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         /*
  51          * The cgroup that this subsystem is attached to. Useful
  52          * for subsystems that want to know about the cgroup
  53          * hierarchy structure
  54          */
  55         struct cgroup *cgroup;
  56
  57         /*
  58          * State maintained by the cgroup system to allow subsystems
  59          * to be "busy". Should be accessed via css_get(),
  60          * css_tryget() and and css_put().
  61          */
  62
  63         atomic_t refcnt;
  64
  65         unsigned long flags;
  66 };
  67
  68 /* bits in struct cgroup_subsys_state flags field */
  69 enum {
  70         CSS_ROOT, /* This CSS is the root of the subsystem */
  71         CSS_REMOVED, /* This CSS is dead */
  72 };
  73
  74 /*
  75  * Call css_get() to hold a reference on the css; it can be used
  76  * for a reference obtained via:
  77  * - an existing ref-counted reference to the css
  78  * - task->cgroups for a locked task
  79  */
  80
  81 static inline void css_get(struct cgroup_subsys_state *css)
  82 {
  83         /* We don't need to reference count the root state */
  84         if (!test_bit(CSS_ROOT, &css->flags))
  85                 atomic_inc(&css->refcnt);
  86 }
  87
  88 static inline bool css_is_removed(struct cgroup_subsys_state *css)
  89 {
  90         return test_bit(CSS_REMOVED, &css->flags);
  91 }
  92
  93 /*
  94  * Call css_tryget() to take a reference on a css if your existing
  95  * (known-valid) reference isn't already ref-counted. Returns false if
  96  * the css has been destroyed.
  97  */
  98
  99 static inline bool css_tryget(struct cgroup_subsys_state *css)
 100 {
 101         if (test_bit(CSS_ROOT, &css->flags))
 102                 return true;
 103         while (!atomic_inc_not_zero(&css->refcnt)) {
 104                 if (test_bit(CSS_REMOVED, &css->flags))
 105                         return false;
 106                 cpu_relax();
 107         }
 108         return true;
 109 }
 110
 111 /*
 112  * css_put() should be called to release a reference taken by
 113  * css_get() or css_tryget()
 114  */
 115
 116 extern void __css_put(struct cgroup_subsys_state *css);
 117 static inline void css_put(struct cgroup_subsys_state *css)
 118 {
 119         if (!test_bit(CSS_ROOT, &css->flags))
 120                 __css_put(css);
 121 }
 122
 123 /* bits in struct cgroup flags field */
 124 enum {
 125         /* Control Group is dead */
 126         CGRP_REMOVED,
 127         /*
 128          * Control Group has previously had a child cgroup or a task,
 129          * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
 130          */
 131         CGRP_RELEASABLE,
 132         /* Control Group requires release notifications to userspace */
 133         CGRP_NOTIFY_ON_RELEASE,
 134 };
 135
 136 struct cgroup {
 137         unsigned long flags;            /* "unsigned long" so bitops work */
 138
 139         /*
 140          * count users of this cgroup. >0 means busy, but doesn't
 141          * necessarily indicate the number of tasks in the cgroup
 142          */
 143         atomic_t count;
 144
 145         /*
 146          * We link our 'sibling' struct into our parent's 'children'.
 147          * Our children link their 'sibling' into our 'children'.
 148          */
 149         struct list_head sibling;       /* my parent's children */
 150         struct list_head children;      /* my children */
 151
 152         struct cgroup *parent;          /* my parent */
 153         struct dentry *dentry;          /* cgroup fs entry, RCU protected */
 154
 155         /* Private pointers for each registered subsystem */
 156         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 157
 158         struct cgroupfs_root *root;
 159         struct cgroup *top_cgroup;
 160
 161         /*
 162          * List of cg_cgroup_links pointing at css_sets with
 163          * tasks in this cgroup. Protected by css_set_lock
 164          */
 165         struct list_head css_sets;
 166
 167         /*
 168          * Linked list running through all cgroups that can
 169          * potentially be reaped by the release agent. Protected by
 170          * release_list_lock
 171          */
 172         struct list_head release_list;
 173
 174         /* pids_mutex protects the fields below */
 175         struct rw_semaphore pids_mutex;
 176         /* Array of process ids in the cgroup */
 177         pid_t *tasks_pids;
 178         /* How many files are using the current tasks_pids array */
 179         int pids_use_count;
 180         /* Length of the current tasks_pids array */
 181         int pids_length;
 182
 183         /* For RCU-protected deletion */
 184         struct rcu_head rcu_head;
 185 };
 186
 187 /*
 188  * A css_set is a structure holding pointers to a set of
 189  * cgroup_subsys_state objects. This saves space in the task struct
 190  * object and speeds up fork()/exit(), since a single inc/dec and a
 191  * list_add()/del() can bump the reference count on the entire cgroup
 192  * set for a task.
