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 int cgroup_lock_is_held(void);
  32 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
  33 extern void cgroup_unlock(void);
  34 extern void cgroup_fork(struct task_struct *p);
  35 extern void cgroup_fork_callbacks(struct task_struct *p);
  36 extern void cgroup_post_fork(struct task_struct *p);
  37 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
  38 extern int cgroupstats_build(struct cgroupstats *stats,
  39                                 struct dentry *dentry);
  40
  41 extern const struct file_operations proc_cgroup_operations;
  42
  43 /* Define the enumeration of all cgroup subsystems */
  44 #define SUBSYS(_x) _x ## _subsys_id,
  45 enum cgroup_subsys_id {
  46 #include <linux/cgroup_subsys.h>
  47         CGROUP_SUBSYS_COUNT
  48 };
  49 #undef SUBSYS
  50
  51 /* Per-subsystem/per-cgroup state maintained by the system. */
  52 struct cgroup_subsys_state {
  53         /*
  54          * The cgroup that this subsystem is attached to. Useful
  55          * for subsystems that want to know about the cgroup
  56          * hierarchy structure
  57          */
  58         struct cgroup *cgroup;
  59
  60         /*
  61          * State maintained by the cgroup system to allow subsystems
  62          * to be "busy". Should be accessed via css_get(),
  63          * css_tryget() and and css_put().
  64          */
  65
  66         atomic_t refcnt;
  67
  68         unsigned long flags;
  69         /* ID for this css, if possible */
  70         struct css_id *id;
  71 };
  72
  73 /* bits in struct cgroup_subsys_state flags field */
  74 enum {
  75         CSS_ROOT, /* This CSS is the root of the subsystem */
  76         CSS_REMOVED, /* This CSS is dead */
  77 };
  78
  79 /*
  80  * Call css_get() to hold a reference on the css; it can be used
  81  * for a reference obtained via:
  82  * - an existing ref-counted reference to the css
  83  * - task->cgroups for a locked task
  84  */
  85
  86 static inline void css_get(struct cgroup_subsys_state *css)
  87 {
  88         /* We don't need to reference count the root state */
  89         if (!test_bit(CSS_ROOT, &css->flags))
  90                 atomic_inc(&css->refcnt);
  91 }
  92
  93 static inline bool css_is_removed(struct cgroup_subsys_state *css)
  94 {
  95         return test_bit(CSS_REMOVED, &css->flags);
  96 }
  97
  98 /*
  99  * Call css_tryget() to take a reference on a css if your existing
 100  * (known-valid) reference isn't already ref-counted. Returns false if
 101  * the css has been destroyed.
 102  */
 103
 104 static inline bool css_tryget(struct cgroup_subsys_state *css)
 105 {
 106         if (test_bit(CSS_ROOT, &css->flags))
 107                 return true;
 108         while (!atomic_inc_not_zero(&css->refcnt)) {
 109                 if (test_bit(CSS_REMOVED, &css->flags))
 110                         return false;
 111                 cpu_relax();
 112         }
 113         return true;
 114 }
 115
 116 /*
 117  * css_put() should be called to release a reference taken by
 118  * css_get() or css_tryget()
 119  */
 120
 121 extern void __css_put(struct cgroup_subsys_state *css);
 122 static inline void css_put(struct cgroup_subsys_state *css)
 123 {
 124         if (!test_bit(CSS_ROOT, &css->flags))
 125                 __css_put(css);
 126 }
 127
 128 /* bits in struct cgroup flags field */
 129 enum {
 130         /* Control Group is dead */
 131         CGRP_REMOVED,
 132         /*
 133          * Control Group has previously had a child cgroup or a task,
 134          * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
 135          */
 136         CGRP_RELEASABLE,
 137         /* Control Group requires release notifications to userspace */
 138         CGRP_NOTIFY_ON_RELEASE,
 139         /*
 140          * A thread in rmdir() is wating for this cgroup.
 141          */
 142         CGRP_WAIT_ON_RMDIR,
 143 };
 144
 145 /* which pidlist file are we talking about? */
 146 enum cgroup_filetype {
 147         CGROUP_FILE_PROCS,
 148         CGROUP_FILE_TASKS,
 149 };
 150
 151 /*
 152  * A pidlist is a list of pids that virtually represents the contents of one
 153  * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
 154  * a pair (one each for procs, tasks) for each pid namespace that's relevant
 155  * to the cgroup.
 156  */
 157 struct cgroup_pidlist {
 158         /*
 159          * used to find which pidlist is wanted. doesn't change as long as
 160          * this particular list stays in the list.
