cgroup.h 23 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765
  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. #include <linux/sched.h>
  11. #include <linux/cpumask.h>
  12. #include <linux/nodemask.h>
  13. #include <linux/rcupdate.h>
  14. #include <linux/rculist.h>
  15. #include <linux/cgroupstats.h>
  16. #include <linux/prio_heap.h>
  17. #include <linux/rwsem.h>
  18. #include <linux/idr.h>
  19. #include <linux/workqueue.h>
  20. #include <linux/xattr.h>
  21. #ifdef CONFIG_CGROUPS
  22. struct cgroupfs_root;
  23. struct cgroup_subsys;
  24. struct inode;
  25. struct cgroup;
  26. struct css_id;
  27. extern int cgroup_init_early(void);
  28. extern int cgroup_init(void);
  29. extern void cgroup_lock(void);
  30. extern int cgroup_lock_is_held(void);
  31. extern bool cgroup_lock_live_group(struct cgroup *cgrp);
  32. extern void cgroup_unlock(void);
  33. extern void cgroup_fork(struct task_struct *p);
  34. extern void cgroup_post_fork(struct task_struct *p);
  35. extern void cgroup_exit(struct task_struct *p, int run_callbacks);
  36. extern int cgroupstats_build(struct cgroupstats *stats,
  37. struct dentry *dentry);
  38. extern int cgroup_load_subsys(struct cgroup_subsys *ss);
  39. extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
  40. extern const struct file_operations proc_cgroup_operations;
  41. /*
  42. * Define the enumeration of all cgroup subsystems.
  43. *
  44. * We define ids for builtin subsystems and then modular ones.
  45. */
  46. #define SUBSYS(_x) _x ## _subsys_id,
  47. enum cgroup_subsys_id {
  48. #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
  49. #include <linux/cgroup_subsys.h>
  50. #undef IS_SUBSYS_ENABLED
  51. CGROUP_BUILTIN_SUBSYS_COUNT,
  52. __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1,
  53. #define IS_SUBSYS_ENABLED(option) IS_MODULE(option)
  54. #include <linux/cgroup_subsys.h>
  55. #undef IS_SUBSYS_ENABLED
  56. CGROUP_SUBSYS_COUNT,
  57. };
  58. #undef SUBSYS
  59. /* Per-subsystem/per-cgroup state maintained by the system. */
  60. struct cgroup_subsys_state {
  61. /*
  62. * The cgroup that this subsystem is attached to. Useful
  63. * for subsystems that want to know about the cgroup
  64. * hierarchy structure
  65. */
  66. struct cgroup *cgroup;
  67. /*
  68. * State maintained by the cgroup system to allow subsystems
  69. * to be "busy". Should be accessed via css_get(),
  70. * css_tryget() and css_put().
  71. */
  72. atomic_t refcnt;
  73. unsigned long flags;
  74. /* ID for this css, if possible */
  75. struct css_id __rcu *id;
  76. /* Used to put @cgroup->dentry on the last css_put() */
  77. struct work_struct dput_work;
  78. };
  79. /* bits in struct cgroup_subsys_state flags field */
  80. enum {
  81. CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
  82. CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
  83. };
  84. /* Caller must verify that the css is not for root cgroup */
  85. static inline void __css_get(struct cgroup_subsys_state *css, int count)
  86. {
  87. atomic_add(count, &css->refcnt);
  88. }
  89. /*
  90. * Call css_get() to hold a reference on the css; it can be used
  91. * for a reference obtained via:
  92. * - an existing ref-counted reference to the css
  93. * - task->cgroups for a locked task
  94. */
  95. static inline void css_get(struct cgroup_subsys_state *css)
  96. {
  97. /* We don't need to reference count the root state */
  98. if (!(css->flags & CSS_ROOT))
  99. __css_get(css, 1);
  100. }
  101. /*
  102. * Call css_tryget() to take a reference on a css if your existing
  103. * (known-valid) reference isn't already ref-counted. Returns false if
  104. * the css has been destroyed.