 193  */
 194
 195 struct css_set {
 196
 197         /* Reference count */
 198         atomic_t refcount;
 199
 200         /*
 201          * List running through all cgroup groups in the same hash
 202          * slot. Protected by css_set_lock
 203          */
 204         struct hlist_node hlist;
 205
 206         /*
 207          * List running through all tasks using this cgroup
 208          * group. Protected by css_set_lock
 209          */
 210         struct list_head tasks;
 211
 212         /*
 213          * List of cg_cgroup_link objects on link chains from
 214          * cgroups referenced from this css_set. Protected by
 215          * css_set_lock
 216          */
 217         struct list_head cg_links;
 218
 219         /*
 220          * Set of subsystem states, one for each subsystem. This array
 221          * is immutable after creation apart from the init_css_set
 222          * during subsystem registration (at boot time).
 223          */
 224         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 225 };
 226
 227 /*
 228  * cgroup_map_cb is an abstract callback API for reporting map-valued
 229  * control files
 230  */
 231
 232 struct cgroup_map_cb {
 233         int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
 234         void *state;
 235 };
 236
 237 /*
 238  * struct cftype: handler definitions for cgroup control files
 239  *
 240  * When reading/writing to a file:
 241  *      - the cgroup to use is file->f_dentry->d_parent->d_fsdata
 242  *      - the 'cftype' of the file is file->f_dentry->d_fsdata
 243  */
 244
 245 #define MAX_CFTYPE_NAME 64
 246 struct cftype {
 247         /*
 248          * By convention, the name should begin with the name of the
 249          * subsystem, followed by a period
 250          */
 251         char name[MAX_CFTYPE_NAME];
 252         int private;
 253
 254         /*
 255          * If non-zero, defines the maximum length of string that can
 256          * be passed to write_string; defaults to 64
 257          */
 258         size_t max_write_len;
 259
 260         int (*open)(struct inode *inode, struct file *file);
 261         ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
 262                         struct file *file,
 263                         char __user *buf, size_t nbytes, loff_t *ppos);
 264         /*
 265          * read_u64() is a shortcut for the common case of returning a
 266          * single integer. Use it in place of read()
 267          */
 268         u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
 269         /*
 270          * read_s64() is a signed version of read_u64()
 271          */
 272         s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
 273         /*
 274          * read_map() is used for defining a map of key/value
 275          * pairs. It should call cb->fill(cb, key, value) for each
 276          * entry. The key/value pairs (and their ordering) should not
 277          * change between reboots.
 278          */
 279         int (*read_map)(struct cgroup *cont, struct cftype *cft,
 280                         struct cgroup_map_cb *cb);
 281         /*
 282          * read_seq_string() is used for outputting a simple sequence
 283          * using seqfile.
 284          */
 285         int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
 286                                struct seq_file *m);
 287
 288         ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
 289                          struct file *file,
 290                          const char __user *buf, size_t nbytes, loff_t *ppos);
 291
 292         /*
 293          * write_u64() is a shortcut for the common case of accepting
 294          * a single integer (as parsed by simple_strtoull) from
 295          * userspace. Use in place of write(); return 0 or error.
 296          */
 297         int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
 298         /*
 299          * write_s64() is a signed version of write_u64()
 300          */
 301         int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
 302
 303         /*
 304          * write_string() is passed a nul-terminated kernelspace
 305          * buffer of maximum length determined by max_write_len.
 306          * Returns 0 or -ve error code.
 307          */
 308         int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
 309                             const char *buffer);
 310         /*
 311          * trigger() callback can be used to get some kick from the
 312          * userspace, when the actual string written is not important
 313          * at all. The private field can be used to determine the
 314          * kick type for multiplexing.
