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
  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 subsystems
  56          * to be "busy". Should be accessed via css_get(),
  57          * css_tryget() and 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         CSS_REMOVED, /* This CSS is dead */
  68 };
  69
  70 /*
  71  * Call css_get() to hold a reference on the css; it can be used
  72  * for a reference obtained via:
  73  * - an existing ref-counted reference to the css
  74  * - task->cgroups for a locked task
  75  */
  76
  77 static inline void css_get(struct cgroup_subsys_state *css)
  78 {
  79         /* We don't need to reference count the root state */
  80         if (!test_bit(CSS_ROOT, &css->flags))
  81                 atomic_inc(&css->refcnt);
  82 }
  83
  84 static inline bool css_is_removed(struct cgroup_subsys_state *css)
  85 {
  86         return test_bit(CSS_REMOVED, &css->flags);
  87 }
  88
  89 /*
  90  * Call css_tryget() to take a reference on a css if your existing
  91  * (known-valid) reference isn't already ref-counted. Returns false if
  92  * the css has been destroyed.
  93  */
  94
  95 static inline bool css_tryget(struct cgroup_subsys_state *css)
  96 {
  97         if (test_bit(CSS_ROOT, &css->flags))
  98                 return true;
  99         while (!atomic_inc_not_zero(&css->refcnt)) {
 100                 if (test_bit(CSS_REMOVED, &css->flags))
 101                         return false;
 102         }
 103         return true;
 104 }
 105
 106 /*
 107  * css_put() should be called to release a reference taken by
 108  * css_get() or css_tryget()
 109  */
 110
 111 extern void __css_put(struct cgroup_subsys_state *css);
 112 static inline void css_put(struct cgroup_subsys_state *css)
 113 {
 114         if (!test_bit(CSS_ROOT, &css->flags))
 115                 __css_put(css);
 116 }
 117
 118 /* bits in struct cgroup flags field */
 119 enum {
 120         /* Control Group is dead */
 121         CGRP_REMOVED,
 122         /* Control Group has previously had a child cgroup or a task,
 123          * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
 124         CGRP_RELEASABLE,
 125         /* Control Group requires release notifications to userspace */
 126         CGRP_NOTIFY_ON_RELEASE,
 127 };
 128
 129 struct cgroup {
 130         unsigned long flags;            /* "unsigned long" so bitops work */
 131
 132         /* count users of this cgroup. >0 means busy, but doesn't
 133          * necessarily indicate the number of tasks in the
 134          * cgroup */
 135         atomic_t count;
 136
 137         /*
 138          * We link our 'sibling' struct into our parent's 'children'.
 139          * Our children link their 'sibling' into our 'children'.
 140          */
 141         struct list_head sibling;       /* my parent's children */
 142         struct list_head children;      /* my children */
 143
 144         struct cgroup *parent;  /* my parent */
 145         struct dentry *dentry;          /* cgroup fs entry, RCU protected */
 146
 147         /* Private pointers for each registered subsystem */
 148         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 149
 150         struct cgroupfs_root *root;
 151         struct cgroup *top_cgroup;
 152
 153         /*
 154          * List of cg_cgroup_links pointing at css_sets with
 155          * tasks in this cgroup. Protected by css_set_lock
 156          */
 157         struct list_head css_sets;
 158
 159         /*
 160          * Linked list running through all cgroups that can
 161          * potentially be reaped by the release agent. Protected by
 162          * release_list_lock
 163          */
 164         struct list_head release_list;
 165
 166         /* pids_mutex protects the fields below */
 167         struct rw_semaphore pids_mutex;
 168         /* Array of process ids in the cgroup */
 169         pid_t *tasks_pids;
 170         /* How many files are using the current tasks_pids array */
 171         int pids_use_count;
 172         /* Length of the current tasks_pids array */
 173         int pids_length;
 174
 175         /* For RCU-protected deletion */
 176         struct rcu_head rcu_head;
 177 };
 178
 179 /* A css_set is a structure holding pointers to a set of
 180  * cgroup_subsys_state objects. This saves space in the task struct
 181  * object and speeds up fork()/exit(), since a single inc/dec and a
 182  * list_add()/del() can bump the reference count on the entire
 183  * cgroup set for a task.
