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