Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1 | #ifndef _LINUX_CGROUP_H |
| 2 | #define _LINUX_CGROUP_H |
| 3 | /* |
| 4 | * cgroup interface |
| 5 | * |
| 6 | * Copyright (C) 2003 BULL SA |
| 7 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. |
| 8 | * |
| 9 | */ |
| 10 | |
| 11 | #include <linux/sched.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 12 | #include <linux/cpumask.h> |
| 13 | #include <linux/nodemask.h> |
| 14 | #include <linux/rcupdate.h> |
Tejun Heo | eb6fd50 | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 15 | #include <linux/rculist.h> |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 16 | #include <linux/cgroupstats.h> |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 17 | #include <linux/prio_heap.h> |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 18 | #include <linux/rwsem.h> |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 19 | #include <linux/idr.h> |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 20 | #include <linux/workqueue.h> |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 21 | #include <linux/xattr.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 22 | |
| 23 | #ifdef CONFIG_CGROUPS |
| 24 | |
| 25 | struct cgroupfs_root; |
| 26 | struct cgroup_subsys; |
| 27 | struct inode; |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 28 | struct cgroup; |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 29 | struct css_id; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 30 | |
| 31 | extern int cgroup_init_early(void); |
| 32 | extern int cgroup_init(void); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 33 | extern void cgroup_lock(void); |
Paul E. McKenney | d11c563 | 2010-02-22 17:04:50 -0800 | [diff] [blame] | 34 | extern int cgroup_lock_is_held(void); |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 35 | extern bool cgroup_lock_live_group(struct cgroup *cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 36 | extern void cgroup_unlock(void); |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 37 | extern void cgroup_fork(struct task_struct *p); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 38 | extern void cgroup_post_fork(struct task_struct *p); |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 39 | extern void cgroup_exit(struct task_struct *p, int run_callbacks); |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 40 | extern int cgroupstats_build(struct cgroupstats *stats, |
| 41 | struct dentry *dentry); |
Ben Blum | e6a1105 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 42 | extern int cgroup_load_subsys(struct cgroup_subsys *ss); |
Ben Blum | cf5d594 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 43 | extern void cgroup_unload_subsys(struct cgroup_subsys *ss); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 44 | |
Alexey Dobriyan | 828c095 | 2009-10-01 15:43:56 -0700 | [diff] [blame] | 45 | extern const struct file_operations proc_cgroup_operations; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 46 | |
Li Zefan | 7d8e0bf | 2013-03-05 10:57:03 +0800 | [diff] [blame^] | 47 | /* |
| 48 | * Define the enumeration of all cgroup subsystems. |
| 49 | * |
| 50 | * We define ids for builtin subsystems and then modular ones. |
| 51 | */ |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 52 | #define SUBSYS(_x) _x ## _subsys_id, |
| 53 | enum cgroup_subsys_id { |
Li Zefan | 7d8e0bf | 2013-03-05 10:57:03 +0800 | [diff] [blame^] | 54 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 55 | #include <linux/cgroup_subsys.h> |
Li Zefan | 7d8e0bf | 2013-03-05 10:57:03 +0800 | [diff] [blame^] | 56 | #undef IS_SUBSYS_ENABLED |
| 57 | CGROUP_BUILTIN_SUBSYS_COUNT, |
| 58 | |
| 59 | __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1, |
| 60 | |
| 61 | #define IS_SUBSYS_ENABLED(option) IS_MODULE(option) |
| 62 | #include <linux/cgroup_subsys.h> |
| 63 | #undef IS_SUBSYS_ENABLED |
Daniel Wagner | a6f0029 | 2012-09-12 16:12:08 +0200 | [diff] [blame] | 64 | CGROUP_SUBSYS_COUNT, |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 65 | }; |
| 66 | #undef SUBSYS |
| 67 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 68 | /* Per-subsystem/per-cgroup state maintained by the system. */ |
| 69 | struct cgroup_subsys_state { |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 70 | /* |
| 71 | * The cgroup that this subsystem is attached to. Useful |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 72 | * for subsystems that want to know about the cgroup |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 73 | * hierarchy structure |
| 74 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 75 | struct cgroup *cgroup; |
| 76 | |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 77 | /* |
| 78 | * State maintained by the cgroup system to allow subsystems |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 79 | * to be "busy". Should be accessed via css_get(), |
Tao Ma | d0b2fdd | 2012-11-20 22:06:18 +0800 | [diff] [blame] | 80 | * css_tryget() and css_put(). |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 81 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 82 | |
| 83 | atomic_t refcnt; |
| 84 | |
| 85 | unsigned long flags; |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 86 | /* ID for this css, if possible */ |
Arnd Bergmann | 2c392b8 | 2010-02-24 19:41:39 +0100 | [diff] [blame] | 87 | struct css_id __rcu *id; |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 88 | |
| 89 | /* Used to put @cgroup->dentry on the last css_put() */ |
| 90 | struct work_struct dput_work; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 91 | }; |