 184  */
 185
 186 struct css_set {
 187
 188         /* Reference count */
 189         atomic_t refcount;
 190
 191         /*
 192          * List running through all cgroup groups in the same hash
 193          * slot. Protected by css_set_lock
 194          */
 195         struct hlist_node hlist;
 196
 197         /*
 198          * List running through all tasks using this cgroup
 199          * group. Protected by css_set_lock
 200          */
 201         struct list_head tasks;
 202
 203         /*
 204          * List of cg_cgroup_link objects on link chains from
 205          * cgroups referenced from this css_set. Protected by
 206          * css_set_lock
 207          */
 208         struct list_head cg_links;
 209
 210         /*
 211          * Set of subsystem states, one for each subsystem. This array
 212          * is immutable after creation apart from the init_css_set
 213          * during subsystem registration (at boot time).
 214          */
 215         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 216 };
 217
 218 /*
 219  * cgroup_map_cb is an abstract callback API for reporting map-valued
 220  * control files
 221  */
 222
 223 struct cgroup_map_cb {
 224         int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
 225         void *state;
 226 };
 227
 228 /* struct cftype:
 229  *
 230  * The files in the cgroup filesystem mostly have a very simple read/write
 231  * handling, some common function will take care of it. Nevertheless some cases
 232  * (read tasks) are special and therefore I define this structure for every
 233  * kind of file.
 234  *
 235  *
 236  * When reading/writing to a file:
 237  *      - the cgroup to use is file->f_dentry->d_parent->d_fsdata
 238  *      - the 'cftype' of the file is file->f_dentry->d_fsdata
 239  */
 240
 241 #define MAX_CFTYPE_NAME 64
 242 struct cftype {
 243         /* By convention, the name should begin with the name of the
 244          * subsystem, followed by a period */
 245         char name[MAX_CFTYPE_NAME];
 246         int private;
 247
 248         /*
 249          * If non-zero, defines the maximum length of string that can
 250          * be passed to write_string; defaults to 64
 251          */
 252         size_t max_write_len;
 253
 254         int (*open)(struct inode *inode, struct file *file);
 255         ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
 256                         struct file *file,
 257                         char __user *buf, size_t nbytes, loff_t *ppos);
 258         /*
 259          * read_u64() is a shortcut for the common case of returning a
 260          * single integer. Use it in place of read()
 261          */
 262         u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
 263         /*
 264          * read_s64() is a signed version of read_u64()
 265          */
 266         s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
 267         /*
 268          * read_map() is used for defining a map of key/value
 269          * pairs. It should call cb->fill(cb, key, value) for each
 270          * entry. The key/value pairs (and their ordering) should not
 271          * change between reboots.
 272          */
 273         int (*read_map)(struct cgroup *cont, struct cftype *cft,
 274                         struct cgroup_map_cb *cb);
 275         /*
 276          * read_seq_string() is used for outputting a simple sequence
 277          * using seqfile.
 278          */
 279         int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
 280                                struct seq_file *m);
 281
 282         ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
 283                          struct file *file,
 284                          const char __user *buf, size_t nbytes, loff_t *ppos);
 285
 286         /*
 287          * write_u64() is a shortcut for the common case of accepting
 288          * a single integer (as parsed by simple_strtoull) from
 289          * userspace. Use in place of write(); return 0 or error.
 290          */
 291         int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
 292         /*
 293          * write_s64() is a signed version of write_u64()
 294          */
 295         int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
 296
 297         /*
 298          * write_string() is passed a nul-terminated kernelspace
 299          * buffer of maximum length determined by max_write_len.
 300          * Returns 0 or -ve error code.
 301          */
 302         int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
 303                             const char *buffer);
 304         /*
 305          * trigger() callback can be used to get some kick from the
 306          * userspace, when the actual string written is not important
 307          * at all. The private field can be used to determine the
 308          * kick type for multiplexing.
