include/linux/cgroup-defs.h

   1 /*
   2  * linux/cgroup-defs.h - basic definitions for cgroup
   3  *
   4  * This file provides basic type and interface.  Include this file directly
   5  * only if necessary to avoid cyclic dependencies.
   6  */
   7 #ifndef _LINUX_CGROUP_DEFS_H
   8 #define _LINUX_CGROUP_DEFS_H
   9
  10 #include <linux/limits.h>
  11 #include <linux/list.h>
  12 #include <linux/idr.h>
  13 #include <linux/wait.h>
  14 #include <linux/mutex.h>
  15 #include <linux/rcupdate.h>
  16 #include <linux/percpu-refcount.h>
  17 #include <linux/percpu-rwsem.h>
  18 #include <linux/workqueue.h>
  19
  20 #ifdef CONFIG_CGROUPS
  21
  22 struct cgroup;
  23 struct cgroup_root;
  24 struct cgroup_subsys;
  25 struct cgroup_taskset;
  26 struct kernfs_node;
  27 struct kernfs_ops;
  28 struct kernfs_open_file;
  29 struct seq_file;
  30
  31 #define MAX_CGROUP_TYPE_NAMELEN 32
  32 #define MAX_CGROUP_ROOT_NAMELEN 64
  33 #define MAX_CFTYPE_NAME         64
  34
  35 /* define the enumeration of all cgroup subsystems */
  36 #define SUBSYS(_x) _x ## _cgrp_id,
  37 #define SUBSYS_TAG(_t) CGROUP_ ## _t, \
  38         __unused_tag_ ## _t = CGROUP_ ## _t - 1,
  39 enum cgroup_subsys_id {
  40 #include <linux/cgroup_subsys.h>
  41         CGROUP_SUBSYS_COUNT,
  42 };
  43 #undef SUBSYS_TAG
  44 #undef SUBSYS
  45
  46 #define CGROUP_CANFORK_COUNT (CGROUP_CANFORK_END - CGROUP_CANFORK_START)
  47
  48 /* bits in struct cgroup_subsys_state flags field */
  49 enum {
  50         CSS_NO_REF      = (1 << 0), /* no reference counting for this css */
  51         CSS_ONLINE      = (1 << 1), /* between ->css_online() and ->css_offline() */
  52         CSS_RELEASED    = (1 << 2), /* refcnt reached zero, released */
  53 };
  54
  55 /* bits in struct cgroup flags field */
  56 enum {
  57         /* Control Group requires release notifications to userspace */
  58         CGRP_NOTIFY_ON_RELEASE,
  59         /*
  60          * Clone the parent's configuration when creating a new child
  61          * cpuset cgroup.  For historical reasons, this option can be
  62          * specified at mount time and thus is implemented here.
  63          */
  64         CGRP_CPUSET_CLONE_CHILDREN,
  65 };
  66
  67 /* cgroup_root->flags */
  68 enum {
  69         CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */
  70         CGRP_ROOT_NOPREFIX      = (1 << 1), /* mounted subsystems have no named prefix */
  71         CGRP_ROOT_XATTR         = (1 << 2), /* supports extended attributes */
  72 };
  73
  74 /* cftype->flags */
  75 enum {
  76         CFTYPE_ONLY_ON_ROOT     = (1 << 0),     /* only create on root cgrp */
  77         CFTYPE_NOT_ON_ROOT      = (1 << 1),     /* don't create on root cgrp */
  78         CFTYPE_NO_PREFIX        = (1 << 3),     /* (DON'T USE FOR NEW FILES) no subsys prefix */
  79
  80         /* internal flags, do not use outside cgroup core proper */
  81         __CFTYPE_ONLY_ON_DFL    = (1 << 16),    /* only on default hierarchy */
  82         __CFTYPE_NOT_ON_DFL     = (1 << 17),    /* not on default hierarchy */
  83 };
  84
  85 /*
  86  * Per-subsystem/per-cgroup state maintained by the system.  This is the
  87  * fundamental structural building block that controllers deal with.
  88  *
  89  * Fields marked with "PI:" are public and immutable and may be accessed
  90  * directly without synchronization.
  91  */
  92 struct cgroup_subsys_state {
  93         /* PI: the cgroup that this css is attached to */
  94         struct cgroup *cgroup;
  95
  96         /* PI: the cgroup subsystem that this css is attached to */
  97         struct cgroup_subsys *ss;
  98
  99         /* reference count - access via css_[try]get() and css_put() */
 100         struct percpu_ref refcnt;
 101
 102         /* PI: the parent css */
 103         struct cgroup_subsys_state *parent;
 104
 105         /* siblings list anchored at the parent's ->children */
 106         struct list_head sibling;
 107         struct list_head children;
 108
 109         /*
 110          * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
 111          * matching css can be looked up using css_from_id().