| 92 | |
| 93 | /* bits in struct cgroup_subsys_state flags field */ |
| 94 | enum { |
Tejun Heo | 38b53ab | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 95 | CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */ |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 96 | CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 97 | }; |
| 98 | |
Daisuke Nishimura | d7b9fff | 2010-03-10 15:22:05 -0800 | [diff] [blame] | 99 | /* Caller must verify that the css is not for root cgroup */ |
| 100 | static inline void __css_get(struct cgroup_subsys_state *css, int count) |
| 101 | { |
| 102 | atomic_add(count, &css->refcnt); |
| 103 | } |
| 104 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 105 | /* |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 106 | * Call css_get() to hold a reference on the css; it can be used |
| 107 | * for a reference obtained via: |
| 108 | * - an existing ref-counted reference to the css |
| 109 | * - task->cgroups for a locked task |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 110 | */ |
| 111 | |
| 112 | static inline void css_get(struct cgroup_subsys_state *css) |
| 113 | { |
| 114 | /* We don't need to reference count the root state */ |
Tejun Heo | 38b53ab | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 115 | if (!(css->flags & CSS_ROOT)) |
Daisuke Nishimura | d7b9fff | 2010-03-10 15:22:05 -0800 | [diff] [blame] | 116 | __css_get(css, 1); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 117 | } |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 118 | |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 119 | /* |
| 120 | * Call css_tryget() to take a reference on a css if your existing |
| 121 | * (known-valid) reference isn't already ref-counted. Returns false if |
| 122 | * the css has been destroyed. |
| 123 | */ |
| 124 | |
Tejun Heo | 28b4c27 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 125 | extern bool __css_tryget(struct cgroup_subsys_state *css); |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 126 | static inline bool css_tryget(struct cgroup_subsys_state *css) |
| 127 | { |
Tejun Heo | 38b53ab | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 128 | if (css->flags & CSS_ROOT) |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 129 | return true; |
Tejun Heo | 28b4c27 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 130 | return __css_tryget(css); |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 131 | } |
| 132 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 133 | /* |
| 134 | * css_put() should be called to release a reference taken by |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 135 | * css_get() or css_tryget() |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 136 | */ |
| 137 | |
Tejun Heo | 28b4c27 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 138 | extern void __css_put(struct cgroup_subsys_state *css); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 139 | static inline void css_put(struct cgroup_subsys_state *css) |
| 140 | { |
Tejun Heo | 38b53ab | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 141 | if (!(css->flags & CSS_ROOT)) |
Tejun Heo | 28b4c27 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 142 | __css_put(css); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 143 | } |
| 144 | |
Paul Menage | 3116f0e | 2008-04-29 01:00:04 -0700 | [diff] [blame] | 145 | /* bits in struct cgroup flags field */ |
| 146 | enum { |
| 147 | /* Control Group is dead */ |
| 148 | CGRP_REMOVED, |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 149 | /* |
| 150 | * Control Group has previously had a child cgroup or a task, |
| 151 | * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) |
| 152 | */ |
Paul Menage | 3116f0e | 2008-04-29 01:00:04 -0700 | [diff] [blame] | 153 | CGRP_RELEASABLE, |
| 154 | /* Control Group requires release notifications to userspace */ |
| 155 | CGRP_NOTIFY_ON_RELEASE, |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 156 | /* |
Tejun Heo | 2260e7f | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 157 | * Clone the parent's configuration when creating a new child |
| 158 | * cpuset cgroup. For historical reasons, this option can be |
| 159 | * specified at mount time and thus is implemented here. |
Daniel Lezcano | 97978e6 | 2010-10-27 15:33:35 -0700 | [diff] [blame] | 160 | */ |
Tejun Heo | 2260e7f | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 161 | CGRP_CPUSET_CLONE_CHILDREN, |
Paul Menage | 3116f0e | 2008-04-29 01:00:04 -0700 | [diff] [blame] | 162 | }; |
| 163 | |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 164 | struct cgroup_name { |
| 165 | struct rcu_head rcu_head; |
| 166 | char name[]; |
| 167 | }; |
| 168 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 169 | struct cgroup { |
| 170 | unsigned long flags; /* "unsigned long" so bitops work */ |
| 171 | |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 172 | /* |
| 173 | * count users of this cgroup. >0 means busy, but doesn't |
| 174 | * necessarily indicate the number of tasks in the cgroup |
| 175 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 176 | atomic_t count; |
| 177 | |
Tejun Heo | 0a950f6 | 2012-11-19 09:02:12 -0800 | [diff] [blame] | 178 | int id; /* ida allocated in-hierarchy ID */ |
| 179 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 180 | /* |
| 181 | * We link our 'sibling' struct into our parent's 'children'. |
| 182 | * Our children link their 'sibling' into our 'children'. |
| 183 | */ |
| 184 | struct list_head sibling; /* my parent's children */ |
| 185 | struct list_head children; /* my children */ |
