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> |
cc31edce | 17 | #include <linux/rwsem.h> |
38460b48 | 18 | #include <linux/idr.h> |
48ddbe19 | 19 | #include <linux/workqueue.h> |
25a7e684 | 20 | #include <linux/fs.h> |
d3daf28d | 21 | #include <linux/percpu-refcount.h> |
7da11279 | 22 | #include <linux/seq_file.h> |
2bd59d48 | 23 | #include <linux/kernfs.h> |
f8f22e53 | 24 | #include <linux/wait.h> |
ddbcc7e8 PM |
25 | |
26 | #ifdef CONFIG_CGROUPS | |
27 | ||
3dd06ffa | 28 | struct cgroup_root; |
ddbcc7e8 PM |
29 | struct cgroup_subsys; |
30 | struct inode; | |
84eea842 | 31 | struct cgroup; |
ddbcc7e8 PM |
32 | |
33 | extern int cgroup_init_early(void); | |
34 | extern int cgroup_init(void); | |
b4f48b63 | 35 | extern void cgroup_fork(struct task_struct *p); |
817929ec | 36 | extern void cgroup_post_fork(struct task_struct *p); |
1ec41830 | 37 | extern void cgroup_exit(struct task_struct *p); |
846c7bb0 BS |
38 | extern int cgroupstats_build(struct cgroupstats *stats, |
39 | struct dentry *dentry); | |
ddbcc7e8 | 40 | |
8d8b97ba | 41 | extern int proc_cgroup_show(struct seq_file *, void *); |
a424316c | 42 | |
3ed80a62 | 43 | /* define the enumeration of all cgroup subsystems */ |
073219e9 | 44 | #define SUBSYS(_x) _x ## _cgrp_id, |
817929ec PM |
45 | enum cgroup_subsys_id { |
46 | #include <linux/cgroup_subsys.h> | |
a6f00298 | 47 | CGROUP_SUBSYS_COUNT, |
817929ec PM |
48 | }; |
49 | #undef SUBSYS | |
50 | ||
5c9d535b TH |
51 | /* |
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 | */ | |
ddbcc7e8 | 58 | struct cgroup_subsys_state { |
5c9d535b | 59 | /* PI: the cgroup that this css is attached to */ |
ddbcc7e8 PM |
60 | struct cgroup *cgroup; |
61 | ||
5c9d535b | 62 | /* PI: the cgroup subsystem that this css is attached to */ |
72c97e54 TH |
63 | struct cgroup_subsys *ss; |
64 | ||
d3daf28d TH |
65 | /* reference count - access via css_[try]get() and css_put() */ |
66 | struct percpu_ref refcnt; | |
ddbcc7e8 | 67 | |
5c9d535b | 68 | /* PI: the parent css */ |
0ae78e0b TH |
69 | struct cgroup_subsys_state *parent; |
70 | ||
d5c419b6 TH |
71 | /* siblings list anchored at the parent's ->children */ |
72 | struct list_head sibling; | |
73 | struct list_head children; | |
74 | ||
15a4c835 | 75 | /* |
5c9d535b | 76 | * PI: Subsys-unique ID. 0 is unused and root is always 1. The |
15a4c835 TH |
77 | * matching css can be looked up using css_from_id(). |
78 | */ | |
79 | int id; | |
80 | ||
69dfa00c | 81 | unsigned int flags; |
48ddbe19 | 82 | |
0cb51d71 TH |
83 | /* |
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 | ||
0c21ead1 TH |
91 | /* percpu_ref killing and RCU release */ |
92 | struct rcu_head rcu_head; | |
35ef10da | 93 | struct work_struct destroy_work; |
ddbcc7e8 PM |
94 | }; |
95 | ||
96 | /* bits in struct cgroup_subsys_state flags field */ | |
97 | enum { | |
3b514d24 | 98 | CSS_NO_REF = (1 << 0), /* no reference counting for this css */ |
92fb9748 | 99 | CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ |
de3f0341 | 100 | CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */ |
ddbcc7e8 PM |
101 | }; |
102 | ||
5de0107e TH |
103 | /** |
104 | * css_get - obtain a reference on the specified css | |
105 | * @css: target css | |
106 | * | |
107 | * The caller must already have a reference. | |
ddbcc7e8 | 108 | */ |
ddbcc7e8 PM |
109 | static inline void css_get(struct cgroup_subsys_state *css) |
110 | { | |
3b514d24 TH |
111 | if (!(css->flags & CSS_NO_REF)) |
112 | percpu_ref_get(&css->refcnt); | |
ddbcc7e8 | 113 | } |
e7c5ec91 | 114 | |
6f4524d3 TH |
115 | /** |
116 | * css_tryget - try to obtain a reference on the specified css | |
117 | * @css: target css | |
118 | * | |
119 | * Obtain a reference on @css unless it already has reached zero and is | |
120 | * being released. This function doesn't care whether @css is on or | |
121 | * offline. The caller naturally needs to ensure that @css is accessible | |
122 | * but doesn't have to be holding a reference on it - IOW, RCU protected | |
123 | * access is good enough for this function. Returns %true if a reference | |
124 | * count was successfully obtained; %false otherwise. | |
125 | */ | |
126 | static inline bool css_tryget(struct cgroup_subsys_state *css) | |
127 | { | |
128 | if (!(css->flags & CSS_NO_REF)) | |
129 | return percpu_ref_tryget(&css->refcnt); | |
130 | return true; | |
131 | } | |
132 | ||
5de0107e | 133 | /** |
ec903c0c | 134 | * css_tryget_online - try to obtain a reference on the specified css if online |
5de0107e TH |
135 | * @css: target css |
136 | * | |
ec903c0c TH |
137 | * Obtain a reference on @css if it's online. The caller naturally needs |
138 | * to ensure that @css is accessible but doesn't have to be holding a | |
5de0107e TH |
139 | * reference on it - IOW, RCU protected access is good enough for this |
140 | * function. Returns %true if a reference count was successfully obtained; | |
141 | * %false otherwise. | |
142 | */ | |
ec903c0c | 143 | static inline bool css_tryget_online(struct cgroup_subsys_state *css) |