 112          */
 113         int id;
 114
 115         unsigned int flags;
 116
 117         /*
 118          * Monotonically increasing unique serial number which defines a
 119          * uniform order among all csses.  It's guaranteed that all
 120          * ->children lists are in the ascending order of ->serial_nr and
 121          * used to allow interrupting and resuming iterations.
 122          */
 123         u64 serial_nr;
 124
 125         /* percpu_ref killing and RCU release */
 126         struct rcu_head rcu_head;
 127         struct work_struct destroy_work;
 128 };
 129
 130 /*
 131  * A css_set is a structure holding pointers to a set of
 132  * cgroup_subsys_state objects. This saves space in the task struct
 133  * object and speeds up fork()/exit(), since a single inc/dec and a
 134  * list_add()/del() can bump the reference count on the entire cgroup
 135  * set for a task.
 136  */
 137 struct css_set {
 138         /* Reference count */
 139         atomic_t refcount;
 140
 141         /*
 142          * List running through all cgroup groups in the same hash
 143          * slot. Protected by css_set_lock
 144          */
 145         struct hlist_node hlist;
 146
 147         /*
 148          * Lists running through all tasks using this cgroup group.
 149          * mg_tasks lists tasks which belong to this cset but are in the
 150          * process of being migrated out or in.  Protected by
 151          * css_set_rwsem, but, during migration, once tasks are moved to
 152          * mg_tasks, it can be read safely while holding cgroup_mutex.
 153          */
 154         struct list_head tasks;
 155         struct list_head mg_tasks;
 156
 157         /*
 158          * List of cgrp_cset_links pointing at cgroups referenced from this
 159          * css_set.  Protected by css_set_lock.
 160          */
 161         struct list_head cgrp_links;
 162
 163         /* the default cgroup associated with this css_set */
 164         struct cgroup *dfl_cgrp;
 165
 166         /*
 167          * Set of subsystem states, one for each subsystem. This array is
 168          * immutable after creation apart from the init_css_set during
 169          * subsystem registration (at boot time).
 170          */
 171         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 172
 173         /*
 174          * List of csets participating in the on-going migration either as
 175          * source or destination.  Protected by cgroup_mutex.
 176          */
 177         struct list_head mg_preload_node;
 178         struct list_head mg_node;
 179
 180         /*
 181          * If this cset is acting as the source of migration the following
 182          * two fields are set.  mg_src_cgrp is the source cgroup of the
 183          * on-going migration and mg_dst_cset is the destination cset the
 184          * target tasks on this cset should be migrated to.  Protected by
 185          * cgroup_mutex.
 186          */
 187         struct cgroup *mg_src_cgrp;
 188         struct css_set *mg_dst_cset;
 189
 190         /*
 191          * On the default hierarhcy, ->subsys[ssid] may point to a css
 192          * attached to an ancestor instead of the cgroup this css_set is
 193          * associated with.  The following node is anchored at
 194          * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
 195          * iterate through all css's attached to a given cgroup.
 196          */
 197         struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
 198
 199         /* For RCU-protected deletion */
 200         struct rcu_head rcu_head;
 201 };
 202
 203 struct cgroup {
 204         /* self css with NULL ->ss, points back to this cgroup */
 205         struct cgroup_subsys_state self;
 206
 207         unsigned long flags;            /* "unsigned long" so bitops work */
 208
 209         /*
 210          * idr allocated in-hierarchy ID.
 211          *
 212          * ID 0 is not used, the ID of the root cgroup is always 1, and a
 213          * new cgroup will be assigned with a smallest available ID.
 214          *
 215          * Allocating/Removing ID must be protected by cgroup_mutex.
 216          */
 217         int id;
 218
 219         /*
 220          * If this cgroup contains any tasks, it contributes one to
 221          * populated_cnt.  All children with non-zero popuplated_cnt of
 222          * their own contribute one.  The count is zero iff there's no task
 223          * in this cgroup or its subtree.
 224          */
 225         int populated_cnt;
 226
 227         struct kernfs_node *kn;         /* cgroup kernfs entry */
 228         struct kernfs_node *procs_kn;   /* kn for "cgroup.procs" */
 229         struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
 230
 231         /*
 232          * The bitmask of subsystems enabled on the child cgroups.
 233          * ->subtree_control is the one configured through
 234          * "cgroup.subtree_control" while ->child_subsys_mask is the
 235          * effective one which may have more subsystems enabled.
 236          * Controller knobs are made available iff it's enabled in
 237          * ->subtree_control.
