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 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 struct cgroup {
 145         unsigned long flags;            /* "unsigned long" so bitops work */
 146
 147         /*
 148          * count users of this cgroup. >0 means busy, but doesn't
 149          * necessarily indicate the number of tasks in the cgroup
 150          */
 151         atomic_t count;
 152
 153         /*
 154          * We link our 'sibling' struct into our parent's 'children'.
 155          * Our children link their 'sibling' into our 'children'.
 156          */
 157         struct list_head sibling;       /* my parent's children */
 158         struct list_head children;      /* my children */
 159
 160         struct cgroup *parent;          /* my parent */
 161         struct dentry *dentry;          /* cgroup fs entry, RCU protected */
 162
 163         /* Private pointers for each registered subsystem */
 164         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 165
 166         struct cgroupfs_root *root;
 167         struct cgroup *top_cgroup;
 168
 169         /*
 170          * List of cg_cgroup_links pointing at css_sets with
 171          * tasks in this cgroup. Protected by css_set_lock
 172          */
 173         struct list_head css_sets;
 174
 175         /*
 176          * Linked list running through all cgroups that can
 177          * potentially be reaped by the release agent. Protected by
 178          * release_list_lock
 179          */
 180         struct list_head release_list;
 181
 182         /* pids_mutex protects the fields below */
 183         struct rw_semaphore pids_mutex;
 184         /* Array of process ids in the cgroup */
 185         pid_t *tasks_pids;
 186         /* How many files are using the current tasks_pids array */
 187         int pids_use_count;
 188         /* Length of the current tasks_pids array */
 189         int pids_length;
 190
 191         /* For RCU-protected deletion */
 192         struct rcu_head rcu_head;
 193 };
 194
 195 /*
 196  * A css_set is a structure holding pointers to a set of
 197  * cgroup_subsys_state objects. This saves space in the task struct
 198  * object and speeds up fork()/exit(), since a single inc/dec and a
 199  * list_add()/del() can bump the reference count on the entire cgroup
 200  * set for a task.
 201  */
 202
 203 struct css_set {
 204
 205         /* Reference count */
 206         atomic_t refcount;
 207
 208         /*
 209          * List running through all cgroup groups in the same hash
 210          * slot. Protected by css_set_lock
 211          */
 212         struct hlist_node hlist;
 213
 214         /*
 215          * List running through all tasks using this cgroup
 216          * group. Protected by css_set_lock
 217          */
 218         struct list_head tasks;
 219
 220         /*
 221          * List of cg_cgroup_link objects on link chains from
 222          * cgroups referenced from this css_set. Protected by
 223          * css_set_lock
 224          */
 225         struct list_head cg_links;
 226
 227         /*
 228          * Set of subsystem states, one for each subsystem. This array
 229          * is immutable after creation apart from the init_css_set
 230          * during subsystem registration (at boot time).
 231          */
 232         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 233 };
 234
 235 /*
 236  * cgroup_map_cb is an abstract callback API for reporting map-valued
 237  * control files
 238  */
 239
 240 struct cgroup_map_cb {
 241         int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
 242         void *state;
 243 };
 244
 245 /*
 246  * struct cftype: handler definitions for cgroup control files
 247  *
 248  * When reading/writing to a file:
 249  *      - the cgroup to use is file->f_dentry->d_parent->d_fsdata
 250  *      - the 'cftype' of the file is file->f_dentry->d_fsdata
 251  */
 252
 253 #define MAX_CFTYPE_NAME 64
 254 struct cftype {
 255         /*
 256          * By convention, the name should begin with the name of the
 257          * subsystem, followed by a period
 258          */
 259         char name[MAX_CFTYPE_NAME];
 260         int private;
 261         /*
 262          * If not 0, file mode is set to this value, otherwise it will
 263          * be figured out automatically
 264          */
 265         mode_t mode;
 266
 267         /*
 268          * If non-zero, defines the maximum length of string that can
 269          * be passed to write_string; defaults to 64
 270          */
 271         size_t max_write_len;
 272
 273         int (*open)(struct inode *inode, struct file *file);
 274         ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
 275                         struct file *file,
 276                         char __user *buf, size_t nbytes, loff_t *ppos);
 277         /*
 278          * read_u64() is a shortcut for the common case of returning a
 279          * single integer. Use it in place of read()
 280          */
 281         u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
 282         /*
 283          * read_s64() is a signed version of read_u64()
 284          */
 285         s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
 286         /*
 287          * read_map() is used for defining a map of key/value
 288          * pairs. It should call cb->fill(cb, key, value) for each
 289          * entry. The key/value pairs (and their ordering) should not
 290          * change between reboots.
