verify @pos is alive and inherit state from its parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -0800863 * Unlock @pos;
864 * }
865 * }
866 *
867 * As long as the inheriting step, including checking the parent state, is
868 * enclosed inside @pos locking, double-locking the parent isn't necessary
869 * while inheriting. The state update to the parent is guaranteed to be
870 * visible by walking order and, as long as inheriting operations to the
871 * same @pos are atomic to each other, multiple updates racing each other
872 * still result in the correct state. It's guaranateed that at least one
Tejun Heo492eb212013-08-08 20:11:25 -0400873 * inheritance happens for any css after the latest update to its parent.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800874 *
875 * If checking parent's state requires locking the parent, each inheriting
876 * iteration should lock and unlock both @pos->parent and @pos.
877 *
878 * Alternatively, a subsystem may choose to use a single global lock to
Tejun Heo92fb9742012-11-19 08:13:38 -0800879 * synchronize ->css_online() and ->css_offline() against tree-walking
Tejun Heo574bd9f2012-11-09 09:12:29 -0800880 * operations.
Tejun Heo75501a62013-05-24 10:55:38 +0900881 *
882 * It is allowed to temporarily drop RCU read lock during iteration. The
883 * caller is responsible for ensuring that @pos remains accessible until
884 * the start of the next iteration by, for example, bumping the css refcnt.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800885 */
Tejun Heo492eb212013-08-08 20:11:25 -0400886#define css_for_each_descendant_pre(pos, css) \
887 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
888 (pos) = css_next_descendant_pre((pos), (css)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800889
Tejun Heo492eb212013-08-08 20:11:25 -0400890struct cgroup_subsys_state *
891css_next_descendant_post(struct cgroup_subsys_state *pos,
892 struct cgroup_subsys_state *css);
Tejun Heo574bd9f2012-11-09 09:12:29 -0800893
894/**
Tejun Heo492eb212013-08-08 20:11:25 -0400895 * css_for_each_descendant_post - post-order walk of a css's descendants
896 * @pos: the css * to use as the loop cursor
897 * @css: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -0800898 *
Tejun Heo492eb212013-08-08 20:11:25 -0400899 * Similar to css_for_each_descendant_pre() but performs post-order
Tejun Heobd8815a2013-08-08 20:11:27 -0400900 * traversal instead. @root is included in the iteration and the last
Tejun Heoc2931b72014-05-16 13:22:51 -0400901 * node to be visited.
902 *
903 * If a subsystem synchronizes ->css_online() and the start of iteration, a
904 * css which finished ->css_online() is guaranteed to be visible in the
905 * future iterations and will stay visible until the last reference is put.
906 * A css which hasn't finished ->css_online() or already finished
907 * ->css_offline() may show up during traversal. It's each subsystem's
908 * responsibility to synchronize against on/offlining.
909 *
910 * Note that the walk visibility guarantee example described in pre-order
911 * walk doesn't apply the same to post-order walks.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800912 */
Tejun Heo492eb212013-08-08 20:11:25 -0400913#define css_for_each_descendant_post(pos, css) \
914 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
915 (pos) = css_next_descendant_post((pos), (css)))
Tejun Heo574bd9f2012-11-09 09:12:29 -0800916
Tejun Heof3d46502014-05-16 13:22:52 -0400917bool css_has_online_children(struct cgroup_subsys_state *css);
918
Tejun Heo72ec7022013-08-08 20:11:26 -0400919/* A css_task_iter should be treated as an opaque object */
920struct css_task_iter {
Tejun Heo3ebb2b62014-04-23 11:13:15 -0400921 struct cgroup_subsys *ss;
922
Tejun Heo0f0a2b42014-04-23 11:13:15 -0400923 struct list_head *cset_pos;
924 struct list_head *cset_head;
925
926 struct list_head *task_pos;
927 struct list_head *tasks_head;
928 struct list_head *mg_tasks_head;
Paul Menage817929e2007-10-18 23:39:36 -0700929};
930
Tejun Heo72ec7022013-08-08 20:11:26 -0400931void css_task_iter_start(struct cgroup_subsys_state *css,
932 struct css_task_iter *it);
933struct task_struct *css_task_iter_next(struct css_task_iter *it);
934void css_task_iter_end(struct css_task_iter *it);
Tejun Heoe5358372013-08-08 20:11:26 -0400935
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700936int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
Tejun Heo8cc99342013-04-07 09:29:50 -0700937int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700938
Tejun Heoeeecbd12014-11-18 02:49:52 -0500939struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
940 struct cgroup_subsys *ss);
Tejun Heoec903c02014-05-13 12:11:01 -0400941struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
942 struct cgroup_subsys *ss);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700943
Paul Menageddbcc7e2007-10-18 23:39:30 -0700944#else /* !CONFIG_CGROUPS */
945
946static inline int cgroup_init_early(void) { return 0; }
947static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700948static inline void cgroup_fork(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700949static inline void cgroup_post_fork(struct task_struct *p) {}
Li Zefan1ec41832014-03-28 15:22:19 +0800950static inline void cgroup_exit(struct task_struct *p) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700951
Balbir Singh846c7bb2007-10-18 23:39:44 -0700952static inline int cgroupstats_build(struct cgroupstats *stats,
953 struct dentry *dentry)
954{
955 return -EINVAL;
956}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700957
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200958/* No cgroups - nothing to do */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700959static inline int cgroup_attach_task_all(struct task_struct *from,
960 struct task_struct *t)
961{
962 return 0;
963}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200964
Paul Menageddbcc7e2007-10-18 23:39:30 -0700965#endif /* !CONFIG_CGROUPS */
966
Paul Menageddbcc7e2007-10-18 23:39:30 -0700967#endif /* _LINUX_CGROUP_H */