cgroup.h 23 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766
  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_fork(struct task_struct *p);
  30. extern void cgroup_post_fork(struct task_struct *p);
  31. extern void cgroup_exit(struct task_struct *p, int run_callbacks);
  32. extern int cgroupstats_build(struct cgroupstats *stats,
  33. struct dentry *dentry);
  34. extern int cgroup_load_subsys(struct cgroup_subsys *ss);
  35. extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
  36. extern const struct file_operations proc_cgroup_operations;
  37. /*
  38. * Define the enumeration of all cgroup subsystems.
  39. *
  40. * We define ids for builtin subsystems and then modular ones.
  41. */
  42. #define SUBSYS(_x) _x ## _subsys_id,
  43. enum cgroup_subsys_id {
  44. #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
  45. #include <linux/cgroup_subsys.h>
  46. #undef IS_SUBSYS_ENABLED
  47. CGROUP_BUILTIN_SUBSYS_COUNT,
  48. __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1,
  49. #define IS_SUBSYS_ENABLED(option) IS_MODULE(option)
  50. #include <linux/cgroup_subsys.h>
  51. #undef IS_SUBSYS_ENABLED
  52. CGROUP_SUBSYS_COUNT,
  53. };
  54. #undef SUBSYS
  55. /* Per-subsystem/per-cgroup state maintained by the system. */
  56. struct cgroup_subsys_state {
  57. /*
  58. * The cgroup that this subsystem is attached to. Useful
  59. * for subsystems that want to know about the cgroup
  60. * hierarchy structure
  61. */
  62. struct cgroup *cgroup;
  63. /*
  64. * State maintained by the cgroup system to allow subsystems
  65. * to be "busy". Should be accessed via css_get(),
  66. * css_tryget() and css_put().
  67. */
  68. atomic_t refcnt;
  69. unsigned long flags;
  70. /* ID for this css, if possible */
  71. struct css_id __rcu *id;
  72. /* Used to put @cgroup->dentry on the last css_put() */
  73. struct work_struct dput_work;
  74. };
  75. /* bits in struct cgroup_subsys_state flags field */
  76. enum {
  77. CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
  78. CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
  79. };
  80. /* Caller must verify that the css is not for root cgroup */
  81. static inline void __css_get(struct cgroup_subsys_state *css, int count)
  82. {
  83. atomic_add(count, &css->refcnt);
  84. }
  85. /*
  86. * Call css_get() to hold a reference on the css; it can be used
  87. * for a reference obtained via:
  88. * - an existing ref-counted reference to the css
  89. * - task->cgroups for a locked task
  90. */
  91. static inline void css_get(struct cgroup_subsys_state *css)
  92. {
  93. /* We don't need to reference count the root state */
  94. if (!(css->flags & CSS_ROOT))
  95. __css_get(css, 1);
  96. }
  97. /*
  98. * Call css_tryget() to take a reference on a css if your existing
  99. * (known-valid) reference isn't already ref-counted. Returns false if
  100. * the css has been destroyed.
  101. */
  102. extern bool __css_tryget(struct cgroup_subsys_state *css);
  103. static inline bool css_tryget(struct cgroup_subsys_state *css)
  104. {
  105. if (css->flags & CSS_ROOT)
  106. return true;
  107. return __css_tryget(css);
  108. }
  109. /*
  110. * css_put() should be called to release a reference taken by
  111. * css_get() or css_tryget()
  112. */
  113. extern void __css_put(struct cgroup_subsys_state *css);
  114. static inline void css_put(struct cgroup_subsys_state *css)
  115. {
  116. if (!(css->flags & CSS_ROOT))
  117. __css_put(css);
  118. }
  119. /* bits in struct cgroup flags field */
  120. enum {
  121. /* Control Group is dead */
  122. CGRP_REMOVED,
  123. /*
  124. * Control Group has previously had a child cgroup or a task,
  125. * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
  126. */
  127. CGRP_RELEASABLE,
  128. /* Control Group requires release notifications to userspace */
  129. CGRP_NOTIFY_ON_RELEASE,
  130. /*
  131. * Clone the parent's configuration when creating a new child
  132. * cpuset cgroup. For historical reasons, this option can be
  133. * specified at mount time and thus is implemented here.
