Commit | Line | Data |
---|---|---|
ddbcc7e8 PM |
1 | #ifndef _LINUX_CGROUP_H |
2 | #define _LINUX_CGROUP_H | |
3 | /* | |
4 | * cgroup interface | |
5 | * | |
6 | * Copyright (C) 2003 BULL SA | |
7 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
8 | * | |
9 | */ | |
10 | ||
11 | #include <linux/sched.h> | |
ddbcc7e8 PM |
12 | #include <linux/cpumask.h> |
13 | #include <linux/nodemask.h> | |
14 | #include <linux/rcupdate.h> | |
eb6fd504 | 15 | #include <linux/rculist.h> |
846c7bb0 | 16 | #include <linux/cgroupstats.h> |
31a7df01 | 17 | #include <linux/prio_heap.h> |
cc31edce | 18 | #include <linux/rwsem.h> |
38460b48 | 19 | #include <linux/idr.h> |
48ddbe19 | 20 | #include <linux/workqueue.h> |
03b1cde6 | 21 | #include <linux/xattr.h> |
25a7e684 | 22 | #include <linux/fs.h> |
d3daf28d | 23 | #include <linux/percpu-refcount.h> |
ddbcc7e8 PM |
24 | |
25 | #ifdef CONFIG_CGROUPS | |
26 | ||
27 | struct cgroupfs_root; | |
28 | struct cgroup_subsys; | |
29 | struct inode; | |
84eea842 | 30 | struct cgroup; |
38460b48 | 31 | struct css_id; |
a27bb332 | 32 | struct eventfd_ctx; |
ddbcc7e8 PM |
33 | |
34 | extern int cgroup_init_early(void); | |
35 | extern int cgroup_init(void); | |
b4f48b63 | 36 | extern void cgroup_fork(struct task_struct *p); |
817929ec | 37 | extern void cgroup_post_fork(struct task_struct *p); |
b4f48b63 | 38 | extern void cgroup_exit(struct task_struct *p, int run_callbacks); |
846c7bb0 BS |
39 | extern int cgroupstats_build(struct cgroupstats *stats, |
40 | struct dentry *dentry); | |
e6a1105b | 41 | extern int cgroup_load_subsys(struct cgroup_subsys *ss); |
cf5d5941 | 42 | extern void cgroup_unload_subsys(struct cgroup_subsys *ss); |
ddbcc7e8 | 43 | |
8d8b97ba | 44 | extern int proc_cgroup_show(struct seq_file *, void *); |
a424316c | 45 | |
7d8e0bf5 LZ |
46 | /* |
47 | * Define the enumeration of all cgroup subsystems. | |
48 | * | |
49 | * We define ids for builtin subsystems and then modular ones. | |
50 | */ | |
817929ec PM |
51 | #define SUBSYS(_x) _x ## _subsys_id, |
52 | enum cgroup_subsys_id { | |
7d8e0bf5 | 53 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
817929ec | 54 | #include <linux/cgroup_subsys.h> |
7d8e0bf5 LZ |
55 | #undef IS_SUBSYS_ENABLED |
56 | CGROUP_BUILTIN_SUBSYS_COUNT, | |
57 | ||
58 | __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1, | |
59 | ||
60 | #define IS_SUBSYS_ENABLED(option) IS_MODULE(option) | |
817929ec | 61 | #include <linux/cgroup_subsys.h> |
7d8e0bf5 | 62 | #undef IS_SUBSYS_ENABLED |
a6f00298 | 63 | CGROUP_SUBSYS_COUNT, |
817929ec PM |
64 | }; |
65 | #undef SUBSYS | |
66 | ||
ddbcc7e8 PM |
67 | /* Per-subsystem/per-cgroup state maintained by the system. */ |
68 | struct cgroup_subsys_state { | |
72c97e54 | 69 | /* the cgroup that this css is attached to */ |
ddbcc7e8 PM |
70 | struct cgroup *cgroup; |
71 | ||
72c97e54 TH |
72 | /* the cgroup subsystem that this css is attached to */ |
73 | struct cgroup_subsys *ss; | |
74 | ||
d3daf28d TH |
75 | /* reference count - access via css_[try]get() and css_put() */ |
76 | struct percpu_ref refcnt; | |
ddbcc7e8 | 77 | |
0ae78e0b TH |
78 | /* the parent css */ |
79 | struct cgroup_subsys_state *parent; | |
80 | ||
ddbcc7e8 | 81 | unsigned long flags; |
38460b48 | 82 | /* ID for this css, if possible */ |
2c392b8c | 83 | struct css_id __rcu *id; |
48ddbe19 TH |
84 | |
85 | /* Used to put @cgroup->dentry on the last css_put() */ | |
35ef10da | 86 | struct work_struct destroy_work; |
ddbcc7e8 PM |
87 | }; |
88 | ||
89 | /* bits in struct cgroup_subsys_state flags field */ | |
90 | enum { | |
38b53aba | 91 | CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */ |
92fb9748 | 92 | CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ |
ddbcc7e8 PM |
93 | }; |
94 | ||
5de0107e TH |
95 | /** |
96 | * css_get - obtain a reference on the specified css | |
97 | * @css: target css | |
98 | * | |
99 | * The caller must already have a reference. | |
ddbcc7e8 | 100 | */ |
ddbcc7e8 PM |
101 | static inline void css_get(struct cgroup_subsys_state *css) |
102 | { | |
103 | /* We don't need to reference count the root state */ | |
38b53aba | 104 | if (!(css->flags & CSS_ROOT)) |
d3daf28d | 105 | percpu_ref_get(&css->refcnt); |
ddbcc7e8 | 106 | } |
e7c5ec91 | 107 | |
5de0107e TH |
108 | /** |
109 | * css_tryget - try to obtain a reference on the specified css | |
110 | * @css: target css | |
111 | * | |
112 | * Obtain a reference on @css if it's alive. The caller naturally needs to | |
113 | * ensure that @css is accessible but doesn't have to be holding a | |
114 | * reference on it - IOW, RCU protected access is good enough for this | |
115 | * function. Returns %true if a reference count was successfully obtained; | |
116 | * %false otherwise. | |
117 | */ | |
e7c5ec91 PM |
118 | static inline bool css_tryget(struct cgroup_subsys_state *css) |
119 | { | |
38b53aba | 120 | if (css->flags & CSS_ROOT) |
e7c5ec91 | 121 | return true; |
d3daf28d | 122 | return percpu_ref_tryget(&css->refcnt); |
e7c5ec91 PM |
123 | } |
124 | ||
5de0107e TH |
125 | /** |
126 | * css_put - put a css reference | |
127 | * @css: target css | |
128 | * | |