  105. */
  106. extern bool __css_tryget(struct cgroup_subsys_state *css);
  107. static inline bool css_tryget(struct cgroup_subsys_state *css)
  108. {
  109. if (css->flags & CSS_ROOT)
  110. return true;
  111. return __css_tryget(css);
  112. }
  113. /*
  114. * css_put() should be called to release a reference taken by
  115. * css_get() or css_tryget()
  116. */
  117. extern void __css_put(struct cgroup_subsys_state *css);
  118. static inline void css_put(struct cgroup_subsys_state *css)
  119. {
  120. if (!(css->flags & CSS_ROOT))
  121. __css_put(css);
  122. }
  123. /* bits in struct cgroup flags field */
  124. enum {
  125. /* Control Group is dead */
  126. CGRP_REMOVED,
  127. /*
  128. * Control Group has previously had a child cgroup or a task,
  129. * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
  130. */
  131. CGRP_RELEASABLE,
  132. /* Control Group requires release notifications to userspace */
  133. CGRP_NOTIFY_ON_RELEASE,
  134. /*
  135. * Clone the parent's configuration when creating a new child
  136. * cpuset cgroup. For historical reasons, this option can be
  137. * specified at mount time and thus is implemented here.
  138. */
  139. CGRP_CPUSET_CLONE_CHILDREN,
  140. };
  141. struct cgroup_name {
  142. struct rcu_head rcu_head;
  143. char name[];
  144. };
  145. struct cgroup {
  146. unsigned long flags; /* "unsigned long" so bitops work */
  147. /*
  148. * count users of this cgroup. >0 means busy, but doesn't
  149. * necessarily indicate the number of tasks in the cgroup
  150. */
  151. atomic_t count;
  152. int id; /* ida allocated in-hierarchy ID */
  153. /*
  154. * We link our 'sibling' struct into our parent's 'children'.
  155. * Our children link their 'sibling' into our 'children'.
  156. */
  157. struct list_head sibling; /* my parent's children */
  158. struct list_head children; /* my children */
  159. struct list_head files; /* my files */
  160. struct cgroup *parent; /* my parent */
  161. struct dentry *dentry; /* cgroup fs entry, RCU protected */
  162. /*
  163. * This is a copy of dentry->d_name, and it's needed because
  164. * we can't use dentry->d_name in cgroup_path().
  165. *
  166. * You must acquire rcu_read_lock() to access cgrp->name, and
  167. * the only place that can change it is rename(), which is
  168. * protected by parent dir's i_mutex.
  169. *
  170. * Normally you should use cgroup_name() wrapper rather than
  171. * access it directly.
  172. */
  173. struct cgroup_name __rcu *name;
  174. /* Private pointers for each registered subsystem */
  175. struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
  176. struct cgroupfs_root *root;
  177. struct cgroup *top_cgroup;
  178. /*
  179. * List of cg_cgroup_links pointing at css_sets with
  180. * tasks in this cgroup. Protected by css_set_lock
  181. */
  182. struct list_head css_sets;
  183. struct list_head allcg_node; /* cgroupfs_root->allcg_list */
  184. struct list_head cft_q_node; /* used during cftype add/rm */
  185. /*
  186. * Linked list running through all cgroups that can
  187. * potentially be reaped by the release agent. Protected by
  188. * release_list_lock
  189. */
  190. struct list_head release_list;
  191. /*
  192. * list of pidlists, up to two for each namespace (one for procs, one
  193. * for tasks); created on demand.
  194. */
  195. struct list_head pidlists;
  196. struct mutex pidlist_mutex;
  197. /* For RCU-protected deletion */
  198. struct rcu_head rcu_head;
  199. struct work_struct free_work;
  200. /* List of events which userspace want to receive */
  201. struct list_head event_list;
  202. spinlock_t event_list_lock;
  203. /* directory xattrs */
  204. struct simple_xattrs xattrs;
  205. };
  206. /*
  207. * A css_set is a structure holding pointers to a set of
  208. * cgroup_subsys_state objects. This saves space in the task struct
  209. * object and speeds up fork()/exit(), since a single inc/dec and a
  210. * list_add()/del() can bump the reference count on the entire cgroup
  211. * set for a task.