e7c5ec91 | 144 | { |
3b514d24 TH |
145 | if (!(css->flags & CSS_NO_REF)) |
146 | return percpu_ref_tryget_live(&css->refcnt); | |
147 | return true; | |
e7c5ec91 PM |
148 | } |
149 | ||
5de0107e TH |
150 | /** |
151 | * css_put - put a css reference | |
152 | * @css: target css | |
153 | * | |
ec903c0c | 154 | * Put a reference obtained via css_get() and css_tryget_online(). |
5de0107e | 155 | */ |
ddbcc7e8 PM |
156 | static inline void css_put(struct cgroup_subsys_state *css) |
157 | { | |
3b514d24 TH |
158 | if (!(css->flags & CSS_NO_REF)) |
159 | percpu_ref_put(&css->refcnt); | |
ddbcc7e8 PM |
160 | } |
161 | ||
3116f0e3 PM |
162 | /* bits in struct cgroup flags field */ |
163 | enum { | |
d20a390a PM |
164 | /* |
165 | * Control Group has previously had a child cgroup or a task, | |
166 | * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) | |
167 | */ | |
3116f0e3 PM |
168 | CGRP_RELEASABLE, |
169 | /* Control Group requires release notifications to userspace */ | |
170 | CGRP_NOTIFY_ON_RELEASE, | |
97978e6d | 171 | /* |
2260e7fc TH |
172 | * Clone the parent's configuration when creating a new child |
173 | * cpuset cgroup. For historical reasons, this option can be | |
174 | * specified at mount time and thus is implemented here. | |
97978e6d | 175 | */ |
2260e7fc | 176 | CGRP_CPUSET_CLONE_CHILDREN, |
3116f0e3 PM |
177 | }; |
178 | ||
ddbcc7e8 | 179 | struct cgroup { |
9d800df1 TH |
180 | /* self css with NULL ->ss, points back to this cgroup */ |
181 | struct cgroup_subsys_state self; | |
182 | ||
ddbcc7e8 PM |
183 | unsigned long flags; /* "unsigned long" so bitops work */ |
184 | ||
b414dc09 LZ |
185 | /* |
186 | * idr allocated in-hierarchy ID. | |
187 | * | |
7d699ddb TH |
188 | * ID 0 is not used, the ID of the root cgroup is always 1, and a |
189 | * new cgroup will be assigned with a smallest available ID. | |
0ab02ca8 LZ |
190 | * |
191 | * Allocating/Removing ID must be protected by cgroup_mutex. | |
b414dc09 LZ |
192 | */ |
193 | int id; | |
0a950f65 | 194 | |
842b597e TH |
195 | /* |
196 | * If this cgroup contains any tasks, it contributes one to | |
197 | * populated_cnt. All children with non-zero popuplated_cnt of | |
198 | * their own contribute one. The count is zero iff there's no task | |
199 | * in this cgroup or its subtree. | |
200 | */ | |
201 | int populated_cnt; | |
202 | ||
2bd59d48 | 203 | struct kernfs_node *kn; /* cgroup kernfs entry */ |
842b597e | 204 | struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */ |
ddbcc7e8 | 205 | |
667c2491 TH |
206 | /* |
207 | * The bitmask of subsystems enabled on the child cgroups. | |
208 | * ->subtree_control is the one configured through | |
209 | * "cgroup.subtree_control" while ->child_subsys_mask is the | |
210 | * effective one which may have more subsystems enabled. | |
f63070d3 TH |
211 | * Controller knobs are made available iff it's enabled in |
212 | * ->subtree_control. | |
667c2491 TH |
213 | */ |
214 | unsigned int subtree_control; | |
69dfa00c | 215 | unsigned int child_subsys_mask; |
94419627 | 216 | |
ddbcc7e8 | 217 | /* Private pointers for each registered subsystem */ |
73e80ed8 | 218 | struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; |
ddbcc7e8 | 219 | |
3dd06ffa | 220 | struct cgroup_root *root; |
817929ec PM |
221 | |
222 | /* | |
69d0206c TH |
223 | * List of cgrp_cset_links pointing at css_sets with tasks in this |
224 | * cgroup. Protected by css_set_lock. | |
817929ec | 225 | */ |
69d0206c | 226 | struct list_head cset_links; |
81a6a5cd | 227 | |
2d8f243a TH |
228 | /* |
229 | * On the default hierarchy, a css_set for a cgroup with some | |
230 | * susbsys disabled will point to css's which are associated with | |
231 | * the closest ancestor which has the subsys enabled. The | |
232 | * following lists all css_sets which point to this cgroup's css | |
233 | * for the given subsystem. | |
234 | */ | |
235 | struct list_head e_csets[CGROUP_SUBSYS_COUNT]; | |
236 | ||
81a6a5cd PM |
237 | /* |
238 | * Linked list running through all cgroups that can | |
239 | * potentially be reaped by the release agent. Protected by | |
240 | * release_list_lock | |
241 | */ | |
242 | struct list_head release_list; | |
cc31edce | 243 | |
72a8cb30 BB |
244 | /* |
245 | * list of pidlists, up to two for each namespace (one for procs, one | |
246 | * for tasks); created on demand. | |
247 | */ | |
248 | struct list_head pidlists; | |
249 | struct mutex pidlist_mutex; | |
a47295e6 | 250 | |
f8f22e53 TH |
251 | /* used to wait for offlining of csses */ |
252 | wait_queue_head_t offline_waitq; | |
817929ec PM |
253 | }; |
254 | ||
25a7e684 TH |
255 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
256 | ||
3dd06ffa | 257 | /* cgroup_root->flags */ |
25a7e684 | 258 | enum { |
aa6ec29b | 259 | CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */ |
25a7e684 TH |
260 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
261 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ | |
262 | }; | |
263 | ||
264 | /* | |
3dd06ffa | 265 | * A cgroup_root represents the root of a cgroup hierarchy, and may be |
2bd59d48 | 266 | * associated with a kernfs_root to form an active hierarchy. This is |
25a7e684 TH |
267 | * internal to cgroup core. Don't access directly from controllers. |
268 | */ | |
3dd06ffa | 269 | struct cgroup_root { |
2bd59d48 | 270 | struct kernfs_root *kf_root; |
25a7e684 | 271 | |
f392e51c | 272 | /* The bitmask of subsystems attached to this hierarchy */ |
69dfa00c | 273 | unsigned int subsys_mask; |
f392e51c | 274 | |
25a7e684 TH |
275 | /* Unique id for this hierarchy. */ |
276 | int hierarchy_id; | |
277 | ||
776f02fa | 278 | /* The root cgroup. Root is destroyed on its release. */ |
3dd06ffa | 279 | struct cgroup cgrp; |
25a7e684 | 280 | |
3c9c825b TH |
281 | /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ |
282 | atomic_t nr_cgrps; | |
25a7e684 TH |
283 | |
284 | /* A list running through the active hierarchies */ | |
285 | struct list_head root_list; | |
286 | ||
25a7e684 | 287 | /* Hierarchy-specific flags */ |
69dfa00c | 288 | unsigned int flags; |
25a7e684 TH |
289 | |
290 | /* IDs for cgroups in this hierarchy */ | |
4e96ee8e | 291 | struct idr cgroup_idr; |
25a7e684 TH |
292 | |
293 | /* The path to use for release notifications. */ | |
294 | char release_agent_path[PATH_MAX]; | |
295 | ||
296 | /* The name for this hierarchy - may be empty */ | |
297 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
298 | }; | |
299 | ||
d20a390a PM |
300 | /* |
301 | * A css_set is a structure holding pointers to a set of | |
817929ec PM |
302 | * cgroup_subsys_state objects. This saves space in the task struct |
303 | * object and speeds up fork()/exit(), since a single inc/dec and a | |
d20a390a PM |
304 | * list_add()/del() can bump the reference count on the entire cgroup |
305 | * set for a task. | |
817929ec PM |
306 | */ |
307 | ||
308 | struct css_set { | |
309 | ||
310 | /* Reference count */ | |
146aa1bd | 311 | atomic_t refcount; |
817929ec | 312 | |
472b1053 LZ |
313 | /* |
314 | * List running through all cgroup groups in the same hash | |
315 | * slot. Protected by css_set_lock | |
316 | */ | |
317 | struct hlist_node hlist; | |
318 | ||
817929ec | 319 | /* |
c7561128 TH |
320 | * Lists running through all tasks using this cgroup group. |
321 | * mg_tasks lists tasks which belong to this cset but are in the | |
322 | * process of being migrated out or in. Protected by | |
323 | * css_set_rwsem, but, during migration, once tasks are moved to | |
324 | * mg_tasks, it can be read safely while holding cgroup_mutex. | |
817929ec PM |
325 | */ |
326 | struct list_head tasks; | |
c7561128 | 327 | struct list_head mg_tasks; |
817929ec PM |
328 | |
329 | /* | |
69d0206c TH |
330 | * List of cgrp_cset_links pointing at cgroups referenced from this |
331 | * css_set. Protected by css_set_lock. | |
817929ec | 332 | */ |
69d0206c | 333 | struct list_head cgrp_links; |
817929ec | 334 | |
6803c006 TH |
335 | /* the default cgroup associated with this css_set */ |
336 | struct cgroup *dfl_cgrp; | |
337 | ||
817929ec | 338 | /* |
3ed80a62 TH |
339 | * Set of subsystem states, one for each subsystem. This array is |
340 | * immutable after creation apart from the init_css_set during | |
341 | * subsystem registration (at boot time). | |
817929ec PM |
342 | */ |
343 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; | |
c378369d | 344 | |
b3dc094e TH |
345 | /* |
346 | * List of csets participating in the on-going migration either as | |
347 | * source or destination. Protected by cgroup_mutex. | |
348 | */ | |
1958d2d5 | 349 | struct list_head mg_preload_node; |
b3dc094e TH |
350 | struct list_head mg_node; |
351 | ||
352 | /* | |
353 | * If this cset is acting as the source of migration the following | |
354 | * two fields are set. mg_src_cgrp is the source cgroup of the | |
355 | * on-going migration and mg_dst_cset is the destination cset the | |
356 | * target tasks on this cset should be migrated to. Protected by | |
357 | * cgroup_mutex. | |
358 | */ | |
359 | struct cgroup *mg_src_cgrp; | |
360 | struct css_set *mg_dst_cset; | |
361 | ||
2d8f243a TH |
362 | /* |
363 | * On the default hierarhcy, ->subsys[ssid] may point to a css | |
364 | * attached to an ancestor instead of the cgroup this css_set is | |
365 | * associated with. The following node is anchored at | |
366 | * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to | |
367 | * iterate through all css's attached to a given cgroup. | |
368 | */ | |
369 | struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; | |
370 | ||
c378369d BB |
371 | /* For RCU-protected deletion */ |
372 | struct rcu_head rcu_head; | |
ddbcc7e8 PM |
373 | }; |
374 | ||
d20a390a PM |
375 | /* |
376 | * struct cftype: handler definitions for cgroup control files | |
ddbcc7e8 PM |
377 | * |
378 | * When reading/writing to a file: | |
a043e3b2 | 379 | * - the cgroup to use is file->f_dentry->d_parent->d_fsdata |
ddbcc7e8 PM |
