cgroup.h 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416
  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/cgroupstats.h>
  15. #include <linux/prio_heap.h>
  16. #include <linux/rwsem.h>
  17. #ifdef CONFIG_CGROUPS
  18. struct cgroupfs_root;
  19. struct cgroup_subsys;
  20. struct inode;
  21. struct cgroup;
  22. extern int cgroup_init_early(void);
  23. extern int cgroup_init(void);
  24. extern void cgroup_lock(void);
  25. extern bool cgroup_lock_live_group(struct cgroup *cgrp);
  26. extern void cgroup_unlock(void);
  27. extern void cgroup_fork(struct task_struct *p);
  28. extern void cgroup_fork_callbacks(struct task_struct *p);
  29. extern void cgroup_post_fork(struct task_struct *p);
  30. extern void cgroup_exit(struct task_struct *p, int run_callbacks);
  31. extern int cgroupstats_build(struct cgroupstats *stats,
  32. struct dentry *dentry);
  33. extern struct file_operations proc_cgroup_operations;
  34. /* Define the enumeration of all cgroup subsystems */
  35. #define SUBSYS(_x) _x ## _subsys_id,
  36. enum cgroup_subsys_id {
  37. #include <linux/cgroup_subsys.h>
  38. CGROUP_SUBSYS_COUNT
  39. };
  40. #undef SUBSYS
  41. /* Per-subsystem/per-cgroup state maintained by the system. */
  42. struct cgroup_subsys_state {
  43. /* The cgroup that this subsystem is attached to. Useful
  44. * for subsystems that want to know about the cgroup
  45. * hierarchy structure */
  46. struct cgroup *cgroup;
  47. /* State maintained by the cgroup system to allow
  48. * subsystems to be "busy". Should be accessed via css_get()
  49. * and css_put() */
  50. atomic_t refcnt;
  51. unsigned long flags;
  52. };
  53. /* bits in struct cgroup_subsys_state flags field */
  54. enum {
  55. CSS_ROOT, /* This CSS is the root of the subsystem */
  56. };
  57. /*
  58. * Call css_get() to hold a reference on the cgroup;
  59. *
  60. */
  61. static inline void css_get(struct cgroup_subsys_state *css)
  62. {
  63. /* We don't need to reference count the root state */
  64. if (!test_bit(CSS_ROOT, &css->flags))
  65. atomic_inc(&css->refcnt);
  66. }
  67. /*
  68. * css_put() should be called to release a reference taken by
  69. * css_get()
  70. */
  71. extern void __css_put(struct cgroup_subsys_state *css);
  72. static inline void css_put(struct cgroup_subsys_state *css)
  73. {
  74. if (!test_bit(CSS_ROOT, &css->flags))
  75. __css_put(css);
  76. }
  77. /* bits in struct cgroup flags field */
  78. enum {
  79. /* Control Group is dead */
  80. CGRP_REMOVED,
  81. /* Control Group has previously had a child cgroup or a task,
  82. * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
  83. CGRP_RELEASABLE,
  84. /* Control Group requires release notifications to userspace */
  85. CGRP_NOTIFY_ON_RELEASE,
  86. };
  87. struct cgroup {
  88. unsigned long flags; /* "unsigned long" so bitops work */
  89. /* count users of this cgroup. >0 means busy, but doesn't
  90. * necessarily indicate the number of tasks in the
  91. * cgroup */
  92. atomic_t count;
  93. /*
  94. * We link our 'sibling' struct into our parent's 'children'.
  95. * Our children link their 'sibling' into our 'children'.
  96. */
  97. struct list_head sibling; /* my parent's children */
  98. struct list_head children; /* my children */
  99. struct cgroup *parent; /* my parent */
  100. struct dentry *dentry; /* cgroup fs entry */
  101. /* Private pointers for each registered subsystem */
  102. struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
  103. struct cgroupfs_root *root;
  104. struct cgroup *top_cgroup;
  105. /*
  106. * List of cg_cgroup_links pointing at css_sets with
  107. * tasks in this cgroup. Protected by css_set_lock
  108. */
  109. struct list_head css_sets;
  110. /*
  111. * Linked list running through all cgroups that can
  112. * potentially be reaped by the release agent. Protected by
  113. * release_list_lock
  114. */
  115. struct list_head release_list;
  116. /* pids_mutex protects the fields below */
  117. struct rw_semaphore pids_mutex;
  118. /* Array of process ids in the cgroup */
  119. pid_t *tasks_pids;
  120. /* How many files are using the current tasks_pids array */
  121. int pids_use_count;
  122. /* Length of the current tasks_pids array */
  123. int pids_length;
  124. };
  125. /* A css_set is a structure holding pointers to a set of
  126. * cgroup_subsys_state objects. This saves space in the task struct
  127. * object and speeds up fork()/exit(), since a single inc/dec and a
  128. * list_add()/del() can bump the reference count on the entire
  129. * cgroup set for a task.