Tejun Heo | 05ef1d7 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 186 | struct list_head files; /* my files */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 187 | |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 188 | struct cgroup *parent; /* my parent */ |
Tejun Heo | febfcef | 2012-11-19 08:13:36 -0800 | [diff] [blame] | 189 | struct dentry *dentry; /* cgroup fs entry, RCU protected */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 190 | |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 191 | /* |
| 192 | * This is a copy of dentry->d_name, and it's needed because |
| 193 | * we can't use dentry->d_name in cgroup_path(). |
| 194 | * |
| 195 | * You must acquire rcu_read_lock() to access cgrp->name, and |
| 196 | * the only place that can change it is rename(), which is |
| 197 | * protected by parent dir's i_mutex. |
| 198 | * |
| 199 | * Normally you should use cgroup_name() wrapper rather than |
| 200 | * access it directly. |
| 201 | */ |
| 202 | struct cgroup_name __rcu *name; |
| 203 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 204 | /* Private pointers for each registered subsystem */ |
| 205 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
| 206 | |
| 207 | struct cgroupfs_root *root; |
| 208 | struct cgroup *top_cgroup; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 209 | |
| 210 | /* |
| 211 | * List of cg_cgroup_links pointing at css_sets with |
| 212 | * tasks in this cgroup. Protected by css_set_lock |
| 213 | */ |
| 214 | struct list_head css_sets; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 215 | |
Tejun Heo | b0ca5a8 | 2012-04-01 12:09:54 -0700 | [diff] [blame] | 216 | struct list_head allcg_node; /* cgroupfs_root->allcg_list */ |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 217 | struct list_head cft_q_node; /* used during cftype add/rm */ |
Tejun Heo | b0ca5a8 | 2012-04-01 12:09:54 -0700 | [diff] [blame] | 218 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 219 | /* |
| 220 | * Linked list running through all cgroups that can |
| 221 | * potentially be reaped by the release agent. Protected by |
| 222 | * release_list_lock |
| 223 | */ |
| 224 | struct list_head release_list; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 225 | |
Ben Blum | 72a8cb3 | 2009-09-23 15:56:27 -0700 | [diff] [blame] | 226 | /* |
| 227 | * list of pidlists, up to two for each namespace (one for procs, one |
| 228 | * for tasks); created on demand. |
| 229 | */ |
| 230 | struct list_head pidlists; |
| 231 | struct mutex pidlist_mutex; |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 232 | |
| 233 | /* For RCU-protected deletion */ |
| 234 | struct rcu_head rcu_head; |
Li Zefan | be44562 | 2013-01-24 14:31:42 +0800 | [diff] [blame] | 235 | struct work_struct free_work; |
Kirill A. Shutemov | 0dea116 | 2010-03-10 15:22:20 -0800 | [diff] [blame] | 236 | |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 237 | /* List of events which userspace want to receive */ |
Kirill A. Shutemov | 0dea116 | 2010-03-10 15:22:20 -0800 | [diff] [blame] | 238 | struct list_head event_list; |
| 239 | spinlock_t event_list_lock; |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 240 | |
| 241 | /* directory xattrs */ |
| 242 | struct simple_xattrs xattrs; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 243 | }; |
| 244 | |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 245 | /* |
| 246 | * A css_set is a structure holding pointers to a set of |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 247 | * cgroup_subsys_state objects. This saves space in the task struct |
| 248 | * object and speeds up fork()/exit(), since a single inc/dec and a |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 249 | * list_add()/del() can bump the reference count on the entire cgroup |
| 250 | * set for a task. |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 251 | */ |
| 252 | |
| 253 | struct css_set { |
| 254 | |
| 255 | /* Reference count */ |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 256 | atomic_t refcount; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 257 | |
| 258 | /* |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 259 | * List running through all cgroup groups in the same hash |
| 260 | * slot. Protected by css_set_lock |
| 261 | */ |
| 262 | struct hlist_node hlist; |
| 263 | |
| 264 | /* |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 265 | * List running through all tasks using this cgroup |
| 266 | * group. Protected by css_set_lock |
| 267 | */ |
| 268 | struct list_head tasks; |
| 269 | |
| 270 | /* |
| 271 | * List of cg_cgroup_link objects on link chains from |
| 272 | * cgroups referenced from this css_set. Protected by |
| 273 | * css_set_lock |
| 274 | */ |
| 275 | struct list_head cg_links; |
| 276 | |
| 277 | /* |
| 278 | * Set of subsystem states, one for each subsystem. This array |
| 279 | * is immutable after creation apart from the init_css_set |
Ben Blum | cf5d594 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 280 | * during subsystem registration (at boot time) and modular subsystem |
| 281 | * loading/unloading. |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 282 | */ |
| 283 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
Ben Blum | c378369 | 2009-09-23 15:56:29 -0700 | [diff] [blame] | 284 | |
| 285 | /* For RCU-protected deletion */ |
| 286 | struct rcu_head rcu_head; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 287 | }; |
| 288 | |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 289 | /* |