  134. */
  135. CGRP_CPUSET_CLONE_CHILDREN,
  136. };
  137. struct cgroup_name {
  138. struct rcu_head rcu_head;
  139. char name[];
  140. };
  141. struct cgroup {
  142. unsigned long flags; /* "unsigned long" so bitops work */
  143. /*
  144. * count users of this cgroup. >0 means busy, but doesn't
  145. * necessarily indicate the number of tasks in the cgroup
  146. */
  147. atomic_t count;
  148. int id; /* ida allocated in-hierarchy ID */
  149. /*
  150. * We link our 'sibling' struct into our parent's 'children'.
  151. * Our children link their 'sibling' into our 'children'.
  152. */
  153. struct list_head sibling; /* my parent's children */
  154. struct list_head children; /* my children */
  155. struct list_head files; /* my files */
  156. struct cgroup *parent; /* my parent */
  157. struct dentry *dentry; /* cgroup fs entry, RCU protected */
  158. /*
  159. * This is a copy of dentry->d_name, and it's needed because
  160. * we can't use dentry->d_name in cgroup_path().
  161. *
  162. * You must acquire rcu_read_lock() to access cgrp->name, and
  163. * the only place that can change it is rename(), which is
  164. * protected by parent dir's i_mutex.
  165. *
  166. * Normally you should use cgroup_name() wrapper rather than
  167. * access it directly.
  168. */
  169. struct cgroup_name __rcu *name;
  170. /* Private pointers for each registered subsystem */
  171. struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
  172. struct cgroupfs_root *root;
  173. struct cgroup *top_cgroup;
  174. /*
  175. * List of cg_cgroup_links pointing at css_sets with
  176. * tasks in this cgroup. Protected by css_set_lock
  177. */
  178. struct list_head css_sets;
  179. struct list_head allcg_node; /* cgroupfs_root->allcg_list */
  180. struct list_head cft_q_node; /* used during cftype add/rm */
  181. /*
  182. * Linked list running through all cgroups that can
  183. * potentially be reaped by the release agent. Protected by
  184. * release_list_lock
  185. */
  186. struct list_head release_list;
  187. /*
  188. * list of pidlists, up to two for each namespace (one for procs, one
  189. * for tasks); created on demand.
  190. */
  191. struct list_head pidlists;
  192. struct mutex pidlist_mutex;
  193. /* For RCU-protected deletion */
  194. struct rcu_head rcu_head;
  195. struct work_struct free_work;
  196. /* List of events which userspace want to receive */
  197. struct list_head event_list;
  198. spinlock_t event_list_lock;
  199. /* directory xattrs */
  200. struct simple_xattrs xattrs;
  201. };
  202. /*
  203. * A css_set is a structure holding pointers to a set of
  204. * cgroup_subsys_state objects. This saves space in the task struct
  205. * object and speeds up fork()/exit(), since a single inc/dec and a
  206. * list_add()/del() can bump the reference count on the entire cgroup
  207. * set for a task.
  208. */
  209. struct css_set {
  210. /* Reference count */
  211. atomic_t refcount;
  212. /*
  213. * List running through all cgroup groups in the same hash
  214. * slot. Protected by css_set_lock
  215. */
  216. struct hlist_node hlist;
  217. /*
  218. * List running through all tasks using this cgroup
  219. * group. Protected by css_set_lock
  220. */
  221. struct list_head tasks;
  222. /*
  223. * List of cg_cgroup_link objects on link chains from
  224. * cgroups referenced from this css_set. Protected by
  225. * css_set_lock
  226. */
  227. struct list_head cg_links;
  228. /*
  229. * Set of subsystem states, one for each subsystem. This array
  230. * is immutable after creation apart from the init_css_set
  231. * during subsystem registration (at boot time) and modular subsystem
  232. * loading/unloading.