  130. */
  131. struct css_set {
  132. /* Reference count */
  133. atomic_t refcount;
  134. /*
  135. * List running through all cgroup groups in the same hash
  136. * slot. Protected by css_set_lock
  137. */
  138. struct hlist_node hlist;
  139. /*
  140. * List running through all tasks using this cgroup
  141. * group. Protected by css_set_lock
  142. */
  143. struct list_head tasks;
  144. /*
  145. * List of cg_cgroup_link objects on link chains from
  146. * cgroups referenced from this css_set. Protected by
  147. * css_set_lock
  148. */
  149. struct list_head cg_links;
  150. /*
  151. * Set of subsystem states, one for each subsystem. This array
  152. * is immutable after creation apart from the init_css_set
  153. * during subsystem registration (at boot time).
  154. */
  155. struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
  156. };
  157. /*
  158. * cgroup_map_cb is an abstract callback API for reporting map-valued
  159. * control files
  160. */
  161. struct cgroup_map_cb {
  162. int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
  163. void *state;
  164. };
  165. /* struct cftype:
  166. *
  167. * The files in the cgroup filesystem mostly have a very simple read/write
  168. * handling, some common function will take care of it. Nevertheless some cases
  169. * (read tasks) are special and therefore I define this structure for every
  170. * kind of file.
  171. *
  172. *
  173. * When reading/writing to a file:
  174. * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
  175. * - the 'cftype' of the file is file->f_dentry->d_fsdata
  176. */
  177. #define MAX_CFTYPE_NAME 64
  178. struct cftype {
  179. /* By convention, the name should begin with the name of the
  180. * subsystem, followed by a period */
  181. char name[MAX_CFTYPE_NAME];
  182. int private;
  183. /*
  184. * If non-zero, defines the maximum length of string that can
  185. * be passed to write_string; defaults to 64
  186. */
  187. size_t max_write_len;
  188. int (*open)(struct inode *inode, struct file *file);
  189. ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
  190. struct file *file,
  191. char __user *buf, size_t nbytes, loff_t *ppos);
  192. /*
  193. * read_u64() is a shortcut for the common case of returning a
  194. * single integer. Use it in place of read()
  195. */
  196. u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
  197. /*
  198. * read_s64() is a signed version of read_u64()
  199. */
  200. s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
  201. /*
  202. * read_map() is used for defining a map of key/value
  203. * pairs. It should call cb->fill(cb, key, value) for each
  204. * entry. The key/value pairs (and their ordering) should not
  205. * change between reboots.
  206. */
  207. int (*read_map)(struct cgroup *cont, struct cftype *cft,
  208. struct cgroup_map_cb *cb);
  209. /*
  210. * read_seq_string() is used for outputting a simple sequence
  211. * using seqfile.
  212. */
  213. int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
  214. struct seq_file *m);
  215. ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
  216. struct file *file,
  217. const char __user *buf, size_t nbytes, loff_t *ppos);
  218. /*
  219. * write_u64() is a shortcut for the common case of accepting
  220. * a single integer (as parsed by simple_strtoull) from
  221. * userspace. Use in place of write(); return 0 or error.
  222. */
  223. int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
  224. /*
  225. * write_s64() is a signed version of write_u64()
  226. */
  227. int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
  228. /*
  229. * write_string() is passed a nul-terminated kernelspace
  230. * buffer of maximum length determined by max_write_len.
  231. * Returns 0 or -ve error code.
  232. */
  233. int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
  234. const char *buffer);
  235. /*
  236. * trigger() callback can be used to get some kick from the
  237. * userspace, when the actual string written is not important
  238. * at all. The private field can be used to determine the
  239. * kick type for multiplexing.