| 290 | * cgroup_map_cb is an abstract callback API for reporting map-valued |
| 291 | * control files |
| 292 | */ |
| 293 | |
| 294 | struct cgroup_map_cb { |
| 295 | int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value); |
| 296 | void *state; |
| 297 | }; |
| 298 | |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 299 | /* |
| 300 | * struct cftype: handler definitions for cgroup control files |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 301 | * |
| 302 | * When reading/writing to a file: |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 303 | * - the cgroup to use is file->f_dentry->d_parent->d_fsdata |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 304 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
| 305 | */ |
| 306 | |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 307 | /* cftype->flags */ |
| 308 | #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */ |
Tao Ma | d0b2fdd | 2012-11-20 22:06:18 +0800 | [diff] [blame] | 309 | #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */ |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 310 | |
| 311 | #define MAX_CFTYPE_NAME 64 |
| 312 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 313 | struct cftype { |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 314 | /* |
| 315 | * By convention, the name should begin with the name of the |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 316 | * subsystem, followed by a period. Zero length string indicates |
| 317 | * end of cftype array. |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 318 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 319 | char name[MAX_CFTYPE_NAME]; |
| 320 | int private; |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 321 | /* |
| 322 | * If not 0, file mode is set to this value, otherwise it will |
| 323 | * be figured out automatically |
| 324 | */ |
Al Viro | a5e7ed3 | 2011-07-26 01:55:55 -0400 | [diff] [blame] | 325 | umode_t mode; |
Paul Menage | db3b149 | 2008-07-25 01:46:58 -0700 | [diff] [blame] | 326 | |
| 327 | /* |
| 328 | * If non-zero, defines the maximum length of string that can |
| 329 | * be passed to write_string; defaults to 64 |
| 330 | */ |
| 331 | size_t max_write_len; |
| 332 | |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 333 | /* CFTYPE_* flags */ |
| 334 | unsigned int flags; |
| 335 | |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 336 | /* file xattrs */ |
| 337 | struct simple_xattrs xattrs; |
| 338 | |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 339 | int (*open)(struct inode *inode, struct file *file); |
| 340 | ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft, |
| 341 | struct file *file, |
| 342 | char __user *buf, size_t nbytes, loff_t *ppos); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 343 | /* |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 344 | * read_u64() is a shortcut for the common case of returning a |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 345 | * single integer. Use it in place of read() |
| 346 | */ |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 347 | u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft); |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 348 | /* |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 349 | * read_s64() is a signed version of read_u64() |
| 350 | */ |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 351 | s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft); |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 352 | /* |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 353 | * read_map() is used for defining a map of key/value |
| 354 | * pairs. It should call cb->fill(cb, key, value) for each |
| 355 | * entry. The key/value pairs (and their ordering) should not |
| 356 | * change between reboots. |
| 357 | */ |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 358 | int (*read_map)(struct cgroup *cont, struct cftype *cft, |
| 359 | struct cgroup_map_cb *cb); |
Serge E. Hallyn | 29486df | 2008-04-29 01:00:14 -0700 | [diff] [blame] | 360 | /* |
| 361 | * read_seq_string() is used for outputting a simple sequence |
| 362 | * using seqfile. |
| 363 | */ |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 364 | int (*read_seq_string)(struct cgroup *cont, struct cftype *cft, |
| 365 | struct seq_file *m); |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 366 | |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 367 | ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft, |
| 368 | struct file *file, |
| 369 | const char __user *buf, size_t nbytes, loff_t *ppos); |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 370 | |
| 371 | /* |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 372 | * write_u64() is a shortcut for the common case of accepting |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 373 | * a single integer (as parsed by simple_strtoull) from |
| 374 | * userspace. Use in place of write(); return 0 or error. |
| 375 | */ |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 376 | int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val); |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 377 | /* |
| 378 | * write_s64() is a signed version of write_u64() |
| 379 | */ |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 380 | int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val); |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 381 | |
Pavel Emelyanov | d447ea2 | 2008-04-29 01:00:08 -0700 | [diff] [blame] | 382 | /* |
Paul Menage | db3b149 | 2008-07-25 01:46:58 -0700 | [diff] [blame] | 383 | * write_string() is passed a nul-terminated kernelspace |