  212. */
  213. struct css_set {
  214. /* Reference count */
  215. atomic_t refcount;
  216. /*
  217. * List running through all cgroup groups in the same hash
  218. * slot. Protected by css_set_lock
  219. */
  220. struct hlist_node hlist;
  221. /*
  222. * List running through all tasks using this cgroup
  223. * group. Protected by css_set_lock
  224. */
  225. struct list_head tasks;
  226. /*
  227. * List of cg_cgroup_link objects on link chains from
  228. * cgroups referenced from this css_set. Protected by
  229. * css_set_lock
  230. */
  231. struct list_head cg_links;
  232. /*
  233. * Set of subsystem states, one for each subsystem. This array
  234. * is immutable after creation apart from the init_css_set
  235. * during subsystem registration (at boot time) and modular subsystem
  236. * loading/unloading.
  237. */
  238. struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
  239. /* For RCU-protected deletion */
  240. struct rcu_head rcu_head;
  241. };
  242. /*
  243. * cgroup_map_cb is an abstract callback API for reporting map-valued
  244. * control files
  245. */
  246. struct cgroup_map_cb {
  247. int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
  248. void *state;
  249. };
  250. /*
  251. * struct cftype: handler definitions for cgroup control files
  252. *
  253. * When reading/writing to a file:
  254. * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
  255. * - the 'cftype' of the file is file->f_dentry->d_fsdata
  256. */
  257. /* cftype->flags */
  258. #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
  259. #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
  260. #define MAX_CFTYPE_NAME 64
  261. struct cftype {
  262. /*
  263. * By convention, the name should begin with the name of the
  264. * subsystem, followed by a period. Zero length string indicates
  265. * end of cftype array.
  266. */
  267. char name[MAX_CFTYPE_NAME];
  268. int private;
  269. /*
  270. * If not 0, file mode is set to this value, otherwise it will
  271. * be figured out automatically
  272. */
  273. umode_t mode;
  274. /*
  275. * If non-zero, defines the maximum length of string that can
  276. * be passed to write_string; defaults to 64
  277. */
  278. size_t max_write_len;
  279. /* CFTYPE_* flags */
  280. unsigned int flags;
  281. /* file xattrs */
  282. struct simple_xattrs xattrs;
  283. int (*open)(struct inode *inode, struct file *file);
  284. ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
  285. struct file *file,
  286. char __user *buf, size_t nbytes, loff_t *ppos);
  287. /*
  288. * read_u64() is a shortcut for the common case of returning a
  289. * single integer. Use it in place of read()
  290. */
  291. u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
  292. /*
  293. * read_s64() is a signed version of read_u64()
  294. */
  295. s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
  296. /*
  297. * read_map() is used for defining a map of key/value
  298. * pairs. It should call cb->fill(cb, key, value) for each
  299. * entry. The key/value pairs (and their ordering) should not
  300. * change between reboots.
  301. */
  302. int (*read_map)(struct cgroup *cont, struct cftype *cft,
  303. struct cgroup_map_cb *cb);
  304. /*
  305. * read_seq_string() is used for outputting a simple sequence
  306. * using seqfile.
  307. */
  308. int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
  309. struct seq_file *m);
  310. ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
  311. struct file *file,
  312. const char __user *buf, size_t nbytes, loff_t *ppos);
  313. /*
  314. * write_u64() is a shortcut for the common case of accepting
  315. * a single integer (as parsed by simple_strtoull) from
  316. * userspace. Use in place of write(); return 0 or error.
  317. */
  318. int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
  319. /*
  320. * write_s64() is a signed version of write_u64()
  321. */
  322. int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
  323. /*
  324. * write_string() is passed a nul-terminated kernelspace
  325. * buffer of maximum length determined by max_write_len.
  326. * Returns 0 or -ve error code.
  327. */
  328. int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
  329. const char *buffer);
  330. /*
  331. * trigger() callback can be used to get some kick from the
  332. * userspace, when the actual string written is not important
  333. * at all. The private field can be used to determine the
  334. * kick type for multiplexing.
  335. */
  336. int (*trigger)(struct cgroup *cgrp, unsigned int event);
  337. int (*release)(struct inode *inode, struct file *file);
  338. /*
  339. * register_event() callback will be used to add new userspace
  340. * waiter for changes related to the cftype. Implement it if
  341. * you want to provide this functionality. Use eventfd_signal()
  342. * on eventfd to send notification to userspace.