 238          */
 239         unsigned int subtree_control;
 240         unsigned int child_subsys_mask;
 241
 242         /* Private pointers for each registered subsystem */
 243         struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
 244
 245         struct cgroup_root *root;
 246
 247         /*
 248          * List of cgrp_cset_links pointing at css_sets with tasks in this
 249          * cgroup.  Protected by css_set_lock.
 250          */
 251         struct list_head cset_links;
 252
 253         /*
 254          * On the default hierarchy, a css_set for a cgroup with some
 255          * susbsys disabled will point to css's which are associated with
 256          * the closest ancestor which has the subsys enabled.  The
 257          * following lists all css_sets which point to this cgroup's css
 258          * for the given subsystem.
 259          */
 260         struct list_head e_csets[CGROUP_SUBSYS_COUNT];
 261
 262         /*
 263          * list of pidlists, up to two for each namespace (one for procs, one
 264          * for tasks); created on demand.
 265          */
 266         struct list_head pidlists;
 267         struct mutex pidlist_mutex;
 268
 269         /* used to wait for offlining of csses */
 270         wait_queue_head_t offline_waitq;
 271
 272         /* used to schedule release agent */
 273         struct work_struct release_agent_work;
 274 };
 275
 276 /*
 277  * A cgroup_root represents the root of a cgroup hierarchy, and may be
 278  * associated with a kernfs_root to form an active hierarchy.  This is
 279  * internal to cgroup core.  Don't access directly from controllers.
 280  */
 281 struct cgroup_root {
 282         struct kernfs_root *kf_root;
 283
 284         /* The bitmask of subsystems attached to this hierarchy */
 285         unsigned int subsys_mask;
 286
 287         /* Unique id for this hierarchy. */
 288         int hierarchy_id;
 289
 290         /* The root cgroup.  Root is destroyed on its release. */
 291         struct cgroup cgrp;
 292
 293         /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
 294         atomic_t nr_cgrps;
 295
 296         /* A list running through the active hierarchies */
 297         struct list_head root_list;
 298
 299         /* Hierarchy-specific flags */
 300         unsigned int flags;
 301
 302         /* IDs for cgroups in this hierarchy */
 303         struct idr cgroup_idr;
 304
 305         /* The path to use for release notifications. */
 306         char release_agent_path[PATH_MAX];
 307
 308         /* The name for this hierarchy - may be empty */
 309         char name[MAX_CGROUP_ROOT_NAMELEN];
 310 };
 311
 312 /*
 313  * struct cftype: handler definitions for cgroup control files
 314  *
 315  * When reading/writing to a file:
 316  *      - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
 317  *      - the 'cftype' of the file is file->f_path.dentry->d_fsdata
 318  */
 319 struct cftype {
 320         /*
 321          * By convention, the name should begin with the name of the
 322          * subsystem, followed by a period.  Zero length string indicates
 323          * end of cftype array.
 324          */
 325         char name[MAX_CFTYPE_NAME];
 326         unsigned long private;
 327         /*
 328          * If not 0, file mode is set to this value, otherwise it will
 329          * be figured out automatically
 330          */
 331         umode_t mode;
 332
 333         /*
 334          * The maximum length of string, excluding trailing nul, that can
 335          * be passed to write.  If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
 336          */
 337         size_t max_write_len;
 338
 339         /* CFTYPE_* flags */
 340         unsigned int flags;
 341
 342         /*
 343          * Fields used for internal bookkeeping.  Initialized automatically
 344          * during registration.
 345          */
 346         struct cgroup_subsys *ss;       /* NULL for cgroup core files */
 347         struct list_head node;          /* anchored at ss->cfts */
 348         struct kernfs_ops *kf_ops;
 349
 350         /*
 351          * read_u64() is a shortcut for the common case of returning a
 352          * single integer. Use it in place of read()
 353          */
 354         u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
 355         /*
 356          * read_s64() is a signed version of read_u64()
 357          */
 358         s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
 359
 360         /* generic seq_file read interface */
 361         int (*seq_show)(struct seq_file *sf, void *v);
 362
 363         /* optional ops, implement all or none */
 364         void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
 365         void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
 366         void (*seq_stop)(struct seq_file *sf, void *v);
 367
 368         /*
 369          * write_u64() is a shortcut for the common case of accepting
 370          * a single integer (as parsed by simple_strtoull) from
 371          * userspace. Use in place of write(); return 0 or error.
 372          */
 373         int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
 374                          u64 val);
 375         /*
 376          * write_s64() is a signed version of write_u64()
 377          */
 378         int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
 379                          s64 val);
 380
 381         /*
 382          * write() is the generic write callback which maps directly to
 383          * kernfs write operation and overrides all other operations.
 384          * Maximum write size is determined by ->max_write_len.  Use
 385          * of_css/cft() to access the associated css and cft.