380 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
381 | */ | |
382 | ||
8e3f6541 | 383 | /* cftype->flags */ |
02c402d9 | 384 | enum { |
6f4b7e63 LZ |
385 | CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ |
386 | CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ | |
02c402d9 | 387 | CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */ |
9fa4db33 | 388 | CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ |
8cbbf2c9 | 389 | CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */ |
02c402d9 | 390 | }; |
8e3f6541 TH |
391 | |
392 | #define MAX_CFTYPE_NAME 64 | |
393 | ||
ddbcc7e8 | 394 | struct cftype { |
d20a390a PM |
395 | /* |
396 | * By convention, the name should begin with the name of the | |
8e3f6541 TH |
397 | * subsystem, followed by a period. Zero length string indicates |
398 | * end of cftype array. | |
d20a390a | 399 | */ |
ddbcc7e8 PM |
400 | char name[MAX_CFTYPE_NAME]; |
401 | int private; | |
099fca32 LZ |
402 | /* |
403 | * If not 0, file mode is set to this value, otherwise it will | |
404 | * be figured out automatically | |
405 | */ | |
a5e7ed32 | 406 | umode_t mode; |
db3b1497 PM |
407 | |
408 | /* | |
5f469907 | 409 | * The maximum length of string, excluding trailing nul, that can |
451af504 | 410 | * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. |
db3b1497 PM |
411 | */ |
412 | size_t max_write_len; | |
413 | ||
8e3f6541 TH |
414 | /* CFTYPE_* flags */ |
415 | unsigned int flags; | |
416 | ||
2bb566cb | 417 | /* |
0adb0704 TH |
418 | * Fields used for internal bookkeeping. Initialized automatically |
419 | * during registration. | |
2bb566cb | 420 | */ |
0adb0704 TH |
421 | struct cgroup_subsys *ss; /* NULL for cgroup core files */ |
422 | struct list_head node; /* anchored at ss->cfts */ | |
2bd59d48 TH |
423 | struct kernfs_ops *kf_ops; |
424 | ||
ddbcc7e8 | 425 | /* |
f4c753b7 | 426 | * read_u64() is a shortcut for the common case of returning a |
ddbcc7e8 PM |
427 | * single integer. Use it in place of read() |
428 | */ | |
182446d0 | 429 | u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); |
e73d2c61 PM |
430 | /* |
431 | * read_s64() is a signed version of read_u64() | |
432 | */ | |
182446d0 | 433 | s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); |
2da8ca82 TH |
434 | |
435 | /* generic seq_file read interface */ | |
436 | int (*seq_show)(struct seq_file *sf, void *v); | |
91796569 | 437 | |
6612f05b TH |
438 | /* optional ops, implement all or none */ |
439 | void *(*seq_start)(struct seq_file *sf, loff_t *ppos); | |
440 | void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); | |
441 | void (*seq_stop)(struct seq_file *sf, void *v); | |
442 | ||
355e0c48 | 443 | /* |
f4c753b7 | 444 | * write_u64() is a shortcut for the common case of accepting |
355e0c48 PM |
445 | * a single integer (as parsed by simple_strtoull) from |
446 | * userspace. Use in place of write(); return 0 or error. | |
447 | */ | |
182446d0 TH |
448 | int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, |
449 | u64 val); | |
e73d2c61 PM |
450 | /* |
451 | * write_s64() is a signed version of write_u64() | |
452 | */ | |
182446d0 TH |
453 | int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, |
454 | s64 val); | |
355e0c48 | 455 | |
b4168640 TH |
456 | /* |
457 | * write() is the generic write callback which maps directly to | |
458 | * kernfs write operation and overrides all other operations. | |
459 | * Maximum write size is determined by ->max_write_len. Use | |
460 | * of_css/cft() to access the associated css and cft. | |
461 | */ | |
462 | ssize_t (*write)(struct kernfs_open_file *of, | |
463 | char *buf, size_t nbytes, loff_t off); | |
464 | ||
2bd59d48 TH |
465 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
466 | struct lock_class_key lockdep_key; | |
467 | #endif | |
ddbcc7e8 PM |
468 | }; |
469 | ||
a2dd4247 | 470 | extern struct cgroup_root cgrp_dfl_root; |
5024ae29 | 471 | extern struct css_set init_css_set; |
a2dd4247 | 472 | |
aa6ec29b TH |
473 | /** |
474 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
475 | * @cgrp: the cgroup of interest | |
476 | * | |
477 | * The default hierarchy is the v2 interface of cgroup and this function | |
478 | * can be used to test whether a cgroup is on the default hierarchy for | |
479 | * cases where a subsystem should behave differnetly depending on the | |
480 | * interface version. | |
481 | * | |
482 | * The set of behaviors which change on the default hierarchy are still | |
483 | * being determined and the mount option is prefixed with __DEVEL__. | |
484 | * | |
485 | * List of changed behaviors: | |
486 | * | |
487 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
488 | * and "name" are disallowed. | |
489 | * | |
490 | * - When mounting an existing superblock, mount options should match. | |
491 | * | |
492 | * - Remount is disallowed. | |
493 | * | |
494 | * - rename(2) is disallowed. | |
495 | * | |
496 | * - "tasks" is removed. Everything should be at process granularity. Use | |