  343. */
  344. int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
  345. struct eventfd_ctx *eventfd, const char *args);
  346. /*
  347. * unregister_event() callback will be called when userspace
  348. * closes the eventfd or on cgroup removing.
  349. * This callback must be implemented, if you want provide
  350. * notification functionality.
  351. */
  352. void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
  353. struct eventfd_ctx *eventfd);
  354. };
  355. /*
  356. * cftype_sets describe cftypes belonging to a subsystem and are chained at
  357. * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
  358. * terminated by zero length name.
  359. */
  360. struct cftype_set {
  361. struct list_head node; /* chained at subsys->cftsets */
  362. struct cftype *cfts;
  363. };
  364. struct cgroup_scanner {
  365. struct cgroup *cg;
  366. int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
  367. void (*process_task)(struct task_struct *p,
  368. struct cgroup_scanner *scan);
  369. struct ptr_heap *heap;
  370. void *data;
  371. };
  372. /* Caller should hold rcu_read_lock() */
  373. static inline const char *cgroup_name(const struct cgroup *cgrp)
  374. {
  375. return rcu_dereference(cgrp->name)->name;
  376. }
  377. int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
  378. int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
  379. int cgroup_is_removed(const struct cgroup *cgrp);
  380. int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
  381. int cgroup_task_count(const struct cgroup *cgrp);
  382. /*
  383. * Control Group taskset, used to pass around set of tasks to cgroup_subsys
  384. * methods.
  385. */
  386. struct cgroup_taskset;
  387. struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
  388. struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
  389. struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
  390. int cgroup_taskset_size(struct cgroup_taskset *tset);
  391. /**
  392. * cgroup_taskset_for_each - iterate cgroup_taskset
  393. * @task: the loop cursor
  394. * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
  395. * @tset: taskset to iterate
  396. */
  397. #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
  398. for ((task) = cgroup_taskset_first((tset)); (task); \
  399. (task) = cgroup_taskset_next((tset))) \
  400. if (!(skip_cgrp) || \
  401. cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
  402. /*
  403. * Control Group subsystem type.
  404. * See Documentation/cgroups/cgroups.txt for details
  405. */
  406. struct cgroup_subsys {
  407. struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
  408. int (*css_online)(struct cgroup *cgrp);
  409. void (*css_offline)(struct cgroup *cgrp);
  410. void (*css_free)(struct cgroup *cgrp);
  411. int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
  412. void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
  413. void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
  414. void (*fork)(struct task_struct *task);
  415. void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
  416. struct task_struct *task);
  417. void (*bind)(struct cgroup *root);
  418. int subsys_id;
  419. int active;
  420. int disabled;
  421. int early_init;
  422. /*
  423. * True if this subsys uses ID. ID is not available before cgroup_init()
  424. * (not available in early_init time.)
  425. */
  426. bool use_id;
  427. /*
  428. * If %false, this subsystem is properly hierarchical -
  429. * configuration, resource accounting and restriction on a parent
  430. * cgroup cover those of its children. If %true, hierarchy support
  431. * is broken in some ways - some subsystems ignore hierarchy
  432. * completely while others are only implemented half-way.
  433. *
  434. * It's now disallowed to create nested cgroups if the subsystem is
  435. * broken and cgroup core will emit a warning message on such
  436. * cases. Eventually, all subsystems will be made properly
  437. * hierarchical and this will go away.
  438. */
  439. bool broken_hierarchy;
  440. bool warned_broken_hierarchy;
  441. #define MAX_CGROUP_TYPE_NAMELEN 32
  442. const char *name;
  443. /*
  444. * Link to parent, and list entry in parent's children.