  233. */
  234. struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
  235. /* For RCU-protected deletion */
  236. struct rcu_head rcu_head;
  237. };
  238. /*
  239. * cgroup_map_cb is an abstract callback API for reporting map-valued
  240. * control files
  241. */
  242. struct cgroup_map_cb {
  243. int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
  244. void *state;
  245. };
  246. /*
  247. * struct cftype: handler definitions for cgroup control files
  248. *
  249. * When reading/writing to a file:
  250. * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
  251. * - the 'cftype' of the file is file->f_dentry->d_fsdata
  252. */
  253. /* cftype->flags */
  254. #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
  255. #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
  256. #define MAX_CFTYPE_NAME 64
  257. struct cftype {
  258. /*
  259. * By convention, the name should begin with the name of the
  260. * subsystem, followed by a period. Zero length string indicates
  261. * end of cftype array.
  262. */
  263. char name[MAX_CFTYPE_NAME];
  264. int private;
  265. /*
  266. * If not 0, file mode is set to this value, otherwise it will
  267. * be figured out automatically
  268. */
  269. umode_t mode;
  270. /*
  271. * If non-zero, defines the maximum length of string that can
  272. * be passed to write_string; defaults to 64
  273. */
  274. size_t max_write_len;
  275. /* CFTYPE_* flags */
  276. unsigned int flags;
  277. /* file xattrs */
  278. struct simple_xattrs xattrs;
  279. int (*open)(struct inode *inode, struct file *file);
  280. ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
  281. struct file *file,
  282. char __user *buf, size_t nbytes, loff_t *ppos);
  283. /*
  284. * read_u64() is a shortcut for the common case of returning a
  285. * single integer. Use it in place of read()
  286. */
  287. u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
  288. /*
  289. * read_s64() is a signed version of read_u64()
  290. */
  291. s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
  292. /*
  293. * read_map() is used for defining a map of key/value
  294. * pairs. It should call cb->fill(cb, key, value) for each
  295. * entry. The key/value pairs (and their ordering) should not
  296. * change between reboots.
  297. */
  298. int (*read_map)(struct cgroup *cont, struct cftype *cft,
  299. struct cgroup_map_cb *cb);
  300. /*
  301. * read_seq_string() is used for outputting a simple sequence
  302. * using seqfile.
  303. */
  304. int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
  305. struct seq_file *m);
  306. ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
  307. struct file *file,
  308. const char __user *buf, size_t nbytes, loff_t *ppos);
  309. /*
  310. * write_u64() is a shortcut for the common case of accepting
  311. * a single integer (as parsed by simple_strtoull) from
  312. * userspace. Use in place of write(); return 0 or error.
  313. */
  314. int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
  315. /*
  316. * write_s64() is a signed version of write_u64()
  317. */
  318. int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
  319. /*
  320. * write_string() is passed a nul-terminated kernelspace
  321. * buffer of maximum length determined by max_write_len.
  322. * Returns 0 or -ve error code.
  323. */
  324. int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
  325. const char *buffer);
  326. /*
  327. * trigger() callback can be used to get some kick from the
  328. * userspace, when the actual string written is not important
  329. * at all. The private field can be used to determine the
  330. * kick type for multiplexing.
  331. */
  332. int (*trigger)(struct cgroup *cgrp, unsigned int event);
  333. int (*release)(struct inode *inode, struct file *file);
  334. /*
  335. * register_event() callback will be used to add new userspace
  336. * waiter for changes related to the cftype. Implement it if
  337. * you want to provide this functionality. Use eventfd_signal()
  338. * on eventfd to send notification to userspace.
  339. */
  340. int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
  341. struct eventfd_ctx *eventfd, const char *args);
  342. /*
  343. * unregister_event() callback will be called when userspace
  344. * closes the eventfd or on cgroup removing.
  345. * This callback must be implemented, if you want provide
  346. * notification functionality.
  347. */
  348. void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
  349. struct eventfd_ctx *eventfd);
  350. };
  351. /*
  352. * cftype_sets describe cftypes belonging to a subsystem and are chained at
  353. * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
  354. * terminated by zero length name.
