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> 12 #include <linux/cpumask.h> 13 #include <linux/nodemask.h> 14 #include <linux/rcupdate.h> 15 #include <linux/cgroupstats.h> 16 #include <linux/prio_heap.h> 17 #include <linux/rwsem.h> 18 19 #ifdef CONFIG_CGROUPS 20 21 struct cgroupfs_root; 22 struct cgroup_subsys; 23 struct inode; 24 struct cgroup; 25 26 extern int cgroup_init_early(void); 27 extern int cgroup_init(void); 28 extern void cgroup_lock(void); 29 extern bool cgroup_lock_live_group(struct cgroup *cgrp); 30 extern void cgroup_unlock(void); 31 extern void cgroup_fork(struct task_struct *p); 32 extern void cgroup_fork_callbacks(struct task_struct *p); 33 extern void cgroup_post_fork(struct task_struct *p); 34 extern void cgroup_exit(struct task_struct *p, int run_callbacks); 35 extern int cgroupstats_build(struct cgroupstats *stats, 36 struct dentry *dentry); 37 38 extern struct file_operations proc_cgroup_operations; 39 40 /* Define the enumeration of all cgroup subsystems */ 41 #define SUBSYS(_x) _x ## _subsys_id, 42 enum cgroup_subsys_id { 43 #include <linux/cgroup_subsys.h> 44 CGROUP_SUBSYS_COUNT 45 }; 46 #undef SUBSYS 47 48 /* Per-subsystem/per-cgroup state maintained by the system. */ 49 struct cgroup_subsys_state { 50 /* The cgroup that this subsystem is attached to. Useful 51 * for subsystems that want to know about the cgroup 52 * hierarchy structure */ 53 struct cgroup *cgroup; 54 55 /* State maintained by the cgroup system to allow 56 * subsystems to be "busy". Should be accessed via css_get() 57 * and css_put() */ 58 59 atomic_t refcnt; 60 61 unsigned long flags; 62 }; 63 64 /* bits in struct cgroup_subsys_state flags field */ 65 enum { 66 CSS_ROOT, /* This CSS is the root of the subsystem */ 67 }; 68 69 /* 70 * Call css_get() to hold a reference on the cgroup; 71 * 72 */ 73 74 static inline void css_get(struct cgroup_subsys_state *css) 75 { 76 /* We don't need to reference count the root state */ 77 if (!test_bit(CSS_ROOT, &css->flags)) 78 atomic_inc(&css->refcnt); 79 } 80 /* 81 * css_put() should be called to release a reference taken by 82 * css_get() 83 */ 84 85 extern void __css_put(struct cgroup_subsys_state *css); 86 static inline void css_put(struct cgroup_subsys_state *css) 87 { 88 if (!test_bit(CSS_ROOT, &css->flags)) 89 __css_put(css); 90 } 91 92 /* bits in struct cgroup flags field */ 93 enum { 94 /* Control Group is dead */ 95 CGRP_REMOVED, 96 /* Control Group has previously had a child cgroup or a task, 97 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */ 98 CGRP_RELEASABLE, 99 /* Control Group requires release notifications to userspace */ 100 CGRP_NOTIFY_ON_RELEASE, 101 }; 102 103 struct cgroup { 104 unsigned long flags; /* "unsigned long" so bitops work */ 105 106 /* count users of this cgroup. >0 means busy, but doesn't 107 * necessarily indicate the number of tasks in the 108 * cgroup */ 109 atomic_t count; 110 111 /* 112 * We link our 'sibling' struct into our parent's 'children'. 113 * Our children link their 'sibling' into our 'children'. 114 */ 115 struct list_head sibling; /* my parent's children */ 116 struct list_head children; /* my children */ 117 118 struct cgroup *parent; /* my parent */ 119 struct dentry *dentry; /* cgroup fs entry */ 120 121 /* Private pointers for each registered subsystem */ 122 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; 123 124 struct cgroupfs_root *root; 125 struct cgroup *top_cgroup; 126 127 /* 128 * List of cg_cgroup_links pointing at css_sets with 129 * tasks in this cgroup. Protected by css_set_lock 130 */ 131 struct list_head css_sets; 132 133 /* 134 * Linked list running through all cgroups that can 135 * potentially be reaped by the release agent. Protected by 136 * release_list_lock 137 */ 138 struct list_head release_list; 139 140 /* pids_mutex protects the fields below */ 141 struct rw_semaphore pids_mutex; 142 /* Array of process ids in the cgroup */ 143 pid_t *tasks_pids; 144 /* How many files are using the current tasks_pids array */ 145 int pids_use_count; 146 /* Length of the current tasks_pids array */ 147 int pids_length; 148 }; 149 150 /* A css_set is a structure holding pointers to a set of 151 * cgroup_subsys_state objects. This saves space in the task struct 152 * object and speeds up fork()/exit(), since a single inc/dec and a 153 * list_add()/del() can bump the reference count on the entire 154 * cgroup set for a task. 155 */ 156 157 struct css_set { 158 159 /* Reference count */ 160 atomic_t refcount; 161 162 /* 163 * List running through all cgroup groups in the same hash 164 * slot. Protected by css_set_lock 165 */ 166 struct hlist_node hlist; 167 168 /* 169 * List running through all tasks using this cgroup 170 * group. Protected by css_set_lock 171 */ 172 struct list_head tasks; 173 174 /* 175 * List of cg_cgroup_link objects on link chains from 176 * cgroups referenced from this css_set. Protected by 177 * css_set_lock 178 */ 179 struct list_head cg_links; 180 181 /* 182 * Set of subsystem states, one for each subsystem. This array 183 * is immutable after creation apart from the init_css_set 184 * during subsystem registration (at boot time). 