  240. */
  241. int (*trigger)(struct cgroup *cgrp, unsigned int event);
  242. int (*release)(struct inode *inode, struct file *file);
  243. };
  244. struct cgroup_scanner {
  245. struct cgroup *cg;
  246. int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
  247. void (*process_task)(struct task_struct *p,
  248. struct cgroup_scanner *scan);
  249. struct ptr_heap *heap;
  250. };
  251. /* Add a new file to the given cgroup directory. Should only be
  252. * called by subsystems from within a populate() method */
  253. int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
  254. const struct cftype *cft);
  255. /* Add a set of new files to the given cgroup directory. Should
  256. * only be called by subsystems from within a populate() method */
  257. int cgroup_add_files(struct cgroup *cgrp,
  258. struct cgroup_subsys *subsys,
  259. const struct cftype cft[],
  260. int count);
  261. int cgroup_is_removed(const struct cgroup *cgrp);
  262. int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
  263. int cgroup_task_count(const struct cgroup *cgrp);
  264. /* Return true if the cgroup is a descendant of the current cgroup */
  265. int cgroup_is_descendant(const struct cgroup *cgrp);
  266. /* Control Group subsystem type. See Documentation/cgroups.txt for details */
  267. struct cgroup_subsys {
  268. struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
  269. struct cgroup *cgrp);
  270. void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
  271. void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
  272. int (*can_attach)(struct cgroup_subsys *ss,
  273. struct cgroup *cgrp, struct task_struct *tsk);
  274. void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
  275. struct cgroup *old_cgrp, struct task_struct *tsk);
  276. void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
  277. void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
  278. int (*populate)(struct cgroup_subsys *ss,
  279. struct cgroup *cgrp);
  280. void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
  281. void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
  282. int subsys_id;
  283. int active;
  284. int disabled;
  285. int early_init;
  286. #define MAX_CGROUP_TYPE_NAMELEN 32
  287. const char *name;
  288. /* Protected by RCU */
  289. struct cgroupfs_root *root;
  290. struct list_head sibling;
  291. };
  292. #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
  293. #include <linux/cgroup_subsys.h>
  294. #undef SUBSYS
  295. static inline struct cgroup_subsys_state *cgroup_subsys_state(
  296. struct cgroup *cgrp, int subsys_id)
  297. {
  298. return cgrp->subsys[subsys_id];
  299. }
  300. static inline struct cgroup_subsys_state *task_subsys_state(
  301. struct task_struct *task, int subsys_id)
  302. {
  303. return rcu_dereference(task->cgroups->subsys[subsys_id]);
  304. }
  305. static inline struct cgroup* task_cgroup(struct task_struct *task,
  306. int subsys_id)
  307. {
  308. return task_subsys_state(task, subsys_id)->cgroup;
  309. }
  310. int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
  311. char *nodename);
  312. /* A cgroup_iter should be treated as an opaque object */
  313. struct cgroup_iter {
  314. struct list_head *cg_link;
  315. struct list_head *task;
  316. };
  317. /* To iterate across the tasks in a cgroup:
  318. *
  319. * 1) call cgroup_iter_start to intialize an iterator
  320. *
  321. * 2) call cgroup_iter_next() to retrieve member tasks until it
  322. * returns NULL or until you want to end the iteration
  323. *
  324. * 3) call cgroup_iter_end() to destroy the iterator.
  325. *
  326. * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
  327. * - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
  328. * callback, but not while calling the process_task() callback.
  329. */
  330. void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
  331. struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
  332. struct cgroup_iter *it);
  333. void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
  334. int cgroup_scan_tasks(struct cgroup_scanner *scan);
  335. int cgroup_attach_task(struct cgroup *, struct task_struct *);
  336. #else /* !CONFIG_CGROUPS */
  337. static inline int cgroup_init_early(void) { return 0; }
  338. static inline int cgroup_init(void) { return 0; }
  339. static inline void cgroup_fork(struct task_struct *p) {}
  340. static inline void cgroup_fork_callbacks(struct task_struct *p) {}
  341. static inline void cgroup_post_fork(struct task_struct *p) {}
  342. static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
  343. static inline void cgroup_lock(void) {}
  344. static inline void cgroup_unlock(void) {}
  345. static inline int cgroupstats_build(struct cgroupstats *stats,
  346. struct dentry *dentry)
  347. {
  348. return -EINVAL;
  349. }
  350. #endif /* !CONFIG_CGROUPS */
  351. #endif /* _LINUX_CGROUP_H */