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