| 384 | * buffer of maximum length determined by max_write_len. |
| 385 | * Returns 0 or -ve error code. |
| 386 | */ |
| 387 | int (*write_string)(struct cgroup *cgrp, struct cftype *cft, |
| 388 | const char *buffer); |
| 389 | /* |
Pavel Emelyanov | d447ea2 | 2008-04-29 01:00:08 -0700 | [diff] [blame] | 390 | * trigger() callback can be used to get some kick from the |
| 391 | * userspace, when the actual string written is not important |
| 392 | * at all. The private field can be used to determine the |
| 393 | * kick type for multiplexing. |
| 394 | */ |
| 395 | int (*trigger)(struct cgroup *cgrp, unsigned int event); |
| 396 | |
Paul Menage | ce16b49 | 2008-07-25 01:46:57 -0700 | [diff] [blame] | 397 | int (*release)(struct inode *inode, struct file *file); |
Kirill A. Shutemov | 0dea116 | 2010-03-10 15:22:20 -0800 | [diff] [blame] | 398 | |
| 399 | /* |
| 400 | * register_event() callback will be used to add new userspace |
| 401 | * waiter for changes related to the cftype. Implement it if |
| 402 | * you want to provide this functionality. Use eventfd_signal() |
| 403 | * on eventfd to send notification to userspace. |
| 404 | */ |
| 405 | int (*register_event)(struct cgroup *cgrp, struct cftype *cft, |
| 406 | struct eventfd_ctx *eventfd, const char *args); |
| 407 | /* |
| 408 | * unregister_event() callback will be called when userspace |
| 409 | * closes the eventfd or on cgroup removing. |
| 410 | * This callback must be implemented, if you want provide |
| 411 | * notification functionality. |
Kirill A. Shutemov | 0dea116 | 2010-03-10 15:22:20 -0800 | [diff] [blame] | 412 | */ |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 413 | void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft, |
Kirill A. Shutemov | 0dea116 | 2010-03-10 15:22:20 -0800 | [diff] [blame] | 414 | struct eventfd_ctx *eventfd); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 415 | }; |
| 416 | |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 417 | /* |
| 418 | * cftype_sets describe cftypes belonging to a subsystem and are chained at |
| 419 | * cgroup_subsys->cftsets. Each cftset points to an array of cftypes |
| 420 | * terminated by zero length name. |
| 421 | */ |
| 422 | struct cftype_set { |
| 423 | struct list_head node; /* chained at subsys->cftsets */ |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 424 | struct cftype *cfts; |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 425 | }; |
| 426 | |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 427 | struct cgroup_scanner { |
| 428 | struct cgroup *cg; |
| 429 | int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan); |
| 430 | void (*process_task)(struct task_struct *p, |
| 431 | struct cgroup_scanner *scan); |
| 432 | struct ptr_heap *heap; |
Li Zefan | bd1a8ab | 2009-04-02 16:57:50 -0700 | [diff] [blame] | 433 | void *data; |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 434 | }; |
| 435 | |
Li Zefan | 65dff75 | 2013-03-01 15:01:56 +0800 | [diff] [blame] | 436 | /* Caller should hold rcu_read_lock() */ |
| 437 | static inline const char *cgroup_name(const struct cgroup *cgrp) |
| 438 | { |
| 439 | return rcu_dereference(cgrp->name)->name; |
| 440 | } |
| 441 | |
Aristeu Rozanski | 03b1cde | 2012-08-23 16:53:30 -0400 | [diff] [blame] | 442 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
| 443 | int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 444 | |
Li Zefan | ffd2d88 | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 445 | int cgroup_is_removed(const struct cgroup *cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 446 | |
Li Zefan | ffd2d88 | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 447 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 448 | |
Li Zefan | ffd2d88 | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 449 | int cgroup_task_count(const struct cgroup *cgrp); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 450 | |
Grzegorz Nosek | 313e924 | 2009-04-02 16:57:23 -0700 | [diff] [blame] | 451 | /* Return true if cgrp is a descendant of the task's cgroup */ |
| 452 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 453 | |
Thadeu Lima de Souza Cascardo | 21acb9c | 2009-02-04 10:12:08 +0100 | [diff] [blame] | 454 | /* |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 455 | * Control Group taskset, used to pass around set of tasks to cgroup_subsys |
| 456 | * methods. |
| 457 | */ |
| 458 | struct cgroup_taskset; |
| 459 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset); |
| 460 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset); |
| 461 | struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset); |
| 462 | int cgroup_taskset_size(struct cgroup_taskset *tset); |
| 463 | |
| 464 | /** |
| 465 | * cgroup_taskset_for_each - iterate cgroup_taskset |
| 466 | * @task: the loop cursor |
| 467 | * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all |
| 468 | * @tset: taskset to iterate |
| 469 | */ |
| 470 | #define cgroup_taskset_for_each(task, skip_cgrp, tset) \ |
| 471 | for ((task) = cgroup_taskset_first((tset)); (task); \ |
| 472 | (task) = cgroup_taskset_next((tset))) \ |
| 473 | if (!(skip_cgrp) || \ |
| 474 | cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp)) |
| 475 | |
| 476 | /* |
Thadeu Lima de Souza Cascardo | 21acb9c | 2009-02-04 10:12:08 +0100 | [diff] [blame] | 477 | * Control Group subsystem type. |
| 478 | * See Documentation/cgroups/cgroups.txt for details |
| 479 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 480 | |