  355. */
  356. struct cftype_set {
  357. struct list_head node; /* chained at subsys->cftsets */
  358. struct cftype *cfts;
  359. };
  360. struct cgroup_scanner {
  361. struct cgroup *cg;
  362. int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
  363. void (*process_task)(struct task_struct *p,
  364. struct cgroup_scanner *scan);
  365. struct ptr_heap *heap;
  366. void *data;
  367. };
  368. /* Caller should hold rcu_read_lock() */
  369. static inline const char *cgroup_name(const struct cgroup *cgrp)
  370. {
  371. return rcu_dereference(cgrp->name)->name;
  372. }
  373. int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
  374. int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
  375. int cgroup_is_removed(const struct cgroup *cgrp);
  376. bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
  377. int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
  378. int cgroup_task_count(const struct cgroup *cgrp);
  379. /*
  380. * Control Group taskset, used to pass around set of tasks to cgroup_subsys
  381. * methods.
  382. */
  383. struct cgroup_taskset;
  384. struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
  385. struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
  386. struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
  387. int cgroup_taskset_size(struct cgroup_taskset *tset);
  388. /**
  389. * cgroup_taskset_for_each - iterate cgroup_taskset
  390. * @task: the loop cursor
  391. * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
  392. * @tset: taskset to iterate
  393. */
  394. #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
  395. for ((task) = cgroup_taskset_first((tset)); (task); \
  396. (task) = cgroup_taskset_next((tset))) \
  397. if (!(skip_cgrp) || \
  398. cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
  399. /*
  400. * Control Group subsystem type.
  401. * See Documentation/cgroups/cgroups.txt for details
  402. */
  403. struct cgroup_subsys {
  404. struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
  405. int (*css_online)(struct cgroup *cgrp);
  406. void (*css_offline)(struct cgroup *cgrp);
  407. void (*css_free)(struct cgroup *cgrp);
  408. int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
  409. void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
  410. void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
  411. void (*fork)(struct task_struct *task);
  412. void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
  413. struct task_struct *task);
  414. void (*bind)(struct cgroup *root);
  415. int subsys_id;
  416. int active;
  417. int disabled;
  418. int early_init;
  419. /*
  420. * True if this subsys uses ID. ID is not available before cgroup_init()
  421. * (not available in early_init time.)
  422. */
  423. bool use_id;
  424. /*
  425. * If %false, this subsystem is properly hierarchical -
  426. * configuration, resource accounting and restriction on a parent
  427. * cgroup cover those of its children. If %true, hierarchy support
  428. * is broken in some ways - some subsystems ignore hierarchy
  429. * completely while others are only implemented half-way.
  430. *
  431. * It's now disallowed to create nested cgroups if the subsystem is
  432. * broken and cgroup core will emit a warning message on such
  433. * cases. Eventually, all subsystems will be made properly
  434. * hierarchical and this will go away.
  435. */
  436. bool broken_hierarchy;
  437. bool warned_broken_hierarchy;
  438. #define MAX_CGROUP_TYPE_NAMELEN 32
  439. const char *name;
  440. /*
  441. * Link to parent, and list entry in parent's children.
  442. * Protected by cgroup_lock()
  443. */
  444. struct cgroupfs_root *root;
  445. struct list_head sibling;
  446. /* used when use_id == true */
  447. struct idr idr;
  448. spinlock_t id_lock;
  449. /* list of cftype_sets */
  450. struct list_head cftsets;
  451. /* base cftypes, automatically [de]registered with subsys itself */
  452. struct cftype *base_cftypes;
  453. struct cftype_set base_cftset;
  454. /* should be defined only by modular subsystems */
  455. struct module *module;
  456. };
  457. #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
  458. #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
  459. #include <linux/cgroup_subsys.h>
  460. #undef IS_SUBSYS_ENABLED
  461. #undef SUBSYS
  462. static inline struct cgroup_subsys_state *cgroup_subsys_state(
  463. struct cgroup *cgrp, int subsys_id)
  464. {
  465. return cgrp->subsys[subsys_id];
  466. }
  467. /*
  468. * function to get the cgroup_subsys_state which allows for extra
  469. * rcu_dereference_check() conditions, such as locks used during the
  470. * cgroup_subsys::attach() methods.