185 */ 186 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; 187 }; 188 189 /* 190 * cgroup_map_cb is an abstract callback API for reporting map-valued 191 * control files 192 */ 193 194 struct cgroup_map_cb { 195 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value); 196 void *state; 197 }; 198 199 /* struct cftype: 200 * 201 * The files in the cgroup filesystem mostly have a very simple read/write 202 * handling, some common function will take care of it. Nevertheless some cases 203 * (read tasks) are special and therefore I define this structure for every 204 * kind of file. 205 * 206 * 207 * When reading/writing to a file: 208 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata 209 * - the 'cftype' of the file is file->f_dentry->d_fsdata 210 */ 211 212 #define MAX_CFTYPE_NAME 64 213 struct cftype { 214 /* By convention, the name should begin with the name of the 215 * subsystem, followed by a period */ 216 char name[MAX_CFTYPE_NAME]; 217 int private; 218 219 /* 220 * If non-zero, defines the maximum length of string that can 221 * be passed to write_string; defaults to 64 222 */ 223 size_t max_write_len; 224 225 int (*open)(struct inode *inode, struct file *file); 226 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft, 227 struct file *file, 228 char __user *buf, size_t nbytes, loff_t *ppos); 229 /* 230 * read_u64() is a shortcut for the common case of returning a 231 * single integer. Use it in place of read() 232 */ 233 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft); 234 /* 235 * read_s64() is a signed version of read_u64() 236 */ 237 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft); 238 /* 239 * read_map() is used for defining a map of key/value 240 * pairs. It should call cb->fill(cb, key, value) for each 241 * entry. The key/value pairs (and their ordering) should not 242 * change between reboots. 243 */ 244 int (*read_map)(struct cgroup *cont, struct cftype *cft, 245 struct cgroup_map_cb *cb); 246 /* 247 * read_seq_string() is used for outputting a simple sequence 248 * using seqfile. 249 */ 250 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft, 251 struct seq_file *m); 252 253 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft, 254 struct file *file, 255 const char __user *buf, size_t nbytes, loff_t *ppos); 256 257 /* 258 * write_u64() is a shortcut for the common case of accepting 259 * a single integer (as parsed by simple_strtoull) from 260 * userspace. Use in place of write(); return 0 or error. 261 */ 262 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val); 263 /* 264 * write_s64() is a signed version of write_u64() 265 */ 266 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val); 267 268 /* 269 * write_string() is passed a nul-terminated kernelspace 270 * buffer of maximum length determined by max_write_len. 271 * Returns 0 or -ve error code. 272 */ 273 int (*write_string)(struct cgroup *cgrp, struct cftype *cft, 274 const char *buffer); 275 /* 276 * trigger() callback can be used to get some kick from the 277 * userspace, when the actual string written is not important 278 * at all. The private field can be used to determine the 279 * kick type for multiplexing. 