| 481 | struct cgroup_subsys { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 482 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp); |
| 483 | int (*css_online)(struct cgroup *cgrp); |
| 484 | void (*css_offline)(struct cgroup *cgrp); |
| 485 | void (*css_free)(struct cgroup *cgrp); |
| 486 | |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 487 | int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); |
| 488 | void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); |
| 489 | void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); |
| 490 | void (*fork)(struct task_struct *task); |
| 491 | void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp, |
| 492 | struct task_struct *task); |
Li Zefan | 761b3ef5 | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 493 | void (*bind)(struct cgroup *root); |
Hugh Dickins | e599137 | 2009-01-06 14:39:22 -0800 | [diff] [blame] | 494 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 495 | int subsys_id; |
| 496 | int active; |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 497 | int disabled; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 498 | int early_init; |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 499 | /* |
| 500 | * True if this subsys uses ID. ID is not available before cgroup_init() |
| 501 | * (not available in early_init time.) |
| 502 | */ |
| 503 | bool use_id; |
Tejun Heo | 48ddbe1 | 2012-04-01 12:09:56 -0700 | [diff] [blame] | 504 | |
| 505 | /* |
Tejun Heo | 8c7f6ed | 2012-09-13 12:20:58 -0700 | [diff] [blame] | 506 | * If %false, this subsystem is properly hierarchical - |
| 507 | * configuration, resource accounting and restriction on a parent |
| 508 | * cgroup cover those of its children. If %true, hierarchy support |
| 509 | * is broken in some ways - some subsystems ignore hierarchy |
| 510 | * completely while others are only implemented half-way. |
| 511 | * |
| 512 | * It's now disallowed to create nested cgroups if the subsystem is |
| 513 | * broken and cgroup core will emit a warning message on such |
| 514 | * cases. Eventually, all subsystems will be made properly |
| 515 | * hierarchical and this will go away. |
| 516 | */ |
| 517 | bool broken_hierarchy; |
| 518 | bool warned_broken_hierarchy; |
| 519 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 520 | #define MAX_CGROUP_TYPE_NAMELEN 32 |
| 521 | const char *name; |
| 522 | |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 523 | /* |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 524 | * Link to parent, and list entry in parent's children. |
Li Zefan | 6be96a5 | 2012-06-06 19:12:30 -0700 | [diff] [blame] | 525 | * Protected by cgroup_lock() |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 526 | */ |
| 527 | struct cgroupfs_root *root; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 528 | struct list_head sibling; |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 529 | /* used when use_id == true */ |
| 530 | struct idr idr; |
Hugh Dickins | 42aee6c | 2012-03-21 16:34:21 -0700 | [diff] [blame] | 531 | spinlock_t id_lock; |
Ben Blum | e6a1105 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 532 | |
Tejun Heo | 8e3f654 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 533 | /* list of cftype_sets */ |
| 534 | struct list_head cftsets; |
| 535 | |
| 536 | /* base cftypes, automatically [de]registered with subsys itself */ |
| 537 | struct cftype *base_cftypes; |
| 538 | struct cftype_set base_cftset; |
| 539 | |
Ben Blum | e6a1105 | 2010-03-10 15:22:09 -0800 | [diff] [blame] | 540 | /* should be defined only by modular subsystems */ |
| 541 | struct module *module; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 542 | }; |
| 543 | |
| 544 | #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys; |
Daniel Wagner | 5fc0b02 | 2012-09-12 16:12:05 +0200 | [diff] [blame] | 545 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 546 | #include <linux/cgroup_subsys.h> |
Daniel Wagner | 5fc0b02 | 2012-09-12 16:12:05 +0200 | [diff] [blame] | 547 | #undef IS_SUBSYS_ENABLED |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 548 | #undef SUBSYS |
| 549 | |
| 550 | static inline struct cgroup_subsys_state *cgroup_subsys_state( |
Li Zefan | ffd2d88 | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 551 | struct cgroup *cgrp, int subsys_id) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 552 | { |
Li Zefan | ffd2d88 | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 553 | return cgrp->subsys[subsys_id]; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 554 | } |
| 555 | |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 556 | /* |
| 557 | * function to get the cgroup_subsys_state which allows for extra |
| 558 | * rcu_dereference_check() conditions, such as locks used during the |
| 559 | * cgroup_subsys::attach() methods. |
| 560 | */ |
| 561 | #define task_subsys_state_check(task, subsys_id, __c) \ |
| 562 | rcu_dereference_check(task->cgroups->subsys[subsys_id], \ |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 563 | lockdep_is_held(&task->alloc_lock) || \ |
| 564 | cgroup_lock_is_held() || (__c)) |
| 565 | |
| 566 | static inline struct cgroup_subsys_state * |
| 567 | task_subsys_state(struct task_struct *task, int subsys_id) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 568 | { |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 569 | return task_subsys_state_check(task, subsys_id, false); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 570 | } |
| 571 | |
| 572 | static inline struct cgroup* task_cgroup(struct task_struct *task, |
| 573 | int subsys_id) |
| 574 | { |
| 575 | return task_subsys_state(task, subsys_id)->cgroup; |
