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/kref.h>
  13 #include <linux/cpumask.h>
  14 #include <linux/nodemask.h>
  15 #include <linux/rcupdate.h>
  16 #include <linux/cgroupstats.h>
  17 #include <linux/prio_heap.h>
  18
  19 #ifdef CONFIG_CGROUPS
  20
  21 struct cgroupfs_root;
  22 struct cgroup_subsys;
  23 struct inode;
  24
  25 extern int cgroup_init_early(void);
  26 extern int cgroup_init(void);
  27 extern void cgroup_init_smp(void);
  28 extern void cgroup_lock(void);
  29 extern void cgroup_unlock(void);
  30 extern void cgroup_fork(struct task_struct *p);
  31 extern void cgroup_fork_callbacks(struct task_struct *p);
  32 extern void cgroup_post_fork(struct task_struct *p);
  33 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
  34 extern int cgroupstats_build(struct cgroupstats *stats,
  35                                 struct dentry *dentry);
  36
  37 extern struct file_operations proc_cgroup_operations;
  38
  39 /* Define the enumeration of all cgroup subsystems */
  40 #define SUBSYS(_x) _x ## _subsys_id,
  41 enum cgroup_subsys_id {
  42 #include <linux/cgroup_subsys.h>
  43         CGROUP_SUBSYS_COUNT
  44 };
  45 #undef SUBSYS
  46
  47 /* Per-subsystem/per-cgroup state maintained by the system. */
  48 struct cgroup_subsys_state {
  49         /* The cgroup that this subsystem is attached to. Useful
  50          * for subsystems that want to know about the cgroup
  51          * hierarchy structure */
  52         struct cgroup *cgroup;
  53
  54         /* State maintained by the cgroup system to allow
  55          * subsystems to be "busy". Should be accessed via css_get()
  56          * and css_put() */
  57
  58         atomic_t refcnt;
  59
  60         unsigned long flags;
  61 };
  62
  63 /* bits in struct cgroup_subsys_state flags field */
  64 enum {
  65         CSS_ROOT, /* This CSS is the root of the subsystem */
  66 };
  67
  68 /*
  69  * Call css_get() to hold a reference on the cgroup;
  70  *
  71  */
  72
  73 static inline void css_get(struct cgroup_subsys_state *css)
  74 {
  75         /* We don't need to reference count the root state */
  76         if (!test_bit(CSS_ROOT, &css->flags))
  77                 atomic_inc(&css->refcnt);
  78 }
  79 /*
  80  * css_put() should be called to release a reference taken by
  81  * css_get()
  82  */
  83
  84 extern void __css_put(struct cgroup_subsys_state *css);
  85 static inline void css_put(struct cgroup_subsys_state *css)
  86 {
  87         if (!test_bit(CSS_ROOT, &css->flags))
  88                 __css_put(css);
  89 }
  90
  91 /* bits in struct cgroup flags field */
  92 enum {
  93         /* Control Group is dead */
  94         CGRP_REMOVED,
  95         /* Control Group has previously had a child cgroup or a task,
  96          * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
  97         CGRP_RELEASABLE,
  98         /* Control Group requires release notifications to userspace */
  99         CGRP_NOTIFY_ON_RELEASE,
 100 };
 101
 102 struct cgroup {
 103         unsigned long flags;            /* "unsigned long" so bitops work */
 104
 105         /* count users of this cgroup. >0 means busy, but doesn't
 106          * necessarily indicate the number of tasks in the
 107          * cgroup */
 108         atomic_t count;
 109
 110         /*
 111          * We link our 'sibling' struct into our parent's 'children'.
 112          * Our children link their 'sibling' into our 'children'.
 113          */
 114         struct list_head sibling;       /* my parent's children */
 115         struct list_head children;      /* my children */
 116
 117         struct cgroup *parent;  /* my parent */
 118         struct dentry *dentry;          /* cgroup fs entry */
 119
 120         /* Private pointers for each registered subsystem */
 121         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 122
 123         struct cgroupfs_root *root;
 124         struct cgroup *top_cgroup;
 125
 126         /*
 127          * List of cg_cgroup_links pointing at css_sets with
 128          * tasks in this cgroup. Protected by css_set_lock
 129          */
 130         struct list_head css_sets;
 131
 132         /*
 133          * Linked list running through all cgroups that can
 134          * potentially be reaped by the release agent. Protected by
 135          * release_list_lock
 136          */
 137         struct list_head release_list;
 138 };
 139
 140 /* A css_set is a structure holding pointers to a set of
 141  * cgroup_subsys_state objects. This saves space in the task struct
 142  * object and speeds up fork()/exit(), since a single inc/dec and a
 143  * list_add()/del() can bump the reference count on the entire
 144  * cgroup set for a task.
