include/linux/cgroup.h

   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 #include <linux/idr.h>
  19
  20 #ifdef CONFIG_CGROUPS
  21
  22 struct cgroupfs_root;
  23 struct cgroup_subsys;
  24 struct inode;
  25 struct cgroup;
  26 struct css_id;
  27
  28 extern int cgroup_init_early(void);
  29 extern int cgroup_init(void);
  30 extern void cgroup_lock(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_fork_callbacks(struct task_struct *p);
  35 extern void cgroup_post_fork(struct task_struct *p);
  36 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
  37 extern int cgroupstats_build(struct cgroupstats *stats,
  38                                 struct dentry *dentry);
  39
  40 extern const struct file_operations proc_cgroup_operations;
  41
  42 /* Define the enumeration of all cgroup subsystems */
  43 #define SUBSYS(_x) _x ## _subsys_id,
  44 enum cgroup_subsys_id {
  45 #include <linux/cgroup_subsys.h>
  46         CGROUP_SUBSYS_COUNT
  47 };
  48 #undef SUBSYS
  49
  50 /* Per-subsystem/per-cgroup state maintained by the system. */
  51 struct cgroup_subsys_state {
  52         /*
  53          * The cgroup that this subsystem is attached to. Useful
  54          * for subsystems that want to know about the cgroup
  55          * hierarchy structure
  56          */
  57         struct cgroup *cgroup;
  58
  59         /*
  60          * State maintained by the cgroup system to allow subsystems
  61          * to be "busy". Should be accessed via css_get(),
  62          * css_tryget() and and css_put().
  63          */
  64
  65         atomic_t refcnt;
  66
  67         unsigned long flags;
  68         /* ID for this css, if possible */
  69         struct css_id *id;
  70 };
  71
  72 /* bits in struct cgroup_subsys_state flags field */
  73 enum {
  74         CSS_ROOT, /* This CSS is the root of the subsystem */
  75         CSS_REMOVED, /* This CSS is dead */
  76 };
  77
  78 /*
  79  * Call css_get() to hold a reference on the css; it can be used
  80  * for a reference obtained via:
  81  * - an existing ref-counted reference to the css
  82  * - task->cgroups for a locked task
  83  */
  84
  85 static inline void css_get(struct cgroup_subsys_state *css)
  86 {
  87         /* We don't need to reference count the root state */
  88         if (!test_bit(CSS_ROOT, &css->flags))
  89                 atomic_inc(&css->refcnt);
  90 }
  91
  92 static inline bool css_is_removed(struct cgroup_subsys_state *css)
  93 {
  94         return test_bit(CSS_REMOVED, &css->flags);
  95 }
  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
 103 static inline bool css_tryget(struct cgroup_subsys_state *css)
 104 {
 105         if (test_bit(CSS_ROOT, &css->flags))
 106                 return true;
 107         while (!atomic_inc_not_zero(&css->refcnt)) {
 108                 if (test_bit(CSS_REMOVED, &css->flags))
 109                         return false;
 110                 cpu_relax();
 111         }
 112         return true;
 113 }
 114
 115 /*
 116  * css_put() should be called to release a reference taken by
 117  * css_get() or css_tryget()
 118  */
 119
 120 extern void __css_put(struct cgroup_subsys_state *css);
 121 static inline void css_put(struct cgroup_subsys_state *css)
 122 {
 123         if (!test_bit(CSS_ROOT, &css->flags))
 124                 __css_put(css);
 125 }
 126
 127 /* bits in struct cgroup flags field */
 128 enum {
 129         /* Control Group is dead */
 130         CGRP_REMOVED,
 131         /*
 132          * Control Group has previously had a child cgroup or a task,
 133          * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
 134          */
 135         CGRP_RELEASABLE,
 136         /* Control Group requires release notifications to userspace */
 137         CGRP_NOTIFY_ON_RELEASE,
 138         /*
 139          * A thread in rmdir() is wating for this cgroup.
 140          */
 141         CGRP_WAIT_ON_RMDIR,
 142 };
 143
 144 /* which pidlist file are we talking about? */
 145 enum cgroup_filetype {
 146         CGROUP_FILE_PROCS,
 147         CGROUP_FILE_TASKS,
 148 };
 149
 150 /*
 151  * A pidlist is a list of pids that virtually represents the contents of one
 152  * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
 153  * a pair (one each for procs, tasks) for each pid namespace that's relevant
 154  * to the cgroup.
 155  */
 156 struct cgroup_pidlist {
 157         /*
 158          * used to find which pidlist is wanted. doesn't change as long as
 159          * this particular list stays in the list.