 386          */
 387         ssize_t (*write)(struct kernfs_open_file *of,
 388                          char *buf, size_t nbytes, loff_t off);
 389
 390 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 391         struct lock_class_key   lockdep_key;
 392 #endif
 393 };
 394
 395 /*
 396  * Control Group subsystem type.
 397  * See Documentation/cgroups/cgroups.txt for details
 398  */
 399 struct cgroup_subsys {
 400         struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
 401         int (*css_online)(struct cgroup_subsys_state *css);
 402         void (*css_offline)(struct cgroup_subsys_state *css);
 403         void (*css_released)(struct cgroup_subsys_state *css);
 404         void (*css_free)(struct cgroup_subsys_state *css);
 405         void (*css_reset)(struct cgroup_subsys_state *css);
 406         void (*css_e_css_changed)(struct cgroup_subsys_state *css);
 407
 408         int (*can_attach)(struct cgroup_subsys_state *css,
 409                           struct cgroup_taskset *tset);
 410         void (*cancel_attach)(struct cgroup_subsys_state *css,
 411                               struct cgroup_taskset *tset);
 412         void (*attach)(struct cgroup_subsys_state *css,
 413                        struct cgroup_taskset *tset);
 414         int (*can_fork)(struct task_struct *task, void **priv_p);
 415         void (*cancel_fork)(struct task_struct *task, void *priv);
 416         void (*fork)(struct task_struct *task, void *priv);
 417         void (*exit)(struct cgroup_subsys_state *css,
 418                      struct cgroup_subsys_state *old_css,
 419                      struct task_struct *task);
 420         void (*bind)(struct cgroup_subsys_state *root_css);
 421
 422         int disabled;
 423         int early_init;
 424
 425         /*
 426          * If %false, this subsystem is properly hierarchical -
 427          * configuration, resource accounting and restriction on a parent
 428          * cgroup cover those of its children.  If %true, hierarchy support
 429          * is broken in some ways - some subsystems ignore hierarchy
 430          * completely while others are only implemented half-way.
 431          *
 432          * It's now disallowed to create nested cgroups if the subsystem is
 433          * broken and cgroup core will emit a warning message on such
 434          * cases.  Eventually, all subsystems will be made properly
 435          * hierarchical and this will go away.
 436          */
 437         bool broken_hierarchy;
 438         bool warned_broken_hierarchy;
 439
 440         /* the following two fields are initialized automtically during boot */
 441         int id;
 442         const char *name;
 443
 444         /* optional, initialized automatically during boot if not set */
 445         const char *legacy_name;
 446
 447         /* link to parent, protected by cgroup_lock() */
 448         struct cgroup_root *root;
 449
 450         /* idr for css->id */
 451         struct idr css_idr;
 452
 453         /*
 454          * List of cftypes.  Each entry is the first entry of an array
 455          * terminated by zero length name.
 456          */
 457         struct list_head cfts;
 458
 459         /*
 460          * Base cftypes which are automatically registered.  The two can
 461          * point to the same array.
 462          */
 463         struct cftype *dfl_cftypes;     /* for the default hierarchy */
 464         struct cftype *legacy_cftypes;  /* for the legacy hierarchies */
 465
 466         /*
 467          * A subsystem may depend on other subsystems.  When such subsystem
 468          * is enabled on a cgroup, the depended-upon subsystems are enabled
 469          * together if available.  Subsystems enabled due to dependency are
 470          * not visible to userland until explicitly enabled.  The following
 471          * specifies the mask of subsystems that this one depends on.
 472          */
 473         unsigned int depends_on;
 474 };
 475
 476 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
 477
 478 /**
 479  * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
 480  * @tsk: target task
 481  *
 482  * Called from threadgroup_change_begin() and allows cgroup operations to
 483  * synchronize against threadgroup changes using a percpu_rw_semaphore.
 484  */
 485 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
 486 {
 487         percpu_down_read(&cgroup_threadgroup_rwsem);
 488 }
 489
 490 /**
 491  * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
 492  * @tsk: target task
 493  *
 494  * Called from threadgroup_change_end().  Counterpart of
 495  * cgroup_threadcgroup_change_begin().
 496  */
 497 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
 498 {
 499         percpu_up_read(&cgroup_threadgroup_rwsem);
 500 }
 501
 502 #else   /* CONFIG_CGROUPS */
 503
 504 #define CGROUP_CANFORK_COUNT 0
 505 #define CGROUP_SUBSYS_COUNT 0
 506
 507 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {}
 508 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
 509
 510 #endif  /* CONFIG_CGROUPS */
 511
 512 #endif  /* _LINUX_CGROUP_DEFS_H */