497 | * "cgroup.procs" instead. | |
498 | * | |
499 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
500 | * recycled inbetween reads. | |
501 | * | |
502 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
503 | * notification mechanism will be implemented. | |
504 | * | |
505 | * - "cgroup.clone_children" is removed. | |
506 | * | |
507 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
508 | * and its descendants contain no task; otherwise, 1. The file also | |
509 | * generates kernfs notification which can be monitored through poll and | |
510 | * [di]notify when the value of the file changes. | |
511 | * | |
512 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
513 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
514 | * moved to an ancestor. | |
515 | * | |
516 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
517 | * masks of ancestors. | |
518 | * | |
519 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
520 | * is not created. | |
521 | * | |
522 | * - blkcg: blk-throttle becomes properly hierarchical. | |
523 | * | |
524 | * - debug: disallowed on the default hierarchy. | |
525 | */ | |
a2dd4247 TH |
526 | static inline bool cgroup_on_dfl(const struct cgroup *cgrp) |
527 | { | |
528 | return cgrp->root == &cgrp_dfl_root; | |
529 | } | |
530 | ||
07bc356e TH |
531 | /* no synchronization, the result can only be used as a hint */ |
532 | static inline bool cgroup_has_tasks(struct cgroup *cgrp) | |
533 | { | |
534 | return !list_empty(&cgrp->cset_links); | |
535 | } | |
536 | ||
b1664924 TH |
537 | /* returns ino associated with a cgroup, 0 indicates unmounted root */ |
538 | static inline ino_t cgroup_ino(struct cgroup *cgrp) | |
539 | { | |
2bd59d48 TH |
540 | if (cgrp->kn) |
541 | return cgrp->kn->ino; | |
b1664924 TH |
542 | else |
543 | return 0; | |
544 | } | |
545 | ||
b4168640 TH |
546 | /* cft/css accessors for cftype->write() operation */ |
547 | static inline struct cftype *of_cft(struct kernfs_open_file *of) | |
7da11279 | 548 | { |
2bd59d48 | 549 | return of->kn->priv; |
7da11279 TH |
550 | } |
551 | ||
b4168640 TH |
552 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of); |
553 | ||
554 | /* cft/css accessors for cftype->seq_*() operations */ | |
555 | static inline struct cftype *seq_cft(struct seq_file *seq) | |
556 | { | |
557 | return of_cft(seq->private); | |
558 | } | |
559 | ||
560 | static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq) | |
561 | { | |
562 | return of_css(seq->private); | |
563 | } | |
59f5296b | 564 | |
e61734c5 TH |
565 | /* |
566 | * Name / path handling functions. All are thin wrappers around the kernfs | |
567 | * counterparts and can be called under any context. | |
568 | */ | |
569 | ||
570 | static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen) | |
571 | { | |
fdce6bf8 | 572 | return kernfs_name(cgrp->kn, buf, buflen); |
e61734c5 TH |
573 | } |
574 | ||
575 | static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf, | |
576 | size_t buflen) | |
577 | { | |
fdce6bf8 | 578 | return kernfs_path(cgrp->kn, buf, buflen); |
e61734c5 TH |
579 | } |
580 | ||
581 | static inline void pr_cont_cgroup_name(struct cgroup *cgrp) | |
582 | { | |
fdce6bf8 | 583 | pr_cont_kernfs_name(cgrp->kn); |
e61734c5 TH |
584 | } |
585 | ||
586 | static inline void pr_cont_cgroup_path(struct cgroup *cgrp) | |
587 | { | |
fdce6bf8 | 588 | pr_cont_kernfs_path(cgrp->kn); |
e61734c5 TH |
589 | } |
590 | ||
591 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen); | |
592 | ||
2cf669a5 | 593 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
2bb566cb | 594 | int cgroup_rm_cftypes(struct cftype *cfts); |
8e3f6541 | 595 | |
78574cf9 | 596 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor); |
ddbcc7e8 | 597 | |
2f7ee569 TH |
598 | /* |
599 | * Control Group taskset, used to pass around set of tasks to cgroup_subsys | |
600 | * methods. | |
601 | */ | |
602 | struct cgroup_taskset; | |
603 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset); | |
604 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset); | |
2f7ee569 TH |
605 | |
606 | /** | |
607 | * cgroup_taskset_for_each - iterate cgroup_taskset | |
608 | * @task: the loop cursor | |
2f7ee569 TH |
609 | * @tset: taskset to iterate |
610 | */ | |
924f0d9a | 611 | #define cgroup_taskset_for_each(task, tset) \ |
2f7ee569 | 612 | for ((task) = cgroup_taskset_first((tset)); (task); \ |
924f0d9a | 613 | (task) = cgroup_taskset_next((tset))) |
2f7ee569 | 614 | |
21acb9ca TLSC |
615 | /* |
616 | * Control Group subsystem type. | |
617 | * See Documentation/cgroups/cgroups.txt for details | |
618 | */ | |
ddbcc7e8 PM |
619 | |
620 | struct cgroup_subsys { | |
eb95419b TH |
621 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); |
622 | int (*css_online)(struct cgroup_subsys_state *css); | |
623 | void (*css_offline)(struct cgroup_subsys_state *css); | |
624 | void (*css_free)(struct cgroup_subsys_state *css); | |
b4536f0c | 625 | void (*css_reset)(struct cgroup_subsys_state *css); |
