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