129 | * Put a reference obtained via css_get() and css_tryget(). | |
130 | */ | |
ddbcc7e8 PM |
131 | static inline void css_put(struct cgroup_subsys_state *css) |
132 | { | |
38b53aba | 133 | if (!(css->flags & CSS_ROOT)) |
d3daf28d | 134 | percpu_ref_put(&css->refcnt); |
ddbcc7e8 PM |
135 | } |
136 | ||
3116f0e3 PM |
137 | /* bits in struct cgroup flags field */ |
138 | enum { | |
139 | /* Control Group is dead */ | |
54766d4a | 140 | CGRP_DEAD, |
d20a390a PM |
141 | /* |
142 | * Control Group has previously had a child cgroup or a task, | |
143 | * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) | |
144 | */ | |
3116f0e3 PM |
145 | CGRP_RELEASABLE, |
146 | /* Control Group requires release notifications to userspace */ | |
147 | CGRP_NOTIFY_ON_RELEASE, | |
97978e6d | 148 | /* |
2260e7fc TH |
149 | * Clone the parent's configuration when creating a new child |
150 | * cpuset cgroup. For historical reasons, this option can be | |
151 | * specified at mount time and thus is implemented here. | |
97978e6d | 152 | */ |
2260e7fc | 153 | CGRP_CPUSET_CLONE_CHILDREN, |
873fe09e TH |
154 | /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */ |
155 | CGRP_SANE_BEHAVIOR, | |
3116f0e3 PM |
156 | }; |
157 | ||
65dff759 LZ |
158 | struct cgroup_name { |
159 | struct rcu_head rcu_head; | |
160 | char name[]; | |
3116f0e3 PM |
161 | }; |
162 | ||
ddbcc7e8 PM |
163 | struct cgroup { |
164 | unsigned long flags; /* "unsigned long" so bitops work */ | |
165 | ||
b414dc09 LZ |
166 | /* |
167 | * idr allocated in-hierarchy ID. | |
168 | * | |
169 | * The ID of the root cgroup is always 0, and a new cgroup | |
170 | * will be assigned with a smallest available ID. | |
171 | */ | |
172 | int id; | |
0a950f65 | 173 | |
ddbcc7e8 PM |
174 | /* |
175 | * We link our 'sibling' struct into our parent's 'children'. | |
176 | * Our children link their 'sibling' into our 'children'. | |
177 | */ | |
178 | struct list_head sibling; /* my parent's children */ | |
179 | struct list_head children; /* my children */ | |
05ef1d7c | 180 | struct list_head files; /* my files */ |
ddbcc7e8 | 181 | |
d20a390a | 182 | struct cgroup *parent; /* my parent */ |
febfcef6 | 183 | struct dentry *dentry; /* cgroup fs entry, RCU protected */ |
ddbcc7e8 | 184 | |
53fa5261 TH |
185 | /* |
186 | * Monotonically increasing unique serial number which defines a | |
187 | * uniform order among all cgroups. It's guaranteed that all | |
188 | * ->children lists are in the ascending order of ->serial_nr. | |
189 | * It's used to allow interrupting and resuming iterations. | |
190 | */ | |
191 | u64 serial_nr; | |
192 | ||
65dff759 LZ |
193 | /* |
194 | * This is a copy of dentry->d_name, and it's needed because | |
195 | * we can't use dentry->d_name in cgroup_path(). | |
196 | * | |
197 | * You must acquire rcu_read_lock() to access cgrp->name, and | |
198 | * the only place that can change it is rename(), which is | |
199 | * protected by parent dir's i_mutex. | |
200 | * | |
201 | * Normally you should use cgroup_name() wrapper rather than | |
202 | * access it directly. | |
203 | */ | |
204 | struct cgroup_name __rcu *name; | |
205 | ||
ddbcc7e8 PM |
206 | /* Private pointers for each registered subsystem */ |
207 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; | |
208 | ||
209 | struct cgroupfs_root *root; | |
817929ec PM |
210 | |
211 | /* | |
69d0206c TH |
212 | * List of cgrp_cset_links pointing at css_sets with tasks in this |
213 | * cgroup. Protected by css_set_lock. | |
817929ec | 214 | */ |
69d0206c | 215 | struct list_head cset_links; |
81a6a5cd PM |
216 | |
217 | /* | |
218 | * Linked list running through all cgroups that can | |
219 | * potentially be reaped by the release agent. Protected by | |
220 | * release_list_lock | |
221 | */ | |
222 | struct list_head release_list; | |
cc31edce | 223 | |
72a8cb30 BB |
224 | /* |
225 | * list of pidlists, up to two for each namespace (one for procs, one | |
226 | * for tasks); created on demand. | |
227 | */ | |
228 | struct list_head pidlists; | |
229 | struct mutex pidlist_mutex; | |
a47295e6 | 230 | |
67f4c36f TH |
231 | /* dummy css with NULL ->ss, points back to this cgroup */ |
232 | struct cgroup_subsys_state dummy_css; | |
233 | ||
d3daf28d | 234 | /* For css percpu_ref killing and RCU-protected deletion */ |
a47295e6 | 235 | struct rcu_head rcu_head; |
ea15f8cc | 236 | struct work_struct destroy_work; |
d3daf28d | 237 | atomic_t css_kill_cnt; |
0dea1168 | 238 | |
25985edc | 239 | /* List of events which userspace want to receive */ |
0dea1168 KS |
240 | struct list_head event_list; |
241 | spinlock_t event_list_lock; | |
03b1cde6 AR |
242 | |
243 | /* directory xattrs */ | |
244 | struct simple_xattrs xattrs; | |
817929ec PM |
245 | }; |
246 | ||
25a7e684 TH |
247 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
248 | ||
249 | /* cgroupfs_root->flags */ | |
250 | enum { | |
873fe09e TH |
251 | /* |
252 | * Unfortunately, cgroup core and various controllers are riddled | |
253 | * with idiosyncrasies and pointless options. The following flag, | |
254 | * when set, will force sane behavior - some options are forced on, | |