 291          */
 292         int (*read_map)(struct cgroup *cont, struct cftype *cft,
 293                         struct cgroup_map_cb *cb);
 294         /*
 295          * read_seq_string() is used for outputting a simple sequence
 296          * using seqfile.
 297          */
 298         int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
 299                                struct seq_file *m);
 300
 301         ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
 302                          struct file *file,
 303                          const char __user *buf, size_t nbytes, loff_t *ppos);
 304
 305         /*
 306          * write_u64() is a shortcut for the common case of accepting
 307          * a single integer (as parsed by simple_strtoull) from
 308          * userspace. Use in place of write(); return 0 or error.
 309          */
 310         int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
 311         /*
 312          * write_s64() is a signed version of write_u64()
 313          */
 314         int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
 315
 316         /*
 317          * write_string() is passed a nul-terminated kernelspace
 318          * buffer of maximum length determined by max_write_len.
 319          * Returns 0 or -ve error code.
 320          */
 321         int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
 322                             const char *buffer);
 323         /*
 324          * trigger() callback can be used to get some kick from the
 325          * userspace, when the actual string written is not important
 326          * at all. The private field can be used to determine the
 327          * kick type for multiplexing.
 328          */
 329         int (*trigger)(struct cgroup *cgrp, unsigned int event);
 330
 331         int (*release)(struct inode *inode, struct file *file);
 332 };
 333
 334 struct cgroup_scanner {
 335         struct cgroup *cg;
 336         int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
 337         void (*process_task)(struct task_struct *p,
 338                         struct cgroup_scanner *scan);
 339         struct ptr_heap *heap;
 340         void *data;
 341 };
 342
 343 /*
 344  * Add a new file to the given cgroup directory. Should only be
 345  * called by subsystems from within a populate() method
 346  */
 347 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
 348                        const struct cftype *cft);
 349
 350 /*
 351  * Add a set of new files to the given cgroup directory. Should
 352  * only be called by subsystems from within a populate() method
 353  */
 354 int cgroup_add_files(struct cgroup *cgrp,
 355                         struct cgroup_subsys *subsys,
 356                         const struct cftype cft[],
 357                         int count);
 358
 359 int cgroup_is_removed(const struct cgroup *cgrp);
 360
 361 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
 362
 363 int cgroup_task_count(const struct cgroup *cgrp);
 364
 365 /* Return true if cgrp is a descendant of the task's cgroup */
 366 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
 367
 368 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
 369
 370 struct cgroup_subsys {
 371         struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
 372                                                   struct cgroup *cgrp);
 373         int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 374         void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 375         int (*can_attach)(struct cgroup_subsys *ss,
 376                           struct cgroup *cgrp, struct task_struct *tsk);
 377         void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 378                         struct cgroup *old_cgrp, struct task_struct *tsk);
 379         void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
 380         void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
 381         int (*populate)(struct cgroup_subsys *ss,
 382                         struct cgroup *cgrp);
 383         void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 384         void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
 385
 386         int subsys_id;
 387         int active;
 388         int disabled;
 389         int early_init;
 390         /*
 391          * True if this subsys uses ID. ID is not available before cgroup_init()
 392          * (not available in early_init time.)
 393          */
 394         bool use_id;
 395 #define MAX_CGROUP_TYPE_NAMELEN 32
 396         const char *name;
 397
 398         /*
 399          * Protects sibling/children links of cgroups in this
 400          * hierarchy, plus protects which hierarchy (or none) the
 401          * subsystem is a part of (i.e. root/sibling).  To avoid
 402          * potential deadlocks, the following operations should not be
 403          * undertaken while holding any hierarchy_mutex:
 404          *
 405          * - allocating memory
 406          * - initiating hotplug events
 407          */
 408         struct mutex hierarchy_mutex;
 409         struct lock_class_key subsys_key;
 410
 411         /*
 412          * Link to parent, and list entry in parent's children.