| 576 | } |
| 577 | |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 578 | /** |
| 579 | * cgroup_for_each_child - iterate through children of a cgroup |
| 580 | * @pos: the cgroup * to use as the loop cursor |
| 581 | * @cgroup: cgroup whose children to walk |
| 582 | * |
| 583 | * Walk @cgroup's children. Must be called under rcu_read_lock(). A child |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 584 | * cgroup which hasn't finished ->css_online() or already has finished |
| 585 | * ->css_offline() may show up during traversal and it's each subsystem's |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 586 | * responsibility to verify that each @pos is alive. |
| 587 | * |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 588 | * If a subsystem synchronizes against the parent in its ->css_online() and |
| 589 | * before starting iterating, a cgroup which finished ->css_online() is |
| 590 | * guaranteed to be visible in the future iterations. |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 591 | */ |
| 592 | #define cgroup_for_each_child(pos, cgroup) \ |
| 593 | list_for_each_entry_rcu(pos, &(cgroup)->children, sibling) |
| 594 | |
| 595 | struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos, |
| 596 | struct cgroup *cgroup); |
Tejun Heo | 12a9d2f | 2013-01-07 08:49:33 -0800 | [diff] [blame] | 597 | struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos); |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 598 | |
| 599 | /** |
| 600 | * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants |
| 601 | * @pos: the cgroup * to use as the loop cursor |
| 602 | * @cgroup: cgroup whose descendants to walk |
| 603 | * |
| 604 | * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 605 | * descendant cgroup which hasn't finished ->css_online() or already has |
| 606 | * finished ->css_offline() may show up during traversal and it's each |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 607 | * subsystem's responsibility to verify that each @pos is alive. |
| 608 | * |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 609 | * If a subsystem synchronizes against the parent in its ->css_online() and |
| 610 | * before starting iterating, and synchronizes against @pos on each |
| 611 | * iteration, any descendant cgroup which finished ->css_offline() is |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 612 | * guaranteed to be visible in the future iterations. |
| 613 | * |
| 614 | * In other words, the following guarantees that a descendant can't escape |
| 615 | * state updates of its ancestors. |
| 616 | * |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 617 | * my_online(@cgrp) |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 618 | * { |
| 619 | * Lock @cgrp->parent and @cgrp; |
| 620 | * Inherit state from @cgrp->parent; |
| 621 | * Unlock both. |
| 622 | * } |
| 623 | * |
| 624 | * my_update_state(@cgrp) |
| 625 | * { |
| 626 | * Lock @cgrp; |
| 627 | * Update @cgrp's state; |
| 628 | * Unlock @cgrp; |
| 629 | * |
| 630 | * cgroup_for_each_descendant_pre(@pos, @cgrp) { |
| 631 | * Lock @pos; |
| 632 | * Verify @pos is alive and inherit state from @pos->parent; |
| 633 | * Unlock @pos; |
| 634 | * } |
| 635 | * } |
| 636 | * |
| 637 | * As long as the inheriting step, including checking the parent state, is |
| 638 | * enclosed inside @pos locking, double-locking the parent isn't necessary |
| 639 | * while inheriting. The state update to the parent is guaranteed to be |
| 640 | * visible by walking order and, as long as inheriting operations to the |
| 641 | * same @pos are atomic to each other, multiple updates racing each other |
| 642 | * still result in the correct state. It's guaranateed that at least one |
| 643 | * inheritance happens for any cgroup after the latest update to its |
| 644 | * parent. |
| 645 | * |
| 646 | * If checking parent's state requires locking the parent, each inheriting |
| 647 | * iteration should lock and unlock both @pos->parent and @pos. |
| 648 | * |
| 649 | * Alternatively, a subsystem may choose to use a single global lock to |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 650 | * synchronize ->css_online() and ->css_offline() against tree-walking |
Tejun Heo | 574bd9f | 2012-11-09 09:12:29 -0800 | [diff] [blame] | 651 | * operations. |
| 652 | */ |
| 653 | #define cgroup_for_each_descendant_pre(pos, cgroup) \ |
| 654 | for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \ |
| 655 | pos = cgroup_next_descendant_pre((pos), (cgroup))) |
| 656 | |
| 657 | struct cgroup *cgroup_next_descendant_post(struct cgroup *pos, |
| 658 | struct cgroup *cgroup); |
| 659 | |
| 660 | /** |
| 661 | * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants |
| 662 | * @pos: the cgroup * to use as the loop cursor |
| 663 | * @cgroup: cgroup whose descendants to walk |
| 664 | * |
| 665 | * Similar to cgroup_for_each_descendant_pre() but performs post-order |
| 666 | * traversal instead. Note that the walk visibility guarantee described in |
| 667 | * pre-order walk doesn't apply the same to post-order walks. |
| 668 | */ |
| 669 | #define cgroup_for_each_descendant_post(pos, cgroup) \ |
| 670 | for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \ |
| 671 | pos = cgroup_next_descendant_post((pos), (cgroup))) |
| 672 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 673 | /* A cgroup_iter should be treated as an opaque object */ |
| 674 | struct cgroup_iter { |
| 675 | struct list_head *cg_link; |
| 676 | struct list_head *task; |
| 677 | }; |
| 678 | |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 679 | /* |