255 | * others are disallowed, and some controllers will change their | |
256 | * hierarchical or other behaviors. | |
257 | * | |
258 | * The set of behaviors affected by this flag are still being | |
259 | * determined and developed and the mount option for this flag is | |
260 | * prefixed with __DEVEL__. The prefix will be dropped once we | |
261 | * reach the point where all behaviors are compatible with the | |
262 | * planned unified hierarchy, which will automatically turn on this | |
263 | * flag. | |
264 | * | |
265 | * The followings are the behaviors currently affected this flag. | |
266 | * | |
267 | * - Mount options "noprefix" and "clone_children" are disallowed. | |
268 | * Also, cgroupfs file cgroup.clone_children is not created. | |
269 | * | |
270 | * - When mounting an existing superblock, mount options should | |
271 | * match. | |
272 | * | |
273 | * - Remount is disallowed. | |
274 | * | |
0b0585c3 LT |
275 | * - rename(2) is disallowed. |
276 | * | |
f63674fd TH |
277 | * - "tasks" is removed. Everything should be at process |
278 | * granularity. Use "cgroup.procs" instead. | |
f00baae7 | 279 | * |
f63674fd TH |
280 | * - "release_agent" and "notify_on_release" are removed. |
281 | * Replacement notification mechanism will be implemented. | |
873fe09e | 282 | * |
5c5cc623 LZ |
283 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens |
284 | * and take masks of ancestors with non-empty cpus/mems, instead of | |
285 | * being moved to an ancestor. | |
286 | * | |
88fa523b LZ |
287 | * - cpuset: a task can be moved into an empty cpuset, and again it |
288 | * takes masks of ancestors. | |
6db8e85c | 289 | * |
f63674fd TH |
290 | * - memcg: use_hierarchy is on by default and the cgroup file for |
291 | * the flag is not created. | |
f00baae7 | 292 | * |
9138125b | 293 | * - blkcg: blk-throttle becomes properly hierarchical. |
873fe09e TH |
294 | */ |
295 | CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), | |
296 | ||
25a7e684 TH |
297 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
298 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ | |
0ce6cba3 TH |
299 | |
300 | /* mount options live below bit 16 */ | |
301 | CGRP_ROOT_OPTION_MASK = (1 << 16) - 1, | |
302 | ||
303 | CGRP_ROOT_SUBSYS_BOUND = (1 << 16), /* subsystems finished binding */ | |
25a7e684 TH |
304 | }; |
305 | ||
306 | /* | |
307 | * A cgroupfs_root represents the root of a cgroup hierarchy, and may be | |
308 | * associated with a superblock to form an active hierarchy. This is | |
309 | * internal to cgroup core. Don't access directly from controllers. | |
310 | */ | |
311 | struct cgroupfs_root { | |
312 | struct super_block *sb; | |
313 | ||
a8a648c4 | 314 | /* The bitmask of subsystems attached to this hierarchy */ |
25a7e684 TH |
315 | unsigned long subsys_mask; |
316 | ||
317 | /* Unique id for this hierarchy. */ | |
318 | int hierarchy_id; | |
319 | ||
25a7e684 TH |
320 | /* A list running through the attached subsystems */ |
321 | struct list_head subsys_list; | |
322 | ||
323 | /* The root cgroup for this hierarchy */ | |
324 | struct cgroup top_cgroup; | |
325 | ||
326 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
327 | int number_of_cgroups; | |
328 | ||
329 | /* A list running through the active hierarchies */ | |
330 | struct list_head root_list; | |
331 | ||
25a7e684 TH |
332 | /* Hierarchy-specific flags */ |
333 | unsigned long flags; | |
334 | ||
335 | /* IDs for cgroups in this hierarchy */ | |
4e96ee8e | 336 | struct idr cgroup_idr; |
25a7e684 TH |
337 | |
338 | /* The path to use for release notifications. */ | |
339 | char release_agent_path[PATH_MAX]; | |
340 | ||
341 | /* The name for this hierarchy - may be empty */ | |
342 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
343 | }; | |
344 | ||
d20a390a PM |
345 | /* |
346 | * A css_set is a structure holding pointers to a set of | |
817929ec PM |
347 | * cgroup_subsys_state objects. This saves space in the task struct |
348 | * object and speeds up fork()/exit(), since a single inc/dec and a | |
d20a390a PM |
349 | * list_add()/del() can bump the reference count on the entire cgroup |
350 | * set for a task. | |
817929ec PM |
351 | */ |
352 | ||
353 | struct css_set { | |
354 | ||
355 | /* Reference count */ | |
146aa1bd | 356 | atomic_t refcount; |
817929ec | 357 | |
472b1053 LZ |
358 | /* |
359 | * List running through all cgroup groups in the same hash | |
360 | * slot. Protected by css_set_lock | |
361 | */ | |
362 | struct hlist_node hlist; | |
363 | ||
817929ec PM |
364 | /* |
365 | * List running through all tasks using this cgroup | |
366 | * group. Protected by css_set_lock | |
367 | */ | |
368 | struct list_head tasks; | |
369 | ||
370 | /* | |
69d0206c TH |
371 | * List of cgrp_cset_links pointing at cgroups referenced from this |
372 | * css_set. Protected by css_set_lock. | |
817929ec | 373 | */ |
69d0206c | 374 | struct list_head cgrp_links; |
817929ec PM |
375 | |
376 | /* | |
377 | * Set of subsystem states, one for each subsystem. This array | |
378 | * is immutable after creation apart from the init_css_set | |
cf5d5941 BB |
379 | * during subsystem registration (at boot time) and modular subsystem |