280 */ 281 int (*trigger)(struct cgroup *cgrp, unsigned int event); 282 283 int (*release)(struct inode *inode, struct file *file); 284 }; 285 286 struct cgroup_scanner { 287 struct cgroup *cg; 288 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan); 289 void (*process_task)(struct task_struct *p, 290 struct cgroup_scanner *scan); 291 struct ptr_heap *heap; 292 }; 293 294 /* Add a new file to the given cgroup directory. Should only be 295 * called by subsystems from within a populate() method */ 296 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys, 297 const struct cftype *cft); 298 299 /* Add a set of new files to the given cgroup directory. Should 300 * only be called by subsystems from within a populate() method */ 301 int cgroup_add_files(struct cgroup *cgrp, 302 struct cgroup_subsys *subsys, 303 const struct cftype cft[], 304 int count); 305 306 int cgroup_is_removed(const struct cgroup *cgrp); 307 308 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen); 309 310 int cgroup_task_count(const struct cgroup *cgrp); 311 312 /* Return true if the cgroup is a descendant of the current cgroup */ 313 int cgroup_is_descendant(const struct cgroup *cgrp); 314 315 /* Control Group subsystem type. See Documentation/cgroups.txt for details */ 316 317 struct cgroup_subsys { 318 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss, 319 struct cgroup *cgrp); 320 void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp); 321 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp); 322 int (*can_attach)(struct cgroup_subsys *ss, 323 struct cgroup *cgrp, struct task_struct *tsk); 324 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp, 325 struct cgroup *old_cgrp, struct task_struct *tsk); 326 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task); 327 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task); 328 int (*populate)(struct cgroup_subsys *ss, 329 struct cgroup *cgrp); 330 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp); 331 void (*bind)(struct cgroup_subsys *ss, struct cgroup *root); 332 /* 333 * This routine is called with the task_lock of mm->owner held 334 */ 335 void (*mm_owner_changed)(struct cgroup_subsys *ss, 336 struct cgroup *old, 337 struct cgroup *new, 338 struct task_struct *p); 339 int subsys_id; 340 int active; 341 int disabled; 342 int early_init; 343 #define MAX_CGROUP_TYPE_NAMELEN 32 344 const char *name; 345 346 /* Protected by RCU */ 347 struct cgroupfs_root *root; 348 349 struct list_head sibling; 350 }; 351 352 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys; 353 #include <linux/cgroup_subsys.h> 354 #undef SUBSYS 355 356 static inline struct cgroup_subsys_state *cgroup_subsys_state( 357 struct cgroup *cgrp, int subsys_id) 358 { 359 return cgrp->subsys[subsys_id]; 360 } 361 362 static inline struct cgroup_subsys_state *task_subsys_state( 363 struct task_struct *task, int subsys_id) 364 { 365 return rcu_dereference(task->cgroups->subsys[subsys_id]); 366 } 367 368 static inline struct cgroup* task_cgroup(struct task_struct *task, 369 int subsys_id) 370 { 371 return task_subsys_state(task, subsys_id)->cgroup; 372 } 373 374 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss, 375 char *nodename); 376 377 /* A cgroup_iter should be treated as an opaque object */ 378 struct cgroup_iter { 379 struct list_head *cg_link; 380 struct list_head *task; 381 }; 382 383 /* To iterate across the tasks in a cgroup: 384 * 385 * 1) call cgroup_iter_start to intialize an iterator 386 * 387 * 2) call cgroup_iter_next() to retrieve member tasks until it 388 * returns NULL or until you want to end the iteration 389 * 390 * 3) call cgroup_iter_end() to destroy the iterator. 391 * 392 * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset. 393 * - cgroup_scan_tasks() holds the css_set_lock when calling the test_task() 394 * callback, but not while calling the process_task() callback. 395 */ 396 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it); 397 struct task_struct *cgroup_iter_next(struct cgroup *cgrp, 398 struct cgroup_iter *it); 399 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it); 400 int cgroup_scan_tasks(struct cgroup_scanner *scan); 401 int cgroup_attach_task(struct cgroup *, struct task_struct *); 402 403 void cgroup_mm_owner_callbacks(struct task_struct *old, 404 struct task_struct *new); 405 406 #else /* !CONFIG_CGROUPS */ 407 408 static inline int cgroup_init_early(void) { return 0; } 409 static inline int cgroup_init(void) { return 0; } 410 static inline void cgroup_fork(struct task_struct *p) {} 411 static inline void cgroup_fork_callbacks(struct task_struct *p) {} 412 static inline void cgroup_post_fork(struct task_struct *p) {} 413 static inline void cgroup_exit(struct task_struct *p, int callbacks) {} 414 415 static inline void cgroup_lock(void) {} 416 static inline void cgroup_unlock(void) {} 417 static inline int cgroupstats_build(struct cgroupstats *stats, 418 struct dentry *dentry) 419 { 420 return -EINVAL; 421 } 422 423 static inline void cgroup_mm_owner_callbacks(struct task_struct *old, 424 struct task_struct *new) {} 425 426 #endif /* !CONFIG_CGROUPS */ 427 428 #endif /* _LINUX_CGROUP_H */ 429