  471. */
  472. #ifdef CONFIG_PROVE_RCU
  473. extern struct mutex cgroup_mutex;
  474. #define task_subsys_state_check(task, subsys_id, __c) \
  475. rcu_dereference_check((task)->cgroups->subsys[(subsys_id)], \
  476. lockdep_is_held(&(task)->alloc_lock) || \
  477. lockdep_is_held(&cgroup_mutex) || (__c))
  478. #else
  479. #define task_subsys_state_check(task, subsys_id, __c) \
  480. rcu_dereference((task)->cgroups->subsys[(subsys_id)])
  481. #endif
  482. static inline struct cgroup_subsys_state *
  483. task_subsys_state(struct task_struct *task, int subsys_id)
  484. {
  485. return task_subsys_state_check(task, subsys_id, false);
  486. }
  487. static inline struct cgroup* task_cgroup(struct task_struct *task,
  488. int subsys_id)
  489. {
  490. return task_subsys_state(task, subsys_id)->cgroup;
  491. }
  492. /**
  493. * cgroup_for_each_child - iterate through children of a cgroup
  494. * @pos: the cgroup * to use as the loop cursor
  495. * @cgroup: cgroup whose children to walk
  496. *
  497. * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
  498. * cgroup which hasn't finished ->css_online() or already has finished
  499. * ->css_offline() may show up during traversal and it's each subsystem's
  500. * responsibility to verify that each @pos is alive.
  501. *
  502. * If a subsystem synchronizes against the parent in its ->css_online() and
  503. * before starting iterating, a cgroup which finished ->css_online() is
  504. * guaranteed to be visible in the future iterations.
  505. */
  506. #define cgroup_for_each_child(pos, cgroup) \
  507. list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
  508. struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
  509. struct cgroup *cgroup);
  510. struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
  511. /**
  512. * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
  513. * @pos: the cgroup * to use as the loop cursor
  514. * @cgroup: cgroup whose descendants to walk
  515. *
  516. * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
  517. * descendant cgroup which hasn't finished ->css_online() or already has
  518. * finished ->css_offline() may show up during traversal and it's each
  519. * subsystem's responsibility to verify that each @pos is alive.
  520. *
  521. * If a subsystem synchronizes against the parent in its ->css_online() and
  522. * before starting iterating, and synchronizes against @pos on each
  523. * iteration, any descendant cgroup which finished ->css_offline() is
  524. * guaranteed to be visible in the future iterations.
  525. *
  526. * In other words, the following guarantees that a descendant can't escape
  527. * state updates of its ancestors.
  528. *
  529. * my_online(@cgrp)
  530. * {
  531. * Lock @cgrp->parent and @cgrp;
  532. * Inherit state from @cgrp->parent;
  533. * Unlock both.
  534. * }
  535. *
  536. * my_update_state(@cgrp)
  537. * {
  538. * Lock @cgrp;
  539. * Update @cgrp's state;
  540. * Unlock @cgrp;
  541. *
  542. * cgroup_for_each_descendant_pre(@pos, @cgrp) {
  543. * Lock @pos;
  544. * Verify @pos is alive and inherit state from @pos->parent;
  545. * Unlock @pos;
  546. * }
  547. * }
  548. *
  549. * As long as the inheriting step, including checking the parent state, is
  550. * enclosed inside @pos locking, double-locking the parent isn't necessary
  551. * while inheriting. The state update to the parent is guaranteed to be
  552. * visible by walking order and, as long as inheriting operations to the
  553. * same @pos are atomic to each other, multiple updates racing each other
  554. * still result in the correct state. It's guaranateed that at least one
  555. * inheritance happens for any cgroup after the latest update to its
  556. * parent.
  557. *
  558. * If checking parent's state requires locking the parent, each inheriting
  559. * iteration should lock and unlock both @pos->parent and @pos.
  560. *
  561. * Alternatively, a subsystem may choose to use a single global lock to
  562. * synchronize ->css_online() and ->css_offline() against tree-walking
  563. * operations.