380 | * loading/unloading. | |
817929ec PM |
381 | */ |
382 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; | |
c378369d BB |
383 | |
384 | /* For RCU-protected deletion */ | |
385 | struct rcu_head rcu_head; | |
ddbcc7e8 PM |
386 | }; |
387 | ||
91796569 PM |
388 | /* |
389 | * cgroup_map_cb is an abstract callback API for reporting map-valued | |
390 | * control files | |
391 | */ | |
392 | ||
393 | struct cgroup_map_cb { | |
394 | int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value); | |
395 | void *state; | |
396 | }; | |
397 | ||
d20a390a PM |
398 | /* |
399 | * struct cftype: handler definitions for cgroup control files | |
ddbcc7e8 PM |
400 | * |
401 | * When reading/writing to a file: | |
a043e3b2 | 402 | * - the cgroup to use is file->f_dentry->d_parent->d_fsdata |
ddbcc7e8 PM |
403 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
404 | */ | |
405 | ||
8e3f6541 | 406 | /* cftype->flags */ |
02c402d9 | 407 | enum { |
6f4b7e63 LZ |
408 | CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ |
409 | CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ | |
02c402d9 TH |
410 | CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */ |
411 | }; | |
8e3f6541 TH |
412 | |
413 | #define MAX_CFTYPE_NAME 64 | |
414 | ||
ddbcc7e8 | 415 | struct cftype { |
d20a390a PM |
416 | /* |
417 | * By convention, the name should begin with the name of the | |
8e3f6541 TH |
418 | * subsystem, followed by a period. Zero length string indicates |
419 | * end of cftype array. | |
d20a390a | 420 | */ |
ddbcc7e8 PM |
421 | char name[MAX_CFTYPE_NAME]; |
422 | int private; | |
099fca32 LZ |
423 | /* |
424 | * If not 0, file mode is set to this value, otherwise it will | |
425 | * be figured out automatically | |
426 | */ | |
a5e7ed32 | 427 | umode_t mode; |
db3b1497 PM |
428 | |
429 | /* | |
430 | * If non-zero, defines the maximum length of string that can | |
431 | * be passed to write_string; defaults to 64 | |
432 | */ | |
433 | size_t max_write_len; | |
434 | ||
8e3f6541 TH |
435 | /* CFTYPE_* flags */ |
436 | unsigned int flags; | |
437 | ||
2bb566cb TH |
438 | /* |
439 | * The subsys this file belongs to. Initialized automatically | |
440 | * during registration. NULL for cgroup core files. | |
441 | */ | |
442 | struct cgroup_subsys *ss; | |
443 | ||
ce16b49d | 444 | int (*open)(struct inode *inode, struct file *file); |
182446d0 | 445 | ssize_t (*read)(struct cgroup_subsys_state *css, struct cftype *cft, |
ce16b49d PM |
446 | struct file *file, |
447 | char __user *buf, size_t nbytes, loff_t *ppos); | |
ddbcc7e8 | 448 | /* |
f4c753b7 | 449 | * read_u64() is a shortcut for the common case of returning a |
ddbcc7e8 PM |
450 | * single integer. Use it in place of read() |
451 | */ | |
182446d0 | 452 | u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); |
e73d2c61 PM |
453 | /* |
454 | * read_s64() is a signed version of read_u64() | |
455 | */ | |
182446d0 | 456 | s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); |
91796569 PM |
457 | /* |
458 | * read_map() is used for defining a map of key/value | |
459 | * pairs. It should call cb->fill(cb, key, value) for each | |
460 | * entry. The key/value pairs (and their ordering) should not | |
461 | * change between reboots. | |
462 | */ | |
182446d0 | 463 | int (*read_map)(struct cgroup_subsys_state *css, struct cftype *cft, |
ce16b49d | 464 | struct cgroup_map_cb *cb); |
29486df3 SH |
465 | /* |
466 | * read_seq_string() is used for outputting a simple sequence | |
467 | * using seqfile. | |
468 | */ | |
182446d0 TH |
469 | int (*read_seq_string)(struct cgroup_subsys_state *css, |
470 | struct cftype *cft, struct seq_file *m); | |
91796569 | 471 | |
182446d0 | 472 | ssize_t (*write)(struct cgroup_subsys_state *css, struct cftype *cft, |
ce16b49d PM |
473 | struct file *file, |
474 | const char __user *buf, size_t nbytes, loff_t *ppos); | |
355e0c48 PM |
475 | |
476 | /* | |
f4c753b7 | 477 | * write_u64() is a shortcut for the common case of accepting |
355e0c48 PM |
478 | * a single integer (as parsed by simple_strtoull) from |
479 | * userspace. Use in place of write(); return 0 or error. | |
480 | */ | |
182446d0 TH |
481 | int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, |
482 | u64 val); | |
e73d2c61 PM |
483 | /* |
484 | * write_s64() is a signed version of write_u64() | |
485 | */ | |
182446d0 TH |
486 | int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, |
487 | s64 val); | |
355e0c48 | 488 | |
db3b1497 PM |
489 | /* |
490 | * write_string() is passed a nul-terminated kernelspace | |
491 | * buffer of maximum length determined by max_write_len. | |
492 | * Returns 0 or -ve error code. | |
493 | */ | |
182446d0 | 494 | int (*write_string)(struct cgroup_subsys_state *css, struct cftype *cft, |
db3b1497 | 495 | const char *buffer); |
d447ea2f PE |
496 | /* |
497 | * trigger() callback can be used to get some kick from the | |
498 | * userspace, when the actual string written is not important | |
499 | * at all. The private field can be used to determine the | |
500 | * kick type for multiplexing. | |
501 | */ | |