 161          */
 162         struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
 163         /* array of xids */
 164         pid_t *list;
 165         /* how many elements the above list has */
 166         int length;
 167         /* how many files are using the current array */
 168         int use_count;
 169         /* each of these stored in a list by its cgroup */
 170         struct list_head links;
 171         /* pointer to the cgroup we belong to, for list removal purposes */
 172         struct cgroup *owner;
 173         /* protects the other fields */
 174         struct rw_semaphore mutex;
 175 };
 176
 177 struct cgroup {
 178         unsigned long flags;            /* "unsigned long" so bitops work */
 179
 180         /*
 181          * count users of this cgroup. >0 means busy, but doesn't
 182          * necessarily indicate the number of tasks in the cgroup
 183          */
 184         atomic_t count;
 185
 186         /*
 187          * We link our 'sibling' struct into our parent's 'children'.
 188          * Our children link their 'sibling' into our 'children'.
 189          */
 190         struct list_head sibling;       /* my parent's children */
 191         struct list_head children;      /* my children */
 192
 193         struct cgroup *parent;          /* my parent */
 194         struct dentry *dentry;          /* cgroup fs entry, RCU protected */
 195
 196         /* Private pointers for each registered subsystem */
 197         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 198
 199         struct cgroupfs_root *root;
 200         struct cgroup *top_cgroup;
 201
 202         /*
 203          * List of cg_cgroup_links pointing at css_sets with
 204          * tasks in this cgroup. Protected by css_set_lock
 205          */
 206         struct list_head css_sets;
 207
 208         /*
 209          * Linked list running through all cgroups that can
 210          * potentially be reaped by the release agent. Protected by
 211          * release_list_lock
 212          */
 213         struct list_head release_list;
 214
 215         /*
 216          * list of pidlists, up to two for each namespace (one for procs, one
 217          * for tasks); created on demand.
 218          */
 219         struct list_head pidlists;
 220         struct mutex pidlist_mutex;
 221
 222         /* For RCU-protected deletion */
 223         struct rcu_head rcu_head;
 224 };
 225
 226 /*
 227  * A css_set is a structure holding pointers to a set of
 228  * cgroup_subsys_state objects. This saves space in the task struct
 229  * object and speeds up fork()/exit(), since a single inc/dec and a
 230  * list_add()/del() can bump the reference count on the entire cgroup
 231  * set for a task.
 232  */
 233
 234 struct css_set {
 235
 236         /* Reference count */
 237         atomic_t refcount;
 238
 239         /*
 240          * List running through all cgroup groups in the same hash
 241          * slot. Protected by css_set_lock
 242          */
 243         struct hlist_node hlist;
 244
 245         /*
 246          * List running through all tasks using this cgroup
 247          * group. Protected by css_set_lock
 248          */
 249         struct list_head tasks;
 250
 251         /*
 252          * List of cg_cgroup_link objects on link chains from
 253          * cgroups referenced from this css_set. Protected by
 254          * css_set_lock
 255          */
 256         struct list_head cg_links;
 257
 258         /*
 259          * Set of subsystem states, one for each subsystem. This array
 260          * is immutable after creation apart from the init_css_set
 261          * during subsystem registration (at boot time).
 262          */
 263         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 264
 265         /* For RCU-protected deletion */
 266         struct rcu_head rcu_head;
 267 };
 268
 269 /*
 270  * cgroup_map_cb is an abstract callback API for reporting map-valued
 271  * control files
 272  */
 273
 274 struct cgroup_map_cb {
 275         int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
 276         void *state;
 277 };
 278
 279 /*
 280  * struct cftype: handler definitions for cgroup control files
 281  *
 282  * When reading/writing to a file:
 283  *      - the cgroup to use is file->f_dentry->d_parent->d_fsdata
 284  *      - the 'cftype' of the file is file->f_dentry->d_fsdata
 285  */
 286
 287 #define MAX_CFTYPE_NAME 64
 288 struct cftype {
 289         /*
 290          * By convention, the name should begin with the name of the
 291          * subsystem, followed by a period
 292          */
 293         char name[MAX_CFTYPE_NAME];
 294         int private;
 295         /*
 296          * If not 0, file mode is set to this value, otherwise it will
 297          * be figured out automatically
 298          */
 299         mode_t mode;
 300
 301         /*
 302          * If non-zero, defines the maximum length of string that can
 303          * be passed to write_string; defaults to 64
 304          */
 305         size_t max_write_len;
 306
 307         int (*open)(struct inode *inode, struct file *file);
 308         ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
 309                         struct file *file,
 310                         char __user *buf, size_t nbytes, loff_t *ppos);
 311         /*
 312          * read_u64() is a shortcut for the common case of returning a
 313          * single integer. Use it in place of read()
 314          */
 315         u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
 316         /*
 317          * read_s64() is a signed version of read_u64()
 318          */
 319         s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
 320         /*
 321          * read_map() is used for defining a map of key/value
 322          * pairs. It should call cb->fill(cb, key, value) for each
 323          * entry. The key/value pairs (and their ordering) should not
 324          * change between reboots.