 145  */
 146
 147 struct css_set {
 148
 149         /* Reference count */
 150         struct kref ref;
 151
 152         /*
 153          * List running through all cgroup groups in the same hash
 154          * slot. Protected by css_set_lock
 155          */
 156         struct hlist_node hlist;
 157
 158         /*
 159          * List running through all tasks using this cgroup
 160          * group. Protected by css_set_lock
 161          */
 162         struct list_head tasks;
 163
 164         /*
 165          * List of cg_cgroup_link objects on link chains from
 166          * cgroups referenced from this css_set. Protected by
 167          * css_set_lock
 168          */
 169         struct list_head cg_links;
 170
 171         /*
 172          * Set of subsystem states, one for each subsystem. This array
 173          * is immutable after creation apart from the init_css_set
 174          * during subsystem registration (at boot time).
 175          */
 176         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 177 };
 178
 179 /*
 180  * cgroup_map_cb is an abstract callback API for reporting map-valued
 181  * control files
 182  */
 183
 184 struct cgroup_map_cb {
 185         int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
 186         void *state;
 187 };
 188
 189 /* struct cftype:
 190  *
 191  * The files in the cgroup filesystem mostly have a very simple read/write
 192  * handling, some common function will take care of it. Nevertheless some cases
 193  * (read tasks) are special and therefore I define this structure for every
 194  * kind of file.
 195  *
 196  *
 197  * When reading/writing to a file:
 198  *      - the cgroup to use is file->f_dentry->d_parent->d_fsdata
 199  *      - the 'cftype' of the file is file->f_dentry->d_fsdata
 200  */
 201
 202 #define MAX_CFTYPE_NAME 64
 203 struct cftype {
 204         /* By convention, the name should begin with the name of the
 205          * subsystem, followed by a period */
 206         char name[MAX_CFTYPE_NAME];
 207         int private;
 208
 209         /*
 210          * If non-zero, defines the maximum length of string that can
 211          * be passed to write_string; defaults to 64
 212          */
 213         size_t max_write_len;
 214
 215         int (*open)(struct inode *inode, struct file *file);
 216         ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
 217                         struct file *file,
 218                         char __user *buf, size_t nbytes, loff_t *ppos);
 219         /*
 220          * read_u64() is a shortcut for the common case of returning a
 221          * single integer. Use it in place of read()
 222          */
 223         u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
 224         /*
 225          * read_s64() is a signed version of read_u64()
 226          */
 227         s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
 228         /*
 229          * read_map() is used for defining a map of key/value
 230          * pairs. It should call cb->fill(cb, key, value) for each
 231          * entry. The key/value pairs (and their ordering) should not
 232          * change between reboots.
 233          */
 234         int (*read_map)(struct cgroup *cont, struct cftype *cft,
 235                         struct cgroup_map_cb *cb);
 236         /*
 237          * read_seq_string() is used for outputting a simple sequence
 238          * using seqfile.
 239          */
 240         int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
 241                                struct seq_file *m);
 242
 243         ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
 244                          struct file *file,
 245                          const char __user *buf, size_t nbytes, loff_t *ppos);
 246
 247         /*
 248          * write_u64() is a shortcut for the common case of accepting
 249          * a single integer (as parsed by simple_strtoull) from
 250          * userspace. Use in place of write(); return 0 or error.
 251          */
 252         int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
 253         /*
 254          * write_s64() is a signed version of write_u64()
 255          */
 256         int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
 257
 258         /*
 259          * write_string() is passed a nul-terminated kernelspace
 260          * buffer of maximum length determined by max_write_len.
 261          * Returns 0 or -ve error code.
 262          */
 263         int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
 264                             const char *buffer);
 265         /*
 266          * trigger() callback can be used to get some kick from the
 267          * userspace, when the actual string written is not important
 268          * at all. The private field can be used to determine the
 269          * kick type for multiplexing.