| 680 | * To iterate across the tasks in a cgroup: |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 681 | * |
Uwe Kleine-König | b595076 | 2010-11-01 15:38:34 -0400 | [diff] [blame] | 682 | * 1) call cgroup_iter_start to initialize an iterator |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 683 | * |
| 684 | * 2) call cgroup_iter_next() to retrieve member tasks until it |
| 685 | * returns NULL or until you want to end the iteration |
| 686 | * |
| 687 | * 3) call cgroup_iter_end() to destroy the iterator. |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 688 | * |
Paul Menage | d20a390 | 2009-04-02 16:57:22 -0700 | [diff] [blame] | 689 | * Or, call cgroup_scan_tasks() to iterate through every task in a |
| 690 | * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling |
| 691 | * the test_task() callback, but not while calling the process_task() |
| 692 | * callback. |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 693 | */ |
Li Zefan | ffd2d88 | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 694 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it); |
| 695 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 696 | struct cgroup_iter *it); |
Li Zefan | ffd2d88 | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 697 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it); |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 698 | int cgroup_scan_tasks(struct cgroup_scanner *scan); |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 699 | int cgroup_attach_task(struct cgroup *, struct task_struct *); |
Michael S. Tsirkin | 31583bb | 2010-09-09 16:37:37 -0700 | [diff] [blame] | 700 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); |
| 701 | |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 702 | /* |
| 703 | * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works |
| 704 | * if cgroup_subsys.use_id == true. It can be used for looking up and scanning. |
| 705 | * CSS ID is assigned at cgroup allocation (create) automatically |
| 706 | * and removed when subsys calls free_css_id() function. This is because |
| 707 | * the lifetime of cgroup_subsys_state is subsys's matter. |
| 708 | * |
| 709 | * Looking up and scanning function should be called under rcu_read_lock(). |
Li Zefan | 6be96a5 | 2012-06-06 19:12:30 -0700 | [diff] [blame] | 710 | * Taking cgroup_mutex is not necessary for following calls. |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 711 | * But the css returned by this routine can be "not populated yet" or "being |
| 712 | * destroyed". The caller should check css and cgroup's status. |
| 713 | */ |
| 714 | |
| 715 | /* |
| 716 | * Typically Called at ->destroy(), or somewhere the subsys frees |
| 717 | * cgroup_subsys_state. |
| 718 | */ |
| 719 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css); |
| 720 | |
| 721 | /* Find a cgroup_subsys_state which has given ID */ |
| 722 | |
| 723 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id); |
| 724 | |
| 725 | /* |
| 726 | * Get a cgroup whose id is greater than or equal to id under tree of root. |
| 727 | * Returning a cgroup_subsys_state or NULL. |
| 728 | */ |
| 729 | struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id, |
| 730 | struct cgroup_subsys_state *root, int *foundid); |
| 731 | |
| 732 | /* Returns true if root is ancestor of cg */ |
| 733 | bool css_is_ancestor(struct cgroup_subsys_state *cg, |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 734 | const struct cgroup_subsys_state *root); |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 735 | |
| 736 | /* Get id and depth of css */ |
| 737 | unsigned short css_id(struct cgroup_subsys_state *css); |
| 738 | unsigned short css_depth(struct cgroup_subsys_state *css); |
Stephane Eranian | e5d1367 | 2011-02-14 11:20:01 +0200 | [diff] [blame] | 739 | struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id); |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 740 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 741 | #else /* !CONFIG_CGROUPS */ |
| 742 | |
| 743 | static inline int cgroup_init_early(void) { return 0; } |
| 744 | static inline int cgroup_init(void) { return 0; } |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 745 | static inline void cgroup_fork(struct task_struct *p) {} |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 746 | static inline void cgroup_post_fork(struct task_struct *p) {} |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 747 | static inline void cgroup_exit(struct task_struct *p, int callbacks) {} |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 748 | |
| 749 | static inline void cgroup_lock(void) {} |
| 750 | static inline void cgroup_unlock(void) {} |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 751 | static inline int cgroupstats_build(struct cgroupstats *stats, |
| 752 | struct dentry *dentry) |
| 753 | { |
| 754 | return -EINVAL; |
| 755 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 756 | |
Sridhar Samudrala | d7926ee | 2010-05-30 22:24:39 +0200 | [diff] [blame] | 757 | /* No cgroups - nothing to do */ |
Michael S. Tsirkin | 31583bb | 2010-09-09 16:37:37 -0700 | [diff] [blame] | 758 | static inline int cgroup_attach_task_all(struct task_struct *from, |
| 759 | struct task_struct *t) |
| 760 | { |
| 761 | return 0; |
| 762 | } |
Sridhar Samudrala | d7926ee | 2010-05-30 22:24:39 +0200 | [diff] [blame] | 763 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 764 | #endif /* !CONFIG_CGROUPS */ |
| 765 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 766 | #endif /* _LINUX_CGROUP_H */ |