eb95419b TH |
626 | |
627 | int (*can_attach)(struct cgroup_subsys_state *css, | |
628 | struct cgroup_taskset *tset); | |
629 | void (*cancel_attach)(struct cgroup_subsys_state *css, | |
630 | struct cgroup_taskset *tset); | |
631 | void (*attach)(struct cgroup_subsys_state *css, | |
632 | struct cgroup_taskset *tset); | |
761b3ef5 | 633 | void (*fork)(struct task_struct *task); |
eb95419b TH |
634 | void (*exit)(struct cgroup_subsys_state *css, |
635 | struct cgroup_subsys_state *old_css, | |
761b3ef5 | 636 | struct task_struct *task); |
eb95419b | 637 | void (*bind)(struct cgroup_subsys_state *root_css); |
e5991371 | 638 | |
8bab8dde | 639 | int disabled; |
ddbcc7e8 | 640 | int early_init; |
48ddbe19 | 641 | |
8c7f6edb TH |
642 | /* |
643 | * If %false, this subsystem is properly hierarchical - | |
644 | * configuration, resource accounting and restriction on a parent | |
645 | * cgroup cover those of its children. If %true, hierarchy support | |
646 | * is broken in some ways - some subsystems ignore hierarchy | |
647 | * completely while others are only implemented half-way. | |
648 | * | |
649 | * It's now disallowed to create nested cgroups if the subsystem is | |
650 | * broken and cgroup core will emit a warning message on such | |
651 | * cases. Eventually, all subsystems will be made properly | |
652 | * hierarchical and this will go away. | |
653 | */ | |
654 | bool broken_hierarchy; | |
655 | bool warned_broken_hierarchy; | |
656 | ||
073219e9 | 657 | /* the following two fields are initialized automtically during boot */ |
aec25020 | 658 | int id; |
ddbcc7e8 PM |
659 | #define MAX_CGROUP_TYPE_NAMELEN 32 |
660 | const char *name; | |
661 | ||
b85d2040 | 662 | /* link to parent, protected by cgroup_lock() */ |
3dd06ffa | 663 | struct cgroup_root *root; |
e6a1105b | 664 | |
15a4c835 TH |
665 | /* idr for css->id */ |
666 | struct idr css_idr; | |
667 | ||
0adb0704 TH |
668 | /* |
669 | * List of cftypes. Each entry is the first entry of an array | |
670 | * terminated by zero length name. | |
671 | */ | |
672 | struct list_head cfts; | |
8e3f6541 | 673 | |
de00ffa5 | 674 | /* base cftypes, automatically registered with subsys itself */ |
5577964e | 675 | struct cftype *legacy_cftypes; /* used on the legacy hierarchies */ |
af0ba678 TH |
676 | |
677 | /* | |
678 | * A subsystem may depend on other subsystems. When such subsystem | |
679 | * is enabled on a cgroup, the depended-upon subsystems are enabled | |
680 | * together if available. Subsystems enabled due to dependency are | |
681 | * not visible to userland until explicitly enabled. The following | |
682 | * specifies the mask of subsystems that this one depends on. | |
683 | */ | |
684 | unsigned int depends_on; | |
ddbcc7e8 PM |
685 | }; |
686 | ||
073219e9 | 687 | #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys; |
ddbcc7e8 PM |
688 | #include <linux/cgroup_subsys.h> |
689 | #undef SUBSYS | |
690 | ||
14611e51 TH |
691 | /** |
692 | * task_css_set_check - obtain a task's css_set with extra access conditions | |
693 | * @task: the task to obtain css_set for | |
694 | * @__c: extra condition expression to be passed to rcu_dereference_check() | |
695 | * | |
696 | * A task's css_set is RCU protected, initialized and exited while holding | |
697 | * task_lock(), and can only be modified while holding both cgroup_mutex | |
698 | * and task_lock() while the task is alive. This macro verifies that the | |
699 | * caller is inside proper critical section and returns @task's css_set. | |
700 | * | |
701 | * The caller can also specify additional allowed conditions via @__c, such | |
702 | * as locks used during the cgroup_subsys::attach() methods. | |
dc61b1d6 | 703 | */ |
2219449a TH |
704 | #ifdef CONFIG_PROVE_RCU |
705 | extern struct mutex cgroup_mutex; | |
0e1d768f | 706 | extern struct rw_semaphore css_set_rwsem; |
14611e51 TH |
707 | #define task_css_set_check(task, __c) \ |
708 | rcu_dereference_check((task)->cgroups, \ | |
0e1d768f TH |
709 | lockdep_is_held(&cgroup_mutex) || \ |
710 | lockdep_is_held(&css_set_rwsem) || \ | |
711 | ((task)->flags & PF_EXITING) || (__c)) | |
2219449a | 712 | #else |
14611e51 TH |
713 | #define task_css_set_check(task, __c) \ |
714 | rcu_dereference((task)->cgroups) | |
2219449a | 715 | #endif |
dc61b1d6 | 716 | |
14611e51 | 717 | /** |
8af01f56 | 718 | * task_css_check - obtain css for (task, subsys) w/ extra access conds |
14611e51 TH |
719 | * @task: the target task |
720 | * @subsys_id: the target subsystem ID | |
721 | * @__c: extra condition expression to be passed to rcu_dereference_check() | |
722 | * | |
723 | * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The | |
724 | * synchronization rules are the same as task_css_set_check(). | |
725 | */ | |
8af01f56 | 726 | #define task_css_check(task, subsys_id, __c) \ |
14611e51 TH |
727 | task_css_set_check((task), (__c))->subsys[(subsys_id)] |
728 | ||
729 | /** | |
730 | * task_css_set - obtain a task's css_set | |