 315          */
 316         int (*trigger)(struct cgroup *cgrp, unsigned int event);
 317
 318         int (*release)(struct inode *inode, struct file *file);
 319 };
 320
 321 struct cgroup_scanner {
 322         struct cgroup *cg;
 323         int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
 324         void (*process_task)(struct task_struct *p,
 325                         struct cgroup_scanner *scan);
 326         struct ptr_heap *heap;
 327 };
 328
 329 /*
 330  * Add a new file to the given cgroup directory. Should only be
 331  * called by subsystems from within a populate() method
 332  */
 333 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
 334                        const struct cftype *cft);
 335
 336 /*
 337  * Add a set of new files to the given cgroup directory. Should
 338  * only be called by subsystems from within a populate() method
 339  */
 340 int cgroup_add_files(struct cgroup *cgrp,
 341                         struct cgroup_subsys *subsys,
 342                         const struct cftype cft[],
 343                         int count);
 344
 345 int cgroup_is_removed(const struct cgroup *cgrp);
 346
 347 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
 348
 349 int cgroup_task_count(const struct cgroup *cgrp);
 350
 351 /* Return true if cgrp is a descendant of the task's cgroup */
 352 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
 353
 354 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
 355
 356 struct cgroup_subsys {
 357         struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
 358                                                   struct cgroup *cgrp);
 359         void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 360         void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 361         int (*can_attach)(struct cgroup_subsys *ss,
 362                           struct cgroup *cgrp, struct task_struct *tsk);
 363         void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 364                         struct cgroup *old_cgrp, struct task_struct *tsk);
 365         void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
 366         void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
 367         int (*populate)(struct cgroup_subsys *ss,
 368                         struct cgroup *cgrp);
 369         void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 370         void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
 371
 372         int subsys_id;
 373         int active;
 374         int disabled;
 375         int early_init;
 376 #define MAX_CGROUP_TYPE_NAMELEN 32
 377         const char *name;
 378
 379         /*
 380          * Protects sibling/children links of cgroups in this
 381          * hierarchy, plus protects which hierarchy (or none) the
 382          * subsystem is a part of (i.e. root/sibling).  To avoid
 383          * potential deadlocks, the following operations should not be
 384          * undertaken while holding any hierarchy_mutex:
 385          *
 386          * - allocating memory
 387          * - initiating hotplug events
 388          */
 389         struct mutex hierarchy_mutex;
 390         struct lock_class_key subsys_key;
 391
 392         /*
 393          * Link to parent, and list entry in parent's children.
 394          * Protected by this->hierarchy_mutex and cgroup_lock()
 395          */
 396         struct cgroupfs_root *root;
 397         struct list_head sibling;
 398 };
 399
 400 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
 401 #include <linux/cgroup_subsys.h>
 402 #undef SUBSYS
 403
 404 static inline struct cgroup_subsys_state *cgroup_subsys_state(
 405         struct cgroup *cgrp, int subsys_id)
 406 {
 407         return cgrp->subsys[subsys_id];
 408 }
 409
 410 static inline struct cgroup_subsys_state *task_subsys_state(
 411         struct task_struct *task, int subsys_id)
 412 {
 413         return rcu_dereference(task->cgroups->subsys[subsys_id]);
 414 }
 415
 416 static inline struct cgroup* task_cgroup(struct task_struct *task,
 417                                                int subsys_id)
 418 {
 419         return task_subsys_state(task, subsys_id)->cgroup;
 420 }
 421
 422 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
 423                                                         char *nodename);
 424
 425 /* A cgroup_iter should be treated as an opaque object */
 426 struct cgroup_iter {
 427         struct list_head *cg_link;
 428         struct list_head *task;
 429 };
 430
 431 /*
 432  * To iterate across the tasks in a cgroup:
 433  *
 434  * 1) call cgroup_iter_start to intialize an iterator
 435  *
 436  * 2) call cgroup_iter_next() to retrieve member tasks until it
 437  *    returns NULL or until you want to end the iteration
 438  *
 439  * 3) call cgroup_iter_end() to destroy the iterator.
 440  *
 441  * Or, call cgroup_scan_tasks() to iterate through every task in a
 442  * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
 443  * the test_task() callback, but not while calling the process_task()
 444  * callback.
 445  */
 446 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
 447 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
 448                                         struct cgroup_iter *it);
 449 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
 450 int cgroup_scan_tasks(struct cgroup_scanner *scan);
 451 int cgroup_attach_task(struct cgroup *, struct task_struct *);
 452
 453 #else /* !CONFIG_CGROUPS */
 454
 455 static inline int cgroup_init_early(void) { return 0; }
 456 static inline int cgroup_init(void) { return 0; }
 457 static inline void cgroup_fork(struct task_struct *p) {}
 458 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
 459 static inline void cgroup_post_fork(struct task_struct *p) {}
 460 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
 461
 462 static inline void cgroup_lock(void) {}
 463 static inline void cgroup_unlock(void) {}
 464 static inline int cgroupstats_build(struct cgroupstats *stats,
 465                                         struct dentry *dentry)
 466 {
 467         return -EINVAL;
 468 }
 469
 470 #endif /* !CONFIG_CGROUPS */
 471
 472 #endif /* _LINUX_CGROUP_H */