 160          */
 161         struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
 162         /* array of xids */
 163         pid_t *list;
 164         /* how many elements the above list has */
 165         int length;
 166         /* how many files are using the current array */
 167         int use_count;
 168         /* each of these stored in a list by its cgroup */
 169         struct list_head links;
 170         /* pointer to the cgroup we belong to, for list removal purposes */
 171         struct cgroup *owner;
 172         /* protects the other fields */
 173         struct rw_semaphore mutex;
 174 };
 175
 176 struct cgroup {
 177         unsigned long flags;            /* "unsigned long" so bitops work */
 178
 179         /*
 180          * count users of this cgroup. >0 means busy, but doesn't
 181          * necessarily indicate the number of tasks in the cgroup
 182          */
 183         atomic_t count;
 184
 185         /*
 186          * We link our 'sibling' struct into our parent's 'children'.
 187          * Our children link their 'sibling' into our 'children'.
 188          */
 189         struct list_head sibling;       /* my parent's children */
 190         struct list_head children;      /* my children */
 191
 192         struct cgroup *parent;          /* my parent */
 193         struct dentry *dentry;          /* cgroup fs entry, RCU protected */
 194
 195         /* Private pointers for each registered subsystem */
 196         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 197
 198         struct cgroupfs_root *root;
 199         struct cgroup *top_cgroup;
 200
 201         /*
 202          * List of cg_cgroup_links pointing at css_sets with
 203          * tasks in this cgroup. Protected by css_set_lock
 204          */
 205         struct list_head css_sets;
 206
 207         /*
 208          * Linked list running through all cgroups that can
 209          * potentially be reaped by the release agent. Protected by
 210          * release_list_lock
 211          */
 212         struct list_head release_list;
 213
 214         /*
 215          * list of pidlists, up to two for each namespace (one for procs, one
 216          * for tasks); created on demand.
 217          */
 218         struct list_head pidlists;
 219         struct mutex pidlist_mutex;
 220
 221         /* For RCU-protected deletion */
 222         struct rcu_head rcu_head;
 223 };
 224
 225 /*
 226  * A css_set is a structure holding pointers to a set of
 227  * cgroup_subsys_state objects. This saves space in the task struct
 228  * object and speeds up fork()/exit(), since a single inc/dec and a
 229  * list_add()/del() can bump the reference count on the entire cgroup
 230  * set for a task.
 231  */
 232
 233 struct css_set {
 234
 235         /* Reference count */
 236         atomic_t refcount;
 237
 238         /*
 239          * List running through all cgroup groups in the same hash
 240          * slot. Protected by css_set_lock
 241          */
 242         struct hlist_node hlist;
 243
 244         /*
 245          * List running through all tasks using this cgroup
 246          * group. Protected by css_set_lock
 247          */
 248         struct list_head tasks;
 249
 250         /*
 251          * List of cg_cgroup_link objects on link chains from
 252          * cgroups referenced from this css_set. Protected by
 253          * css_set_lock
 254          */
 255         struct list_head cg_links;
 256
 257         /*
 258          * Set of subsystem states, one for each subsystem. This array
 259          * is immutable after creation apart from the init_css_set
 260          * during subsystem registration (at boot time).
 261          */
 262         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 263
 264         /* For RCU-protected deletion */
 265         struct rcu_head rcu_head;
 266 };
 267
 268 /*
 269  * cgroup_map_cb is an abstract callback API for reporting map-valued
 270  * control files
 271  */
 272
 273 struct cgroup_map_cb {
 274         int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
 275         void *state;
 276 };
 277
 278 /*
 279  * struct cftype: handler definitions for cgroup control files
 280  *
 281  * When reading/writing to a file:
 282  *      - the cgroup to use is file->f_dentry->d_parent->d_fsdata
 283  *      - the 'cftype' of the file is file->f_dentry->d_fsdata
 284  */
 285
 286 #define MAX_CFTYPE_NAME 64
 287 struct cftype {
 288         /*
 289          * By convention, the name should begin with the name of the
 290          * subsystem, followed by a period
 291          */
 292         char name[MAX_CFTYPE_NAME];
 293         int private;
 294         /*
 295          * If not 0, file mode is set to this value, otherwise it will
 296          * be figured out automatically
 297          */
 298         mode_t mode;
 299
 300         /*
 301          * If non-zero, defines the maximum length of string that can
 302          * be passed to write_string; defaults to 64
 303          */
 304         size_t max_write_len;
 305
 306         int (*open)(struct inode *inode, struct file *file);
 307         ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
 308                         struct file *file,
 309                         char __user *buf, size_t nbytes, loff_t *ppos);
 310         /*
 311          * read_u64() is a shortcut for the common case of returning a
 312          * single integer. Use it in place of read()
 313          */
 314         u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
 315         /*
 316          * read_s64() is a signed version of read_u64()
 317          */
 318         s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
 319         /*
 320          * read_map() is used for defining a map of key/value
 321          * pairs. It should call cb->fill(cb, key, value) for each
 322          * entry. The key/value pairs (and their ordering) should not
 323          * change between reboots.