731 | * @task: the task to obtain css_set for | |
732 | * | |
733 | * See task_css_set_check(). | |
734 | */ | |
735 | static inline struct css_set *task_css_set(struct task_struct *task) | |
736 | { | |
737 | return task_css_set_check(task, false); | |
738 | } | |
739 | ||
740 | /** | |
8af01f56 | 741 | * task_css - obtain css for (task, subsys) |
14611e51 TH |
742 | * @task: the target task |
743 | * @subsys_id: the target subsystem ID | |
744 | * | |
8af01f56 | 745 | * See task_css_check(). |
14611e51 | 746 | */ |
8af01f56 TH |
747 | static inline struct cgroup_subsys_state *task_css(struct task_struct *task, |
748 | int subsys_id) | |
ddbcc7e8 | 749 | { |
8af01f56 | 750 | return task_css_check(task, subsys_id, false); |
ddbcc7e8 PM |
751 | } |
752 | ||
5024ae29 TH |
753 | /** |
754 | * task_css_is_root - test whether a task belongs to the root css | |
755 | * @task: the target task | |
756 | * @subsys_id: the target subsystem ID | |
757 | * | |
758 | * Test whether @task belongs to the root css on the specified subsystem. | |
759 | * May be invoked in any context. | |
760 | */ | |
761 | static inline bool task_css_is_root(struct task_struct *task, int subsys_id) | |
762 | { | |
763 | return task_css_check(task, subsys_id, true) == | |
764 | init_css_set.subsys[subsys_id]; | |
765 | } | |
766 | ||
8af01f56 TH |
767 | static inline struct cgroup *task_cgroup(struct task_struct *task, |
768 | int subsys_id) | |
ddbcc7e8 | 769 | { |
8af01f56 | 770 | return task_css(task, subsys_id)->cgroup; |
ddbcc7e8 PM |
771 | } |
772 | ||
492eb21b TH |
773 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
774 | struct cgroup_subsys_state *parent); | |
53fa5261 | 775 | |
1cb650b9 LZ |
776 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss); |
777 | ||
574bd9f7 | 778 | /** |
492eb21b TH |
779 | * css_for_each_child - iterate through children of a css |
780 | * @pos: the css * to use as the loop cursor | |
781 | * @parent: css whose children to walk | |
574bd9f7 | 782 | * |
c2931b70 | 783 | * Walk @parent's children. Must be called under rcu_read_lock(). |
574bd9f7 | 784 | * |
c2931b70 TH |
785 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
786 | * css which finished ->css_online() is guaranteed to be visible in the | |
787 | * future iterations and will stay visible until the last reference is put. | |
788 | * A css which hasn't finished ->css_online() or already finished | |
789 | * ->css_offline() may show up during traversal. It's each subsystem's | |
790 | * responsibility to synchronize against on/offlining. | |
75501a6d TH |
791 | * |
792 | * It is allowed to temporarily drop RCU read lock during iteration. The | |
793 | * caller is responsible for ensuring that @pos remains accessible until | |
794 | * the start of the next iteration by, for example, bumping the css refcnt. | |
574bd9f7 | 795 | */ |
492eb21b TH |
796 | #define css_for_each_child(pos, parent) \ |
797 | for ((pos) = css_next_child(NULL, (parent)); (pos); \ | |
798 | (pos) = css_next_child((pos), (parent))) | |
574bd9f7 | 799 | |
492eb21b TH |
800 | struct cgroup_subsys_state * |
801 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
802 | struct cgroup_subsys_state *css); | |
803 | ||
804 | struct cgroup_subsys_state * | |
805 | css_rightmost_descendant(struct cgroup_subsys_state *pos); | |
574bd9f7 TH |
806 | |
807 | /** | |
492eb21b TH |
808 | * css_for_each_descendant_pre - pre-order walk of a css's descendants |
809 | * @pos: the css * to use as the loop cursor | |
810 | * @root: css whose descendants to walk | |
574bd9f7 | 811 | * |
bd8815a6 | 812 | * Walk @root's descendants. @root is included in the iteration and the |
c2931b70 | 813 | * first node to be visited. Must be called under rcu_read_lock(). |
574bd9f7 | 814 | * |
c2931b70 TH |
815 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
816 | * css which finished ->css_online() is guaranteed to be visible in the | |
817 | * future iterations and will stay visible until the last reference is put. | |
818 | * A css which hasn't finished ->css_online() or already finished | |
819 | * ->css_offline() may show up during traversal. It's each subsystem's | |
820 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 821 | * |
c2931b70 | 822 | * For example, the following guarantees that a descendant can't escape |
574bd9f7 TH |
823 | * state updates of its ancestors. |
824 | * | |
492eb21b | 825 | * my_online(@css) |
574bd9f7 | 826 | * { |
492eb21b TH |
827 | * Lock @css's parent and @css; |
828 | * Inherit state from the parent; | |
574bd9f7 TH |
829 | * Unlock both. |
830 | * } | |
831 | * | |
492eb21b | 832 | * my_update_state(@css) |
574bd9f7 | 833 | * { |
492eb21b | 834 | * css_for_each_descendant_pre(@pos, @css) { |
574bd9f7 | 835 | * Lock @pos; |
bd8815a6 TH |
836 | * if (@pos == @css) |
837 | * Update @css's state; | |
838 | * else | |
839 | * Verify @pos is alive and inherit state from its parent; | |
574bd9f7 TH |
840 | * Unlock @pos; |
841 | * } | |
842 | * } | |
843 | * | |