  445. * Protected by cgroup_lock()
  446. */
  447. struct cgroupfs_root *root;
  448. struct list_head sibling;
  449. /* used when use_id == true */
  450. struct idr idr;
  451. spinlock_t id_lock;
  452. /* list of cftype_sets */
  453. struct list_head cftsets;
  454. /* base cftypes, automatically [de]registered with subsys itself */
  455. struct cftype *base_cftypes;
  456. struct cftype_set base_cftset;
  457. /* should be defined only by modular subsystems */
  458. struct module *module;
  459. };
  460. #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
  461. #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
  462. #include <linux/cgroup_subsys.h>
  463. #undef IS_SUBSYS_ENABLED
  464. #undef SUBSYS
  465. static inline struct cgroup_subsys_state *cgroup_subsys_state(
  466. struct cgroup *cgrp, int subsys_id)
  467. {
  468. return cgrp->subsys[subsys_id];
  469. }
  470. /*
  471. * function to get the cgroup_subsys_state which allows for extra
  472. * rcu_dereference_check() conditions, such as locks used during the
  473. * cgroup_subsys::attach() methods.
  474. */
  475. #define task_subsys_state_check(task, subsys_id, __c) \
  476. rcu_dereference_check(task->cgroups->subsys[subsys_id], \
  477. lockdep_is_held(&task->alloc_lock) || \
  478. cgroup_lock_is_held() || (__c))
  479. static inline struct cgroup_subsys_state *
  480. task_subsys_state(struct task_struct *task, int subsys_id)
  481. {
  482. return task_subsys_state_check(task, subsys_id, false);
  483. }
  484. static inline struct cgroup* task_cgroup(struct task_struct *task,
  485. int subsys_id)
  486. {
  487. return task_subsys_state(task, subsys_id)->cgroup;
  488. }
  489. /**
  490. * cgroup_for_each_child - iterate through children of a cgroup
  491. * @pos: the cgroup * to use as the loop cursor
  492. * @cgroup: cgroup whose children to walk
  493. *
  494. * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
  495. * cgroup which hasn't finished ->css_online() or already has finished
  496. * ->css_offline() may show up during traversal and it's each subsystem's
  497. * responsibility to verify that each @pos is alive.
  498. *
  499. * If a subsystem synchronizes against the parent in its ->css_online() and
  500. * before starting iterating, a cgroup which finished ->css_online() is
  501. * guaranteed to be visible in the future iterations.
  502. */
  503. #define cgroup_for_each_child(pos, cgroup) \
  504. list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
  505. struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
  506. struct cgroup *cgroup);
  507. struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
  508. /**
  509. * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
  510. * @pos: the cgroup * to use as the loop cursor
  511. * @cgroup: cgroup whose descendants to walk
  512. *
  513. * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
  514. * descendant cgroup which hasn't finished ->css_online() or already has
  515. * finished ->css_offline() may show up during traversal and it's each
  516. * subsystem's responsibility to verify that each @pos is alive.
  517. *
  518. * If a subsystem synchronizes against the parent in its ->css_online() and
  519. * before starting iterating, and synchronizes against @pos on each
  520. * iteration, any descendant cgroup which finished ->css_offline() is
  521. * guaranteed to be visible in the future iterations.
  522. *
  523. * In other words, the following guarantees that a descendant can't escape
  524. * state updates of its ancestors.
  525. *
  526. * my_online(@cgrp)
  527. * {
  528. * Lock @cgrp->parent and @cgrp;
  529. * Inherit state from @cgrp->parent;
  530. * Unlock both.
  531. * }
  532. *
  533. * my_update_state(@cgrp)
  534. * {
  535. * Lock @cgrp;
  536. * Update @cgrp's state;
  537. * Unlock @cgrp;
  538. *
  539. * cgroup_for_each_descendant_pre(@pos, @cgrp) {
  540. * Lock @pos;
  541. * Verify @pos is alive and inherit state from @pos->parent;
  542. * Unlock @pos;
  543. * }
  544. * }
  545. *
  546. * As long as the inheriting step, including checking the parent state, is
  547. * enclosed inside @pos locking, double-locking the parent isn't necessary
  548. * while inheriting. The state update to the parent is guaranteed to be
  549. * visible by walking order and, as long as inheriting operations to the
  550. * same @pos are atomic to each other, multiple updates racing each other
  551. * still result in the correct state. It's guaranateed that at least one
  552. * inheritance happens for any cgroup after the latest update to its
  553. * parent.
  554. *
  555. * If checking parent's state requires locking the parent, each inheriting
  556. * iteration should lock and unlock both @pos->parent and @pos.