 324          */
 325         int (*read_map)(struct cgroup *cont, struct cftype *cft,
 326                         struct cgroup_map_cb *cb);
 327         /*
 328          * read_seq_string() is used for outputting a simple sequence
 329          * using seqfile.
 330          */
 331         int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
 332                                struct seq_file *m);
 333
 334         ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
 335                          struct file *file,
 336                          const char __user *buf, size_t nbytes, loff_t *ppos);
 337
 338         /*
 339          * write_u64() is a shortcut for the common case of accepting
 340          * a single integer (as parsed by simple_strtoull) from
 341          * userspace. Use in place of write(); return 0 or error.
 342          */
 343         int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
 344         /*
 345          * write_s64() is a signed version of write_u64()
 346          */
 347         int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
 348
 349         /*
 350          * write_string() is passed a nul-terminated kernelspace
 351          * buffer of maximum length determined by max_write_len.
 352          * Returns 0 or -ve error code.
 353          */
 354         int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
 355                             const char *buffer);
 356         /*
 357          * trigger() callback can be used to get some kick from the
 358          * userspace, when the actual string written is not important
 359          * at all. The private field can be used to determine the
 360          * kick type for multiplexing.
 361          */
 362         int (*trigger)(struct cgroup *cgrp, unsigned int event);
 363
 364         int (*release)(struct inode *inode, struct file *file);
 365 };
 366
 367 struct cgroup_scanner {
 368         struct cgroup *cg;
 369         int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
 370         void (*process_task)(struct task_struct *p,
 371                         struct cgroup_scanner *scan);
 372         struct ptr_heap *heap;
 373         void *data;
 374 };
 375
 376 /*
 377  * Add a new file to the given cgroup directory. Should only be
 378  * called by subsystems from within a populate() method
 379  */
 380 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
 381                        const struct cftype *cft);
 382
 383 /*
 384  * Add a set of new files to the given cgroup directory. Should
 385  * only be called by subsystems from within a populate() method
 386  */
 387 int cgroup_add_files(struct cgroup *cgrp,
 388                         struct cgroup_subsys *subsys,
 389                         const struct cftype cft[],
 390                         int count);
 391
 392 int cgroup_is_removed(const struct cgroup *cgrp);
 393
 394 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
 395
 396 int cgroup_task_count(const struct cgroup *cgrp);
 397
 398 /* Return true if cgrp is a descendant of the task's cgroup */
 399 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
 400
 401 /*
 402  * When the subsys has to access css and may add permanent refcnt to css,
 403  * it should take care of racy conditions with rmdir(). Following set of
 404  * functions, is for stop/restart rmdir if necessary.
 405  * Because these will call css_get/put, "css" should be alive css.
 406  *
 407  *  cgroup_exclude_rmdir();
 408  *  ...do some jobs which may access arbitrary empty cgroup
 409  *  cgroup_release_and_wakeup_rmdir();
 410  *
 411  *  When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
 412  *  it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
 413  */
 414
 415 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
 416 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
 417
 418 /*
 419  * Control Group subsystem type.
 420  * See Documentation/cgroups/cgroups.txt for details
 421  */
 422
 423 struct cgroup_subsys {
 424         struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
 425                                                   struct cgroup *cgrp);
 426         int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 427         void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 428         int (*can_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 429                           struct task_struct *tsk, bool threadgroup);
 430         void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 431                         struct cgroup *old_cgrp, struct task_struct *tsk,
 432                         bool threadgroup);
 433         void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
 434         void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
 435         int (*populate)(struct cgroup_subsys *ss,
 436                         struct cgroup *cgrp);
 437         void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 438         void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
 439
 440         int subsys_id;
 441         int active;
 442         int disabled;
 443         int early_init;
 444         /*
 445          * True if this subsys uses ID. ID is not available before cgroup_init()
 446          * (not available in early_init time.)