844 | * As long as the inheriting step, including checking the parent state, is | |
845 | * enclosed inside @pos locking, double-locking the parent isn't necessary | |
846 | * while inheriting. The state update to the parent is guaranteed to be | |
847 | * visible by walking order and, as long as inheriting operations to the | |
848 | * same @pos are atomic to each other, multiple updates racing each other | |
849 | * still result in the correct state. It's guaranateed that at least one | |
492eb21b | 850 | * inheritance happens for any css after the latest update to its parent. |
574bd9f7 TH |
851 | * |
852 | * If checking parent's state requires locking the parent, each inheriting | |
853 | * iteration should lock and unlock both @pos->parent and @pos. | |
854 | * | |
855 | * Alternatively, a subsystem may choose to use a single global lock to | |
92fb9748 | 856 | * synchronize ->css_online() and ->css_offline() against tree-walking |
574bd9f7 | 857 | * operations. |
75501a6d TH |
858 | * |
859 | * It is allowed to temporarily drop RCU read lock during iteration. The | |
860 | * caller is responsible for ensuring that @pos remains accessible until | |
861 | * the start of the next iteration by, for example, bumping the css refcnt. | |
574bd9f7 | 862 | */ |
492eb21b TH |
863 | #define css_for_each_descendant_pre(pos, css) \ |
864 | for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \ | |
865 | (pos) = css_next_descendant_pre((pos), (css))) | |
574bd9f7 | 866 | |
492eb21b TH |
867 | struct cgroup_subsys_state * |
868 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
869 | struct cgroup_subsys_state *css); | |
574bd9f7 TH |
870 | |
871 | /** | |
492eb21b TH |
872 | * css_for_each_descendant_post - post-order walk of a css's descendants |
873 | * @pos: the css * to use as the loop cursor | |
874 | * @css: css whose descendants to walk | |
574bd9f7 | 875 | * |
492eb21b | 876 | * Similar to css_for_each_descendant_pre() but performs post-order |
bd8815a6 | 877 | * traversal instead. @root is included in the iteration and the last |
c2931b70 TH |
878 | * node to be visited. |
879 | * | |
880 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
881 | * css which finished ->css_online() is guaranteed to be visible in the | |
882 | * future iterations and will stay visible until the last reference is put. | |
883 | * A css which hasn't finished ->css_online() or already finished | |
884 | * ->css_offline() may show up during traversal. It's each subsystem's | |
885 | * responsibility to synchronize against on/offlining. | |
886 | * | |
887 | * Note that the walk visibility guarantee example described in pre-order | |
888 | * walk doesn't apply the same to post-order walks. | |
574bd9f7 | 889 | */ |
492eb21b TH |
890 | #define css_for_each_descendant_post(pos, css) \ |
891 | for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \ | |
892 | (pos) = css_next_descendant_post((pos), (css))) | |
574bd9f7 | 893 | |
f3d46500 TH |
894 | bool css_has_online_children(struct cgroup_subsys_state *css); |
895 | ||
72ec7029 TH |
896 | /* A css_task_iter should be treated as an opaque object */ |
897 | struct css_task_iter { | |
3ebb2b6e TH |
898 | struct cgroup_subsys *ss; |
899 | ||
0f0a2b4f TH |
900 | struct list_head *cset_pos; |
901 | struct list_head *cset_head; | |
902 | ||
903 | struct list_head *task_pos; | |
904 | struct list_head *tasks_head; | |
905 | struct list_head *mg_tasks_head; | |
817929ec PM |
906 | }; |
907 | ||
72ec7029 TH |
908 | void css_task_iter_start(struct cgroup_subsys_state *css, |
909 | struct css_task_iter *it); | |
910 | struct task_struct *css_task_iter_next(struct css_task_iter *it); | |
911 | void css_task_iter_end(struct css_task_iter *it); | |
e535837b | 912 | |
31583bb0 | 913 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); |
8cc99345 | 914 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from); |
31583bb0 | 915 | |
ec903c0c TH |
916 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
917 | struct cgroup_subsys *ss); | |
38460b48 | 918 | |
ddbcc7e8 PM |
919 | #else /* !CONFIG_CGROUPS */ |
920 | ||
921 | static inline int cgroup_init_early(void) { return 0; } | |
922 | static inline int cgroup_init(void) { return 0; } | |
b4f48b63 | 923 | static inline void cgroup_fork(struct task_struct *p) {} |
817929ec | 924 | static inline void cgroup_post_fork(struct task_struct *p) {} |
1ec41830 | 925 | static inline void cgroup_exit(struct task_struct *p) {} |
ddbcc7e8 | 926 | |
846c7bb0 BS |
927 | static inline int cgroupstats_build(struct cgroupstats *stats, |
928 | struct dentry *dentry) | |
929 | { | |
930 | return -EINVAL; | |
931 | } | |
ddbcc7e8 | 932 | |
d7926ee3 | 933 | /* No cgroups - nothing to do */ |
31583bb0 MT |
934 | static inline int cgroup_attach_task_all(struct task_struct *from, |
935 | struct task_struct *t) | |
936 | { | |
937 | return 0; | |
938 | } | |
d7926ee3 | 939 | |
ddbcc7e8 PM |
940 | #endif /* !CONFIG_CGROUPS */ |
941 | ||
942 | #endif /* _LINUX_CGROUP_H */ |