182446d0 | 502 | int (*trigger)(struct cgroup_subsys_state *css, unsigned int event); |
d447ea2f | 503 | |
ce16b49d | 504 | int (*release)(struct inode *inode, struct file *file); |
0dea1168 KS |
505 | |
506 | /* | |
507 | * register_event() callback will be used to add new userspace | |
508 | * waiter for changes related to the cftype. Implement it if | |
509 | * you want to provide this functionality. Use eventfd_signal() | |
510 | * on eventfd to send notification to userspace. | |
511 | */ | |
81eeaf04 TH |
512 | int (*register_event)(struct cgroup_subsys_state *css, |
513 | struct cftype *cft, struct eventfd_ctx *eventfd, | |
514 | const char *args); | |
0dea1168 KS |
515 | /* |
516 | * unregister_event() callback will be called when userspace | |
517 | * closes the eventfd or on cgroup removing. | |
518 | * This callback must be implemented, if you want provide | |
519 | * notification functionality. | |
0dea1168 | 520 | */ |
81eeaf04 TH |
521 | void (*unregister_event)(struct cgroup_subsys_state *css, |
522 | struct cftype *cft, | |
523 | struct eventfd_ctx *eventfd); | |
ddbcc7e8 PM |
524 | }; |
525 | ||
8e3f6541 TH |
526 | /* |
527 | * cftype_sets describe cftypes belonging to a subsystem and are chained at | |
528 | * cgroup_subsys->cftsets. Each cftset points to an array of cftypes | |
529 | * terminated by zero length name. | |
530 | */ | |
531 | struct cftype_set { | |
532 | struct list_head node; /* chained at subsys->cftsets */ | |
03b1cde6 | 533 | struct cftype *cfts; |
8e3f6541 TH |
534 | }; |
535 | ||
873fe09e TH |
536 | /* |
537 | * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This | |
538 | * function can be called as long as @cgrp is accessible. | |
539 | */ | |
540 | static inline bool cgroup_sane_behavior(const struct cgroup *cgrp) | |
541 | { | |
542 | return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR; | |
543 | } | |
544 | ||
65dff759 LZ |
545 | /* Caller should hold rcu_read_lock() */ |
546 | static inline const char *cgroup_name(const struct cgroup *cgrp) | |
547 | { | |
548 | return rcu_dereference(cgrp->name)->name; | |
549 | } | |
550 | ||
03b1cde6 | 551 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
2bb566cb | 552 | int cgroup_rm_cftypes(struct cftype *cfts); |
8e3f6541 | 553 | |
78574cf9 | 554 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor); |
ddbcc7e8 | 555 | |
ffd2d883 | 556 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen); |
913ffdb5 | 557 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen); |
ddbcc7e8 | 558 | |
ffd2d883 | 559 | int cgroup_task_count(const struct cgroup *cgrp); |
bbcb81d0 | 560 | |
2f7ee569 TH |
561 | /* |
562 | * Control Group taskset, used to pass around set of tasks to cgroup_subsys | |
563 | * methods. | |
564 | */ | |
565 | struct cgroup_taskset; | |
566 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset); | |
567 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset); | |
d99c8727 TH |
568 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
569 | int subsys_id); | |
2f7ee569 TH |
570 | int cgroup_taskset_size(struct cgroup_taskset *tset); |
571 | ||
572 | /** | |
573 | * cgroup_taskset_for_each - iterate cgroup_taskset | |
574 | * @task: the loop cursor | |
d99c8727 | 575 | * @skip_css: skip if task's css matches this, %NULL to iterate through all |
2f7ee569 TH |
576 | * @tset: taskset to iterate |
577 | */ | |
d99c8727 | 578 | #define cgroup_taskset_for_each(task, skip_css, tset) \ |
2f7ee569 TH |
579 | for ((task) = cgroup_taskset_first((tset)); (task); \ |
580 | (task) = cgroup_taskset_next((tset))) \ | |
d99c8727 TH |
581 | if (!(skip_css) || \ |
582 | cgroup_taskset_cur_css((tset), \ | |
583 | (skip_css)->ss->subsys_id) != (skip_css)) | |
2f7ee569 | 584 | |
21acb9ca TLSC |
585 | /* |
586 | * Control Group subsystem type. | |
587 | * See Documentation/cgroups/cgroups.txt for details | |
588 | */ | |
ddbcc7e8 PM |
589 | |
590 | struct cgroup_subsys { | |
eb95419b TH |
591 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); |
592 | int (*css_online)(struct cgroup_subsys_state *css); | |
593 | void (*css_offline)(struct cgroup_subsys_state *css); | |
594 | void (*css_free)(struct cgroup_subsys_state *css); | |
595 | ||
596 | int (*can_attach)(struct cgroup_subsys_state *css, | |
597 | struct cgroup_taskset *tset); | |
598 | void (*cancel_attach)(struct cgroup_subsys_state *css, | |
599 | struct cgroup_taskset *tset); | |
600 | void (*attach)(struct cgroup_subsys_state *css, | |
601 | struct cgroup_taskset *tset); | |
761b3ef5 | 602 | void (*fork)(struct task_struct *task); |
eb95419b TH |
603 | void (*exit)(struct cgroup_subsys_state *css, |
604 | struct cgroup_subsys_state *old_css, | |
761b3ef5 | 605 | struct task_struct *task); |
eb95419b | 606 | void (*bind)(struct cgroup_subsys_state *root_css); |
e5991371 | 607 | |
ddbcc7e8 | 608 | int subsys_id; |
8bab8dde | 609 | int disabled; |
ddbcc7e8 | 610 | int early_init; |
38460b48 KH |
611 | /* |
612 | * True if this subsys uses ID. ID is not available before cgroup_init() | |
613 | * (not available in early_init time.) | |
614 | */ | |
615 | bool use_id; | |
48ddbe19 | 616 | |
8c7f6edb TH |
617 | /* |