 447          */
 448         bool use_id;
 449 #define MAX_CGROUP_TYPE_NAMELEN 32
 450         const char *name;
 451
 452         /*
 453          * Protects sibling/children links of cgroups in this
 454          * hierarchy, plus protects which hierarchy (or none) the
 455          * subsystem is a part of (i.e. root/sibling).  To avoid
 456          * potential deadlocks, the following operations should not be
 457          * undertaken while holding any hierarchy_mutex:
 458          *
 459          * - allocating memory
 460          * - initiating hotplug events
 461          */
 462         struct mutex hierarchy_mutex;
 463         struct lock_class_key subsys_key;
 464
 465         /*
 466          * Link to parent, and list entry in parent's children.
 467          * Protected by this->hierarchy_mutex and cgroup_lock()
 468          */
 469         struct cgroupfs_root *root;
 470         struct list_head sibling;
 471         /* used when use_id == true */
 472         struct idr idr;
 473         spinlock_t id_lock;
 474 };
 475
 476 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
 477 #include <linux/cgroup_subsys.h>
 478 #undef SUBSYS
 479
 480 static inline struct cgroup_subsys_state *cgroup_subsys_state(
 481         struct cgroup *cgrp, int subsys_id)
 482 {
 483         return cgrp->subsys[subsys_id];
 484 }
 485
 486 static inline struct cgroup_subsys_state *task_subsys_state(
 487         struct task_struct *task, int subsys_id)
 488 {
 489         return rcu_dereference(task->cgroups->subsys[subsys_id]);
 490 }
 491
 492 static inline struct cgroup* task_cgroup(struct task_struct *task,
 493                                                int subsys_id)
 494 {
 495         return task_subsys_state(task, subsys_id)->cgroup;
 496 }
 497
 498 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
 499                                                         char *nodename);
 500
 501 /* A cgroup_iter should be treated as an opaque object */
 502 struct cgroup_iter {
 503         struct list_head *cg_link;
 504         struct list_head *task;
 505 };
 506
 507 /*
 508  * To iterate across the tasks in a cgroup:
 509  *
 510  * 1) call cgroup_iter_start to intialize an iterator
 511  *
 512  * 2) call cgroup_iter_next() to retrieve member tasks until it
 513  *    returns NULL or until you want to end the iteration
 514  *
 515  * 3) call cgroup_iter_end() to destroy the iterator.
 516  *
 517  * Or, call cgroup_scan_tasks() to iterate through every task in a
 518  * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
 519  * the test_task() callback, but not while calling the process_task()
 520  * callback.
 521  */
 522 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
 523 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
 524                                         struct cgroup_iter *it);
 525 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
 526 int cgroup_scan_tasks(struct cgroup_scanner *scan);
 527 int cgroup_attach_task(struct cgroup *, struct task_struct *);
 528
 529 /*
 530  * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
 531  * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
 532  * CSS ID is assigned at cgroup allocation (create) automatically
 533  * and removed when subsys calls free_css_id() function. This is because
 534  * the lifetime of cgroup_subsys_state is subsys's matter.
 535  *
 536  * Looking up and scanning function should be called under rcu_read_lock().
 537  * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
 538  * But the css returned by this routine can be "not populated yet" or "being
 539  * destroyed". The caller should check css and cgroup's status.
 540  */
 541
 542 /*
 543  * Typically Called at ->destroy(), or somewhere the subsys frees
 544  * cgroup_subsys_state.
 545  */
 546 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
 547
 548 /* Find a cgroup_subsys_state which has given ID */
 549
 550 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
 551
 552 /*
 553  * Get a cgroup whose id is greater than or equal to id under tree of root.
 554  * Returning a cgroup_subsys_state or NULL.
 555  */
 556 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
 557                 struct cgroup_subsys_state *root, int *foundid);
 558
 559 /* Returns true if root is ancestor of cg */
 560 bool css_is_ancestor(struct cgroup_subsys_state *cg,
 561                      const struct cgroup_subsys_state *root);
 562
 563 /* Get id and depth of css */
 564 unsigned short css_id(struct cgroup_subsys_state *css);
 565 unsigned short css_depth(struct cgroup_subsys_state *css);
 566
 567 #else /* !CONFIG_CGROUPS */
 568
 569 static inline int cgroup_init_early(void) { return 0; }
 570 static inline int cgroup_init(void) { return 0; }
 571 static inline void cgroup_fork(struct task_struct *p) {}
 572 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
 573 static inline void cgroup_post_fork(struct task_struct *p) {}
 574 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
 575
 576 static inline void cgroup_lock(void) {}
 577 static inline void cgroup_unlock(void) {}
 578 static inline int cgroupstats_build(struct cgroupstats *stats,
 579                                         struct dentry *dentry)
 580 {
 581         return -EINVAL;
 582 }
 583
 584 #endif /* !CONFIG_CGROUPS */
 585
 586 #endif /* _LINUX_CGROUP_H */