 270          */
 271         int (*trigger)(struct cgroup *cgrp, unsigned int event);
 272
 273         int (*release)(struct inode *inode, struct file *file);
 274 };
 275
 276 struct cgroup_scanner {
 277         struct cgroup *cg;
 278         int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
 279         void (*process_task)(struct task_struct *p,
 280                         struct cgroup_scanner *scan);
 281         struct ptr_heap *heap;
 282 };
 283
 284 /* Add a new file to the given cgroup directory. Should only be
 285  * called by subsystems from within a populate() method */
 286 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
 287                        const struct cftype *cft);
 288
 289 /* Add a set of new files to the given cgroup directory. Should
 290  * only be called by subsystems from within a populate() method */
 291 int cgroup_add_files(struct cgroup *cgrp,
 292                         struct cgroup_subsys *subsys,
 293                         const struct cftype cft[],
 294                         int count);
 295
 296 int cgroup_is_removed(const struct cgroup *cgrp);
 297
 298 int cgroup_lock_live_group(struct cgroup *cgrp);
 299
 300 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
 301
 302 int cgroup_task_count(const struct cgroup *cgrp);
 303
 304 /* Return true if the cgroup is a descendant of the current cgroup */
 305 int cgroup_is_descendant(const struct cgroup *cgrp);
 306
 307 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
 308
 309 struct cgroup_subsys {
 310         struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
 311                                                   struct cgroup *cgrp);
 312         void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 313         void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 314         int (*can_attach)(struct cgroup_subsys *ss,
 315                           struct cgroup *cgrp, struct task_struct *tsk);
 316         void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
 317                         struct cgroup *old_cgrp, struct task_struct *tsk);
 318         void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
 319         void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
 320         int (*populate)(struct cgroup_subsys *ss,
 321                         struct cgroup *cgrp);
 322         void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
 323         void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
 324         /*
 325          * This routine is called with the task_lock of mm->owner held
 326          */
 327         void (*mm_owner_changed)(struct cgroup_subsys *ss,
 328                                         struct cgroup *old,
 329                                         struct cgroup *new);
 330         int subsys_id;
 331         int active;
 332         int disabled;
 333         int early_init;
 334 #define MAX_CGROUP_TYPE_NAMELEN 32
 335         const char *name;
 336
 337         /* Protected by RCU */
 338         struct cgroupfs_root *root;
 339
 340         struct list_head sibling;
 341
 342         void *private;
 343 };
 344
 345 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
 346 #include <linux/cgroup_subsys.h>
 347 #undef SUBSYS
 348
 349 static inline struct cgroup_subsys_state *cgroup_subsys_state(
 350         struct cgroup *cgrp, int subsys_id)
 351 {
 352         return cgrp->subsys[subsys_id];
 353 }
 354
 355 static inline struct cgroup_subsys_state *task_subsys_state(
 356         struct task_struct *task, int subsys_id)
 357 {
 358         return rcu_dereference(task->cgroups->subsys[subsys_id]);
 359 }
 360
 361 static inline struct cgroup* task_cgroup(struct task_struct *task,
 362                                                int subsys_id)
 363 {
 364         return task_subsys_state(task, subsys_id)->cgroup;
 365 }
 366
 367 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss);
 368
 369 /* A cgroup_iter should be treated as an opaque object */
 370 struct cgroup_iter {
 371         struct list_head *cg_link;
 372         struct list_head *task;
 373 };
 374
 375 /* To iterate across the tasks in a cgroup:
 376  *
 377  * 1) call cgroup_iter_start to intialize an iterator
 378  *
 379  * 2) call cgroup_iter_next() to retrieve member tasks until it
 380  *    returns NULL or until you want to end the iteration
 381  *
 382  * 3) call cgroup_iter_end() to destroy the iterator.
 383  *
 384  * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
 385  *    - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
 386  *      callback, but not while calling the process_task() callback.
 387  */
 388 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
 389 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
 390                                         struct cgroup_iter *it);
 391 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
 392 int cgroup_scan_tasks(struct cgroup_scanner *scan);
 393 int cgroup_attach_task(struct cgroup *, struct task_struct *);
 394
 395 #else /* !CONFIG_CGROUPS */
 396
 397 static inline int cgroup_init_early(void) { return 0; }
 398 static inline int cgroup_init(void) { return 0; }
 399 static inline void cgroup_init_smp(void) {}
 400 static inline void cgroup_fork(struct task_struct *p) {}
 401 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
 402 static inline void cgroup_post_fork(struct task_struct *p) {}
 403 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
 404
 405 static inline void cgroup_lock(void) {}
 406 static inline void cgroup_unlock(void) {}
 407 static inline int cgroupstats_build(struct cgroupstats *stats,
 408                                         struct dentry *dentry)
 409 {
 410         return -EINVAL;
 411 }
 412
 413 #endif /* !CONFIG_CGROUPS */
 414
 415 #ifdef CONFIG_MM_OWNER
 416 extern void
 417 cgroup_mm_owner_callbacks(struct task_struct *old, struct task_struct *new);
 418 #else /* !CONFIG_MM_OWNER */
 419 static inline void
 420 cgroup_mm_owner_callbacks(struct task_struct *old, struct task_struct *new)
 421 {
 422 }
 423 #endif /* CONFIG_MM_OWNER */
 424 #endif /* _LINUX_CGROUP_H */