618 | * If %false, this subsystem is properly hierarchical - | |
619 | * configuration, resource accounting and restriction on a parent | |
620 | * cgroup cover those of its children. If %true, hierarchy support | |
621 | * is broken in some ways - some subsystems ignore hierarchy | |
622 | * completely while others are only implemented half-way. | |
623 | * | |
624 | * It's now disallowed to create nested cgroups if the subsystem is | |
625 | * broken and cgroup core will emit a warning message on such | |
626 | * cases. Eventually, all subsystems will be made properly | |
627 | * hierarchical and this will go away. | |
628 | */ | |
629 | bool broken_hierarchy; | |
630 | bool warned_broken_hierarchy; | |
631 | ||
ddbcc7e8 PM |
632 | #define MAX_CGROUP_TYPE_NAMELEN 32 |
633 | const char *name; | |
634 | ||
999cd8a4 PM |
635 | /* |
636 | * Link to parent, and list entry in parent's children. | |
6be96a5c | 637 | * Protected by cgroup_lock() |
999cd8a4 PM |
638 | */ |
639 | struct cgroupfs_root *root; | |
ddbcc7e8 | 640 | struct list_head sibling; |
38460b48 KH |
641 | /* used when use_id == true */ |
642 | struct idr idr; | |
42aee6c4 | 643 | spinlock_t id_lock; |
e6a1105b | 644 | |
8e3f6541 TH |
645 | /* list of cftype_sets */ |
646 | struct list_head cftsets; | |
647 | ||
648 | /* base cftypes, automatically [de]registered with subsys itself */ | |
649 | struct cftype *base_cftypes; | |
650 | struct cftype_set base_cftset; | |
651 | ||
e6a1105b BB |
652 | /* should be defined only by modular subsystems */ |
653 | struct module *module; | |
ddbcc7e8 PM |
654 | }; |
655 | ||
656 | #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys; | |
5fc0b025 | 657 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
ddbcc7e8 | 658 | #include <linux/cgroup_subsys.h> |
5fc0b025 | 659 | #undef IS_SUBSYS_ENABLED |
ddbcc7e8 PM |
660 | #undef SUBSYS |
661 | ||
63876986 TH |
662 | /** |
663 | * css_parent - find the parent css | |
664 | * @css: the target cgroup_subsys_state | |
665 | * | |
666 | * Return the parent css of @css. This function is guaranteed to return | |
667 | * non-NULL parent as long as @css isn't the root. | |
668 | */ | |
669 | static inline | |
670 | struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css) | |
671 | { | |
0ae78e0b | 672 | return css->parent; |
63876986 TH |
673 | } |
674 | ||
14611e51 TH |
675 | /** |
676 | * task_css_set_check - obtain a task's css_set with extra access conditions | |
677 | * @task: the task to obtain css_set for | |
678 | * @__c: extra condition expression to be passed to rcu_dereference_check() | |
679 | * | |
680 | * A task's css_set is RCU protected, initialized and exited while holding | |
681 | * task_lock(), and can only be modified while holding both cgroup_mutex | |
682 | * and task_lock() while the task is alive. This macro verifies that the | |
683 | * caller is inside proper critical section and returns @task's css_set. | |
684 | * | |
685 | * The caller can also specify additional allowed conditions via @__c, such | |
686 | * as locks used during the cgroup_subsys::attach() methods. | |
dc61b1d6 | 687 | */ |
2219449a TH |
688 | #ifdef CONFIG_PROVE_RCU |
689 | extern struct mutex cgroup_mutex; | |
14611e51 TH |
690 | #define task_css_set_check(task, __c) \ |
691 | rcu_dereference_check((task)->cgroups, \ | |
692 | lockdep_is_held(&(task)->alloc_lock) || \ | |
693 | lockdep_is_held(&cgroup_mutex) || (__c)) | |
2219449a | 694 | #else |
14611e51 TH |
695 | #define task_css_set_check(task, __c) \ |
696 | rcu_dereference((task)->cgroups) | |
2219449a | 697 | #endif |
dc61b1d6 | 698 | |
14611e51 | 699 | /** |
8af01f56 | 700 | * task_css_check - obtain css for (task, subsys) w/ extra access conds |
14611e51 TH |
701 | * @task: the target task |
702 | * @subsys_id: the target subsystem ID | |
703 | * @__c: extra condition expression to be passed to rcu_dereference_check() | |
704 | * | |
705 | * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The | |
706 | * synchronization rules are the same as task_css_set_check(). | |
707 | */ | |
8af01f56 | 708 | #define task_css_check(task, subsys_id, __c) \ |
14611e51 TH |
709 | task_css_set_check((task), (__c))->subsys[(subsys_id)] |
710 | ||
711 | /** | |
712 | * task_css_set - obtain a task's css_set | |
713 | * @task: the task to obtain css_set for | |
714 | * | |
715 | * See task_css_set_check(). | |
716 | */ | |
717 | static inline struct css_set *task_css_set(struct task_struct *task) | |
718 | { | |
719 | return task_css_set_check(task, false); | |
720 | } | |
721 | ||
722 | /** | |
8af01f56 | 723 | * task_css - obtain css for (task, subsys) |
14611e51 TH |
724 | * @task: the target task |
725 | * @subsys_id: the target subsystem ID | |
726 | * | |
8af01f56 | 727 | * See task_css_check(). |
14611e51 | 728 | */ |
8af01f56 TH |
729 | static inline struct cgroup_subsys_state *task_css(struct task_struct *task, |
730 | int subsys_id) | |
ddbcc7e8 | 731 | { |
8af01f56 | 732 | return task_css_check(task, subsys_id, false); |
ddbcc7e8 PM |
733 | } |
734 | ||
8af01f56 TH |
735 | static inline struct cgroup *task_cgroup(struct task_struct *task, |
736 | int subsys_id) | |