  564. */
  565. #define cgroup_for_each_descendant_pre(pos, cgroup) \
  566. for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
  567. pos = cgroup_next_descendant_pre((pos), (cgroup)))
  568. struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
  569. struct cgroup *cgroup);
  570. /**
  571. * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
  572. * @pos: the cgroup * to use as the loop cursor
  573. * @cgroup: cgroup whose descendants to walk
  574. *
  575. * Similar to cgroup_for_each_descendant_pre() but performs post-order
  576. * traversal instead. Note that the walk visibility guarantee described in
  577. * pre-order walk doesn't apply the same to post-order walks.
  578. */
  579. #define cgroup_for_each_descendant_post(pos, cgroup) \
  580. for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
  581. pos = cgroup_next_descendant_post((pos), (cgroup)))
  582. /* A cgroup_iter should be treated as an opaque object */
  583. struct cgroup_iter {
  584. struct list_head *cg_link;
  585. struct list_head *task;
  586. };
  587. /*
  588. * To iterate across the tasks in a cgroup:
  589. *
  590. * 1) call cgroup_iter_start to initialize an iterator
  591. *
  592. * 2) call cgroup_iter_next() to retrieve member tasks until it
  593. * returns NULL or until you want to end the iteration
  594. *
  595. * 3) call cgroup_iter_end() to destroy the iterator.
  596. *
  597. * Or, call cgroup_scan_tasks() to iterate through every task in a
  598. * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
  599. * the test_task() callback, but not while calling the process_task()
  600. * callback.
  601. */
  602. void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
  603. struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
  604. struct cgroup_iter *it);
  605. void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
  606. int cgroup_scan_tasks(struct cgroup_scanner *scan);
  607. int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
  608. int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
  609. /*
  610. * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
  611. * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
  612. * CSS ID is assigned at cgroup allocation (create) automatically
  613. * and removed when subsys calls free_css_id() function. This is because
  614. * the lifetime of cgroup_subsys_state is subsys's matter.
  615. *
  616. * Looking up and scanning function should be called under rcu_read_lock().
  617. * Taking cgroup_mutex is not necessary for following calls.
  618. * But the css returned by this routine can be "not populated yet" or "being
  619. * destroyed". The caller should check css and cgroup's status.
  620. */
  621. /*
  622. * Typically Called at ->destroy(), or somewhere the subsys frees
  623. * cgroup_subsys_state.
  624. */
  625. void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
  626. /* Find a cgroup_subsys_state which has given ID */
  627. struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
  628. /*
  629. * Get a cgroup whose id is greater than or equal to id under tree of root.
  630. * Returning a cgroup_subsys_state or NULL.
  631. */
  632. struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
  633. struct cgroup_subsys_state *root, int *foundid);
  634. /* Returns true if root is ancestor of cg */
  635. bool css_is_ancestor(struct cgroup_subsys_state *cg,
  636. const struct cgroup_subsys_state *root);
  637. /* Get id and depth of css */
  638. unsigned short css_id(struct cgroup_subsys_state *css);
  639. unsigned short css_depth(struct cgroup_subsys_state *css);
  640. struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
  641. #else /* !CONFIG_CGROUPS */
  642. static inline int cgroup_init_early(void) { return 0; }
  643. static inline int cgroup_init(void) { return 0; }
  644. static inline void cgroup_fork(struct task_struct *p) {}
  645. static inline void cgroup_post_fork(struct task_struct *p) {}
  646. static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
  647. static inline void cgroup_lock(void) {}
  648. static inline void cgroup_unlock(void) {}
  649. static inline int cgroupstats_build(struct cgroupstats *stats,
  650. struct dentry *dentry)
  651. {
  652. return -EINVAL;
  653. }
  654. /* No cgroups - nothing to do */
  655. static inline int cgroup_attach_task_all(struct task_struct *from,
  656. struct task_struct *t)
  657. {
  658. return 0;
  659. }
  660. #endif /* !CONFIG_CGROUPS */
  661. #endif /* _LINUX_CGROUP_H */