  557. *
  558. * Alternatively, a subsystem may choose to use a single global lock to
  559. * synchronize ->css_online() and ->css_offline() against tree-walking
  560. * operations.
  561. */
  562. #define cgroup_for_each_descendant_pre(pos, cgroup) \
  563. for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
  564. pos = cgroup_next_descendant_pre((pos), (cgroup)))
  565. struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
  566. struct cgroup *cgroup);
  567. /**
  568. * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
  569. * @pos: the cgroup * to use as the loop cursor
  570. * @cgroup: cgroup whose descendants to walk
  571. *
  572. * Similar to cgroup_for_each_descendant_pre() but performs post-order
  573. * traversal instead. Note that the walk visibility guarantee described in
  574. * pre-order walk doesn't apply the same to post-order walks.
  575. */
  576. #define cgroup_for_each_descendant_post(pos, cgroup) \
  577. for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
  578. pos = cgroup_next_descendant_post((pos), (cgroup)))
  579. /* A cgroup_iter should be treated as an opaque object */
  580. struct cgroup_iter {
  581. struct list_head *cg_link;
  582. struct list_head *task;
  583. };
  584. /*
  585. * To iterate across the tasks in a cgroup:
  586. *
  587. * 1) call cgroup_iter_start to initialize an iterator
  588. *
  589. * 2) call cgroup_iter_next() to retrieve member tasks until it
  590. * returns NULL or until you want to end the iteration
  591. *
  592. * 3) call cgroup_iter_end() to destroy the iterator.
  593. *
  594. * Or, call cgroup_scan_tasks() to iterate through every task in a
  595. * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
  596. * the test_task() callback, but not while calling the process_task()
  597. * callback.
  598. */
  599. void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
  600. struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
  601. struct cgroup_iter *it);
  602. void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
  603. int cgroup_scan_tasks(struct cgroup_scanner *scan);
  604. int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
  605. bool threadgroup);
  606. int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
  607. int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
  608. /*
  609. * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
  610. * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
  611. * CSS ID is assigned at cgroup allocation (create) automatically
  612. * and removed when subsys calls free_css_id() function. This is because
  613. * the lifetime of cgroup_subsys_state is subsys's matter.
  614. *
  615. * Looking up and scanning function should be called under rcu_read_lock().
  616. * Taking cgroup_mutex is not necessary for following calls.
  617. * But the css returned by this routine can be "not populated yet" or "being
  618. * destroyed". The caller should check css and cgroup's status.
  619. */
  620. /*
  621. * Typically Called at ->destroy(), or somewhere the subsys frees
  622. * cgroup_subsys_state.
  623. */
  624. void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
  625. /* Find a cgroup_subsys_state which has given ID */
  626. struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
  627. /*
  628. * Get a cgroup whose id is greater than or equal to id under tree of root.
  629. * Returning a cgroup_subsys_state or NULL.
  630. */
  631. struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
  632. struct cgroup_subsys_state *root, int *foundid);
  633. /* Returns true if root is ancestor of cg */
  634. bool css_is_ancestor(struct cgroup_subsys_state *cg,
  635. const struct cgroup_subsys_state *root);
  636. /* Get id and depth of css */
  637. unsigned short css_id(struct cgroup_subsys_state *css);
  638. unsigned short css_depth(struct cgroup_subsys_state *css);
  639. struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
  640. #else /* !CONFIG_CGROUPS */
  641. static inline int cgroup_init_early(void) { return 0; }
  642. static inline int cgroup_init(void) { return 0; }
  643. static inline void cgroup_fork(struct task_struct *p) {}
  644. static inline void cgroup_post_fork(struct task_struct *p) {}
  645. static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
  646. static inline void cgroup_lock(void) {}
  647. static inline void cgroup_unlock(void) {}
  648. static inline int cgroupstats_build(struct cgroupstats *stats,
  649. struct dentry *dentry)
  650. {
  651. return -EINVAL;
  652. }
  653. /* No cgroups - nothing to do */
  654. static inline int cgroup_attach_task_all(struct task_struct *from,
  655. struct task_struct *t)
  656. {
  657. return 0;
  658. }
  659. #endif /* !CONFIG_CGROUPS */
  660. #endif /* _LINUX_CGROUP_H */