ddbcc7e8 | 737 | { |
8af01f56 | 738 | return task_css(task, subsys_id)->cgroup; |
ddbcc7e8 PM |
739 | } |
740 | ||
e14880f7 LZ |
741 | /** |
742 | * cgroup_from_id - lookup cgroup by id | |
743 | * @ss: cgroup subsys to be looked into | |
744 | * @id: the cgroup id | |
745 | * | |
746 | * Returns the cgroup if there's valid one with @id, otherwise returns NULL. | |
747 | * Should be called under rcu_read_lock(). | |
748 | */ | |
749 | static inline struct cgroup *cgroup_from_id(struct cgroup_subsys *ss, int id) | |
750 | { | |
751 | #ifdef CONFIG_PROVE_RCU | |
752 | rcu_lockdep_assert(rcu_read_lock_held() || | |
753 | lockdep_is_held(&cgroup_mutex), | |
754 | "cgroup_from_id() needs proper protection"); | |
755 | #endif | |
756 | return idr_find(&ss->root->cgroup_idr, id); | |
757 | } | |
758 | ||
492eb21b TH |
759 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
760 | struct cgroup_subsys_state *parent); | |
53fa5261 | 761 | |
574bd9f7 | 762 | /** |
492eb21b TH |
763 | * css_for_each_child - iterate through children of a css |
764 | * @pos: the css * to use as the loop cursor | |
765 | * @parent: css whose children to walk | |
574bd9f7 | 766 | * |
492eb21b TH |
767 | * Walk @parent's children. Must be called under rcu_read_lock(). A child |
768 | * css which hasn't finished ->css_online() or already has finished | |
92fb9748 | 769 | * ->css_offline() may show up during traversal and it's each subsystem's |
574bd9f7 TH |
770 | * responsibility to verify that each @pos is alive. |
771 | * | |
92fb9748 | 772 | * If a subsystem synchronizes against the parent in its ->css_online() and |
492eb21b | 773 | * before starting iterating, a css which finished ->css_online() is |
92fb9748 | 774 | * guaranteed to be visible in the future iterations. |
75501a6d TH |
775 | * |
776 | * It is allowed to temporarily drop RCU read lock during iteration. The | |
777 | * caller is responsible for ensuring that @pos remains accessible until | |
778 | * the start of the next iteration by, for example, bumping the css refcnt. | |
574bd9f7 | 779 | */ |
492eb21b TH |
780 | #define css_for_each_child(pos, parent) \ |
781 | for ((pos) = css_next_child(NULL, (parent)); (pos); \ | |
782 | (pos) = css_next_child((pos), (parent))) | |
574bd9f7 | 783 | |
492eb21b TH |
784 | struct cgroup_subsys_state * |
785 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
786 | struct cgroup_subsys_state *css); | |
787 | ||
788 | struct cgroup_subsys_state * | |
789 | css_rightmost_descendant(struct cgroup_subsys_state *pos); | |
574bd9f7 TH |
790 | |
791 | /** | |
492eb21b TH |
792 | * css_for_each_descendant_pre - pre-order walk of a css's descendants |
793 | * @pos: the css * to use as the loop cursor | |
794 | * @root: css whose descendants to walk | |
574bd9f7 | 795 | * |
bd8815a6 TH |
796 | * Walk @root's descendants. @root is included in the iteration and the |
797 | * first node to be visited. Must be called under rcu_read_lock(). A | |
492eb21b | 798 | * descendant css which hasn't finished ->css_online() or already has |
92fb9748 | 799 | * finished ->css_offline() may show up during traversal and it's each |
574bd9f7 TH |
800 | * subsystem's responsibility to verify that each @pos is alive. |
801 | * | |
92fb9748 TH |
802 | * If a subsystem synchronizes against the parent in its ->css_online() and |
803 | * before starting iterating, and synchronizes against @pos on each | |
492eb21b | 804 | * iteration, any descendant css which finished ->css_online() is |
574bd9f7 TH |
805 | * guaranteed to be visible in the future iterations. |
806 | * | |
807 | * In other words, the following guarantees that a descendant can't escape | |
808 | * state updates of its ancestors. | |
809 | * | |
492eb21b | 810 | * my_online(@css) |
574bd9f7 | 811 | * { |
492eb21b TH |
812 | * Lock @css's parent and @css; |
813 | * Inherit state from the parent; | |
574bd9f7 TH |
814 | * Unlock both. |
815 | * } | |
816 | * | |
492eb21b | 817 | * my_update_state(@css) |
574bd9f7 | 818 | * { |
492eb21b | 819 | * css_for_each_descendant_pre(@pos, @css) { |
574bd9f7 | 820 | * Lock @pos; |
bd8815a6 TH |
821 | * if (@pos == @css) |
822 | * Update @css's state; | |
823 | * else | |
824 | * Verify @pos is alive and inherit state from its parent; | |
574bd9f7 TH |
825 | * Unlock @pos; |
826 | * } | |
827 | * } | |
828 | * | |
829 | * As long as the inheriting step, including checking the parent state, is | |
830 | * enclosed inside @pos locking, double-locking the parent isn't necessary | |
831 | * while inheriting. The state update to the parent is guaranteed to be | |
832 | * visible by walking order and, as long as inheriting operations to the | |
833 | * same @pos are atomic to each other, multiple updates racing each other | |
834 | * still result in the correct state. It's guaranateed that at least one | |
492eb21b | 835 | * inheritance happens for any css after the latest update to its parent. |
574bd9f7 TH |
836 | * |
837 | * If checking parent's state requires locking the parent, each inheriting | |
838 | * iteration should lock and unlock both @pos->parent and @pos. | |
839 | * | |