 309          */
 310         int (*trigger)(struct cgroup *cgrp, unsigned int event);
 311
 312         int (*release)(struct inode *inode, struct file *file);
 313 };
 314
 315 struct cgroup_scanner {
 316         struct cgroup *cg;
 317         int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
 318         void (*process_task)(struct task_struct *p,
 319                         struct cgroup_scanner *scan);
 320         struct ptr_heap *heap;
 321 };
 322
 323 /* Add a new file to the given cgroup directory. Should only be
 324  * called by subsystems from within a populate() method */
 325 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
 326                        const struct cftype *cft);
 327
 328 /* Add a set of new files to the given cgroup directory. Should
 329  * only be called by subsystems from within a populate() method */
 330 int cgroup_add_files(struct cgroup *cgrp,
 331                         struct cgroup_subsys *subsys,
 332                         const struct cftype cft[],
 333                         int count);
 334
 335 int cgroup_is_removed(const struct cgroup *cgrp);
 336
 337 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
 338
 339 int cgroup_task_count(const struct cgroup *cgrp);
 340
 341 /* Return true if the cgroup is a descendant of the current cgroup */
 342 int cgroup_is_descendant(const struct cgroup *cgrp);
 343
 344 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
 345
 346 struct cgroup_subsys {
 347         struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
 348                                                   struct cgroup *cgrp);
 349         void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 350         void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 351         int (*can_attach)(struct cgroup_subsys *ss,
 352                           struct cgroup *cgrp, struct task_struct *tsk);
 353         void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 354                         struct cgroup *old_cgrp, struct task_struct *tsk);
 355         void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
 356         void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
 357         int (*populate)(struct cgroup_subsys *ss,
 358                         struct cgroup *cgrp);
 359         void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 360         void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
 361
 362         int subsys_id;
 363         int active;
 364         int disabled;
 365         int early_init;
 366 #define MAX_CGROUP_TYPE_NAMELEN 32
 367         const char *name;
 368
 369         /*
 370          * Protects sibling/children links of cgroups in this
 371          * hierarchy, plus protects which hierarchy (or none) the
 372          * subsystem is a part of (i.e. root/sibling).  To avoid
 373          * potential deadlocks, the following operations should not be
 374          * undertaken while holding any hierarchy_mutex:
 375          *
 376          * - allocating memory
 377          * - initiating hotplug events
 378          */
 379         struct mutex hierarchy_mutex;
 380
 381         /*
 382          * Link to parent, and list entry in parent's children.
 383          * Protected by this->hierarchy_mutex and cgroup_lock()
 384          */
 385         struct cgroupfs_root *root;
 386         struct list_head sibling;
 387 };
 388
 389 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
 390 #include <linux/cgroup_subsys.h>
 391 #undef SUBSYS
 392
 393 static inline struct cgroup_subsys_state *cgroup_subsys_state(
 394         struct cgroup *cgrp, int subsys_id)
 395 {
 396         return cgrp->subsys[subsys_id];
 397 }
 398
 399 static inline struct cgroup_subsys_state *task_subsys_state(
 400         struct task_struct *task, int subsys_id)
 401 {
 402         return rcu_dereference(task->cgroups->subsys[subsys_id]);
 403 }
 404
 405 static inline struct cgroup* task_cgroup(struct task_struct *task,
 406                                                int subsys_id)
 407 {
 408         return task_subsys_state(task, subsys_id)->cgroup;
 409 }
 410
 411 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
 412                                                         char *nodename);
 413
 414 /* A cgroup_iter should be treated as an opaque object */
 415 struct cgroup_iter {
 416         struct list_head *cg_link;
 417         struct list_head *task;
 418 };
 419
 420 /* To iterate across the tasks in a cgroup:
 421  *
 422  * 1) call cgroup_iter_start to intialize an iterator
 423  *
 424  * 2) call cgroup_iter_next() to retrieve member tasks until it
 425  *    returns NULL or until you want to end the iteration
 426  *
 427  * 3) call cgroup_iter_end() to destroy the iterator.
 428  *
 429  * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
 430  *    - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
 431  *      callback, but not while calling the process_task() callback.
 432  */
 433 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
 434 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
 435                                         struct cgroup_iter *it);
 436 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
 437 int cgroup_scan_tasks(struct cgroup_scanner *scan);
 438 int cgroup_attach_task(struct cgroup *, struct task_struct *);
 439
 440 #else /* !CONFIG_CGROUPS */
 441
 442 static inline int cgroup_init_early(void) { return 0; }
 443 static inline int cgroup_init(void) { return 0; }
 444 static inline void cgroup_fork(struct task_struct *p) {}
 445 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
 446 static inline void cgroup_post_fork(struct task_struct *p) {}
 447 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
 448
 449 static inline void cgroup_lock(void) {}
 450 static inline void cgroup_unlock(void) {}
 451 static inline int cgroupstats_build(struct cgroupstats *stats,
 452                                         struct dentry *dentry)
 453 {
 454         return -EINVAL;
 455 }
 456
 457 #endif /* !CONFIG_CGROUPS */
 458
 459 #endif /* _LINUX_CGROUP_H */