 325          */
 326         int (*read_map)(struct cgroup *cont, struct cftype *cft,
 327                         struct cgroup_map_cb *cb);
 328         /*
 329          * read_seq_string() is used for outputting a simple sequence
 330          * using seqfile.
 331          */
 332         int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
 333                                struct seq_file *m);
 334
 335         ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
 336                          struct file *file,
 337                          const char __user *buf, size_t nbytes, loff_t *ppos);
 338
 339         /*
 340          * write_u64() is a shortcut for the common case of accepting
 341          * a single integer (as parsed by simple_strtoull) from
 342          * userspace. Use in place of write(); return 0 or error.
 343          */
 344         int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
 345         /*
 346          * write_s64() is a signed version of write_u64()
 347          */
 348         int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
 349
 350         /*
 351          * write_string() is passed a nul-terminated kernelspace
 352          * buffer of maximum length determined by max_write_len.
 353          * Returns 0 or -ve error code.
 354          */
 355         int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
 356                             const char *buffer);
 357         /*
 358          * trigger() callback can be used to get some kick from the
 359          * userspace, when the actual string written is not important
 360          * at all. The private field can be used to determine the
 361          * kick type for multiplexing.
 362          */
 363         int (*trigger)(struct cgroup *cgrp, unsigned int event);
 364
 365         int (*release)(struct inode *inode, struct file *file);
 366 };
 367
 368 struct cgroup_scanner {
 369         struct cgroup *cg;
 370         int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
 371         void (*process_task)(struct task_struct *p,
 372                         struct cgroup_scanner *scan);
 373         struct ptr_heap *heap;
 374         void *data;
 375 };
 376
 377 /*
 378  * Add a new file to the given cgroup directory. Should only be
 379  * called by subsystems from within a populate() method
 380  */
 381 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
 382                        const struct cftype *cft);
 383
 384 /*
 385  * Add a set of new files to the given cgroup directory. Should
 386  * only be called by subsystems from within a populate() method
 387  */
 388 int cgroup_add_files(struct cgroup *cgrp,
 389                         struct cgroup_subsys *subsys,
 390                         const struct cftype cft[],
 391                         int count);
 392
 393 int cgroup_is_removed(const struct cgroup *cgrp);
 394
 395 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
 396
 397 int cgroup_task_count(const struct cgroup *cgrp);
 398
 399 /* Return true if cgrp is a descendant of the task's cgroup */
 400 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
 401
 402 /*
 403  * When the subsys has to access css and may add permanent refcnt to css,
 404  * it should take care of racy conditions with rmdir(). Following set of
 405  * functions, is for stop/restart rmdir if necessary.
 406  * Because these will call css_get/put, "css" should be alive css.
 407  *
 408  *  cgroup_exclude_rmdir();
 409  *  ...do some jobs which may access arbitrary empty cgroup
 410  *  cgroup_release_and_wakeup_rmdir();
 411  *
 412  *  When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
 413  *  it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
 414  */
 415
 416 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
 417 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
 418
 419 /*
 420  * Control Group subsystem type.
 421  * See Documentation/cgroups/cgroups.txt for details
 422  */
 423
 424 struct cgroup_subsys {
 425         struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
 426                                                   struct cgroup *cgrp);
 427         int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 428         void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 429         int (*can_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 430                           struct task_struct *tsk, bool threadgroup);
 431         void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 432                         struct cgroup *old_cgrp, struct task_struct *tsk,
 433                         bool threadgroup);
 434         void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
 435         void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
 436         int (*populate)(struct cgroup_subsys *ss,
 437                         struct cgroup *cgrp);
 438         void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 439         void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
 440
 441         int subsys_id;
 442         int active;
 443         int disabled;
 444         int early_init;
 445         /*
 446          * True if this subsys uses ID. ID is not available before cgroup_init()
 447          * (not available in early_init time.)
 448          */
 449         bool use_id;
 450 #define MAX_CGROUP_TYPE_NAMELEN 32
 451         const char *name;
 452
 453         /*
 454          * Protects sibling/children links of cgroups in this
 455          * hierarchy, plus protects which hierarchy (or none) the
 456          * subsystem is a part of (i.e. root/sibling).  To avoid
 457          * potential deadlocks, the following operations should not be
 458          * undertaken while holding any hierarchy_mutex:
 459          *
 460          * - allocating memory
 461          * - initiating hotplug events
 462          */
 463         struct mutex hierarchy_mutex;
 464         struct lock_class_key subsys_key;
 465
 466         /*
 467          * Link to parent, and list entry in parent's children.