840 | * Alternatively, a subsystem may choose to use a single global lock to | |
92fb9748 | 841 | * synchronize ->css_online() and ->css_offline() against tree-walking |
574bd9f7 | 842 | * operations. |
75501a6d TH |
843 | * |
844 | * It is allowed to temporarily drop RCU read lock during iteration. The | |
845 | * caller is responsible for ensuring that @pos remains accessible until | |
846 | * the start of the next iteration by, for example, bumping the css refcnt. | |
574bd9f7 | 847 | */ |
492eb21b TH |
848 | #define css_for_each_descendant_pre(pos, css) \ |
849 | for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \ | |
850 | (pos) = css_next_descendant_pre((pos), (css))) | |
574bd9f7 | 851 | |
492eb21b TH |
852 | struct cgroup_subsys_state * |
853 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
854 | struct cgroup_subsys_state *css); | |
574bd9f7 TH |
855 | |
856 | /** | |
492eb21b TH |
857 | * css_for_each_descendant_post - post-order walk of a css's descendants |
858 | * @pos: the css * to use as the loop cursor | |
859 | * @css: css whose descendants to walk | |
574bd9f7 | 860 | * |
492eb21b | 861 | * Similar to css_for_each_descendant_pre() but performs post-order |
bd8815a6 TH |
862 | * traversal instead. @root is included in the iteration and the last |
863 | * node to be visited. Note that the walk visibility guarantee described | |
864 | * in pre-order walk doesn't apply the same to post-order walks. | |
574bd9f7 | 865 | */ |
492eb21b TH |
866 | #define css_for_each_descendant_post(pos, css) \ |
867 | for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \ | |
868 | (pos) = css_next_descendant_post((pos), (css))) | |
574bd9f7 | 869 | |
72ec7029 TH |
870 | /* A css_task_iter should be treated as an opaque object */ |
871 | struct css_task_iter { | |
872 | struct cgroup_subsys_state *origin_css; | |
0942eeee TH |
873 | struct list_head *cset_link; |
874 | struct list_head *task; | |
817929ec PM |
875 | }; |
876 | ||
72ec7029 TH |
877 | void css_task_iter_start(struct cgroup_subsys_state *css, |
878 | struct css_task_iter *it); | |
879 | struct task_struct *css_task_iter_next(struct css_task_iter *it); | |
880 | void css_task_iter_end(struct css_task_iter *it); | |
e535837b | 881 | |
72ec7029 TH |
882 | int css_scan_tasks(struct cgroup_subsys_state *css, |
883 | bool (*test)(struct task_struct *, void *), | |
884 | void (*process)(struct task_struct *, void *), | |
885 | void *data, struct ptr_heap *heap); | |
e535837b | 886 | |
31583bb0 | 887 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); |
8cc99345 | 888 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from); |
31583bb0 | 889 | |
38460b48 KH |
890 | /* |
891 | * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works | |
892 | * if cgroup_subsys.use_id == true. It can be used for looking up and scanning. | |
893 | * CSS ID is assigned at cgroup allocation (create) automatically | |
894 | * and removed when subsys calls free_css_id() function. This is because | |
895 | * the lifetime of cgroup_subsys_state is subsys's matter. | |
896 | * | |
897 | * Looking up and scanning function should be called under rcu_read_lock(). | |
6be96a5c | 898 | * Taking cgroup_mutex is not necessary for following calls. |
38460b48 KH |
899 | * But the css returned by this routine can be "not populated yet" or "being |
900 | * destroyed". The caller should check css and cgroup's status. | |
901 | */ | |
902 | ||
903 | /* | |
904 | * Typically Called at ->destroy(), or somewhere the subsys frees | |
905 | * cgroup_subsys_state. | |
906 | */ | |
907 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css); | |
908 | ||
909 | /* Find a cgroup_subsys_state which has given ID */ | |
910 | ||
911 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id); | |
912 | ||
38460b48 KH |
913 | /* Returns true if root is ancestor of cg */ |
914 | bool css_is_ancestor(struct cgroup_subsys_state *cg, | |
0b7f569e | 915 | const struct cgroup_subsys_state *root); |
38460b48 KH |
916 | |
917 | /* Get id and depth of css */ | |
918 | unsigned short css_id(struct cgroup_subsys_state *css); | |
e5d1367f | 919 | struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id); |
38460b48 | 920 | |
ddbcc7e8 PM |
921 | #else /* !CONFIG_CGROUPS */ |
922 | ||
923 | static inline int cgroup_init_early(void) { return 0; } | |
924 | static inline int cgroup_init(void) { return 0; } | |
b4f48b63 | 925 | static inline void cgroup_fork(struct task_struct *p) {} |
817929ec | 926 | static inline void cgroup_post_fork(struct task_struct *p) {} |
b4f48b63 | 927 | static inline void cgroup_exit(struct task_struct *p, int callbacks) {} |
ddbcc7e8 | 928 | |
846c7bb0 BS |
929 | static inline int cgroupstats_build(struct cgroupstats *stats, |
930 | struct dentry *dentry) | |
931 | { | |
932 | return -EINVAL; | |
933 | } | |
ddbcc7e8 | 934 | |
d7926ee3 | 935 | /* No cgroups - nothing to do */ |
31583bb0 MT |
936 | static inline int cgroup_attach_task_all(struct task_struct *from, |
937 | struct task_struct *t) | |
938 | { | |
939 | return 0; | |
940 | } | |
d7926ee3 | 941 | |
ddbcc7e8 PM |
942 | #endif /* !CONFIG_CGROUPS */ |
943 | ||
944 | #endif /* _LINUX_CGROUP_H */ |