 413          * Protected by this->hierarchy_mutex and cgroup_lock()
 414          */
 415         struct cgroupfs_root *root;
 416         struct list_head sibling;
 417         /* used when use_id == true */
 418         struct idr idr;
 419         spinlock_t id_lock;
 420 };
 421
 422 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
 423 #include <linux/cgroup_subsys.h>
 424 #undef SUBSYS
 425
 426 static inline struct cgroup_subsys_state *cgroup_subsys_state(
 427         struct cgroup *cgrp, int subsys_id)
 428 {
 429         return cgrp->subsys[subsys_id];
 430 }
 431
 432 static inline struct cgroup_subsys_state *task_subsys_state(
 433         struct task_struct *task, int subsys_id)
 434 {
 435         return rcu_dereference(task->cgroups->subsys[subsys_id]);
 436 }
 437
 438 static inline struct cgroup* task_cgroup(struct task_struct *task,
 439                                                int subsys_id)
 440 {
 441         return task_subsys_state(task, subsys_id)->cgroup;
 442 }
 443
 444 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
 445                                                         char *nodename);
 446
 447 /* A cgroup_iter should be treated as an opaque object */
 448 struct cgroup_iter {
 449         struct list_head *cg_link;
 450         struct list_head *task;
 451 };
 452
 453 /*
 454  * To iterate across the tasks in a cgroup:
 455  *
 456  * 1) call cgroup_iter_start to intialize an iterator
 457  *
 458  * 2) call cgroup_iter_next() to retrieve member tasks until it
 459  *    returns NULL or until you want to end the iteration
 460  *
 461  * 3) call cgroup_iter_end() to destroy the iterator.
 462  *
 463  * Or, call cgroup_scan_tasks() to iterate through every task in a
 464  * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
 465  * the test_task() callback, but not while calling the process_task()
 466  * callback.
 467  */
 468 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
 469 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
 470                                         struct cgroup_iter *it);
 471 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
 472 int cgroup_scan_tasks(struct cgroup_scanner *scan);
 473 int cgroup_attach_task(struct cgroup *, struct task_struct *);
 474
 475 /*
 476  * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
 477  * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
 478  * CSS ID is assigned at cgroup allocation (create) automatically
 479  * and removed when subsys calls free_css_id() function. This is because
 480  * the lifetime of cgroup_subsys_state is subsys's matter.
 481  *
 482  * Looking up and scanning function should be called under rcu_read_lock().
 483  * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
 484  * But the css returned by this routine can be "not populated yet" or "being
 485  * destroyed". The caller should check css and cgroup's status.
 486  */
 487
 488 /*
 489  * Typically Called at ->destroy(), or somewhere the subsys frees
 490  * cgroup_subsys_state.
 491  */
 492 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
 493
 494 /* Find a cgroup_subsys_state which has given ID */
 495
 496 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
 497
 498 /*
 499  * Get a cgroup whose id is greater than or equal to id under tree of root.
 500  * Returning a cgroup_subsys_state or NULL.
 501  */
 502 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
 503                 struct cgroup_subsys_state *root, int *foundid);
 504
 505 /* Returns true if root is ancestor of cg */
 506 bool css_is_ancestor(struct cgroup_subsys_state *cg,
 507                      const struct cgroup_subsys_state *root);
 508
 509 /* Get id and depth of css */
 510 unsigned short css_id(struct cgroup_subsys_state *css);
 511 unsigned short css_depth(struct cgroup_subsys_state *css);
 512
 513 #else /* !CONFIG_CGROUPS */
 514
 515 static inline int cgroup_init_early(void) { return 0; }
 516 static inline int cgroup_init(void) { return 0; }
 517 static inline void cgroup_fork(struct task_struct *p) {}
 518 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
 519 static inline void cgroup_post_fork(struct task_struct *p) {}
 520 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
 521
 522 static inline void cgroup_lock(void) {}
 523 static inline void cgroup_unlock(void) {}
 524 static inline int cgroupstats_build(struct cgroupstats *stats,
 525                                         struct dentry *dentry)
 526 {
 527         return -EINVAL;
 528 }
 529
 530 #endif /* !CONFIG_CGROUPS */
 531
 532 #endif /* _LINUX_CGROUP_H */