 468          * Protected by this->hierarchy_mutex and cgroup_lock()
 469          */
 470         struct cgroupfs_root *root;
 471         struct list_head sibling;
 472         /* used when use_id == true */
 473         struct idr idr;
 474         spinlock_t id_lock;
 475 };
 476
 477 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
 478 #include <linux/cgroup_subsys.h>
 479 #undef SUBSYS
 480
 481 static inline struct cgroup_subsys_state *cgroup_subsys_state(
 482         struct cgroup *cgrp, int subsys_id)
 483 {
 484         return cgrp->subsys[subsys_id];
 485 }
 486
 487 static inline struct cgroup_subsys_state *task_subsys_state(
 488         struct task_struct *task, int subsys_id)
 489 {
 490         return rcu_dereference_check(task->cgroups->subsys[subsys_id],
 491                                      rcu_read_lock_held() ||
 492                                      cgroup_lock_is_held());
 493 }
 494
 495 static inline struct cgroup* task_cgroup(struct task_struct *task,
 496                                                int subsys_id)
 497 {
 498         return task_subsys_state(task, subsys_id)->cgroup;
 499 }
 500
 501 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
 502                                                         char *nodename);
 503
 504 /* A cgroup_iter should be treated as an opaque object */
 505 struct cgroup_iter {
 506         struct list_head *cg_link;
 507         struct list_head *task;
 508 };
 509
 510 /*
 511  * To iterate across the tasks in a cgroup:
 512  *
 513  * 1) call cgroup_iter_start to intialize an iterator
 514  *
 515  * 2) call cgroup_iter_next() to retrieve member tasks until it
 516  *    returns NULL or until you want to end the iteration
 517  *
 518  * 3) call cgroup_iter_end() to destroy the iterator.
 519  *
 520  * Or, call cgroup_scan_tasks() to iterate through every task in a
 521  * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
 522  * the test_task() callback, but not while calling the process_task()
 523  * callback.
 524  */
 525 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
 526 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
 527                                         struct cgroup_iter *it);
 528 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
 529 int cgroup_scan_tasks(struct cgroup_scanner *scan);
 530 int cgroup_attach_task(struct cgroup *, struct task_struct *);
 531
 532 /*
 533  * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
 534  * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
 535  * CSS ID is assigned at cgroup allocation (create) automatically
 536  * and removed when subsys calls free_css_id() function. This is because
 537  * the lifetime of cgroup_subsys_state is subsys's matter.
 538  *
 539  * Looking up and scanning function should be called under rcu_read_lock().
 540  * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
 541  * But the css returned by this routine can be "not populated yet" or "being
 542  * destroyed". The caller should check css and cgroup's status.
 543  */
 544
 545 /*
 546  * Typically Called at ->destroy(), or somewhere the subsys frees
 547  * cgroup_subsys_state.
 548  */
 549 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
 550
 551 /* Find a cgroup_subsys_state which has given ID */
 552
 553 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
 554
 555 /*
 556  * Get a cgroup whose id is greater than or equal to id under tree of root.
 557  * Returning a cgroup_subsys_state or NULL.
 558  */
 559 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
 560                 struct cgroup_subsys_state *root, int *foundid);
 561
 562 /* Returns true if root is ancestor of cg */
 563 bool css_is_ancestor(struct cgroup_subsys_state *cg,
 564                      const struct cgroup_subsys_state *root);
 565
 566 /* Get id and depth of css */
 567 unsigned short css_id(struct cgroup_subsys_state *css);
 568 unsigned short css_depth(struct cgroup_subsys_state *css);
 569
 570 #else /* !CONFIG_CGROUPS */
 571
 572 static inline int cgroup_init_early(void) { return 0; }
 573 static inline int cgroup_init(void) { return 0; }
 574 static inline void cgroup_fork(struct task_struct *p) {}
 575 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
 576 static inline void cgroup_post_fork(struct task_struct *p) {}
 577 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
 578
 579 static inline void cgroup_lock(void) {}
 580 static inline void cgroup_unlock(void) {}
 581 static inline int cgroupstats_build(struct cgroupstats *stats,
 582                                         struct dentry *dentry)
 583 {
 584         return -EINVAL;
 585 }
 586
 587 #endif /* !CONFIG_CGROUPS */
 588
 589 #endif /* _LINUX_CGROUP_H */