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