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 /* 119 * Incremented by online self and children. Used to guarantee that 120 * parents are not offlined before their children. 121 */ 122 atomic_t online_cnt; 123 124 /* percpu_ref killing and RCU release */ 125 struct rcu_head rcu_head; 126 struct work_struct destroy_work; 127 }; 128 129 /* 130 * A css_set is a structure holding pointers to a set of 131 * cgroup_subsys_state objects. This saves space in the task struct 132 * object and speeds up fork()/exit(), since a single inc/dec and a 133 * list_add()/del() can bump the reference count on the entire cgroup 134 * set for a task. 135 */ 136 struct css_set { 137 /* Reference count */ 138 atomic_t refcount; 139 140 /* 141 * List running through all cgroup groups in the same hash 142 * slot. Protected by css_set_lock 143 */ 144 struct hlist_node hlist; 145 146 /* 147 * Lists running through all tasks using this cgroup group. 148 * mg_tasks lists tasks which belong to this cset but are in the 149 * process of being migrated out or in. Protected by 150 * css_set_rwsem, but, during migration, once tasks are moved to 151 * mg_tasks, it can be read safely while holding cgroup_mutex. 152 */ 153 struct list_head tasks; 154 struct list_head mg_tasks; 155 156 /* 157 * List of cgrp_cset_links pointing at cgroups referenced from this 158 * css_set. Protected by css_set_lock. 159 */ 160 struct list_head cgrp_links; 161 162 /* the default cgroup associated with this css_set */ 163 struct cgroup *dfl_cgrp; 164 165 /* 166 * Set of subsystem states, one for each subsystem. This array is 167 * immutable after creation apart from the init_css_set during 168 * subsystem registration (at boot time). 169 */ 170 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; 171 172 /* 173 * List of csets participating in the on-going migration either as 174 * source or destination. Protected by cgroup_mutex. 175 */ 176 struct list_head mg_preload_node; 177 struct list_head mg_node; 178 179 /* 180 * If this cset is acting as the source of migration the following 181 * two fields are set. mg_src_cgrp is the source cgroup of the 182 * on-going migration and mg_dst_cset is the destination cset the 183 * target tasks on this cset should be migrated to. Protected by 184 * cgroup_mutex. 185 */ 186 struct cgroup *mg_src_cgrp; 187 struct css_set *mg_dst_cset; 188 189 /* 190 * On the default hierarhcy, ->subsys[ssid] may point to a css 191 * attached to an ancestor instead of the cgroup this css_set is 192 * associated with. The following node is anchored at 193 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to 194 * iterate through all css's attached to a given cgroup. 195 */ 196 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; 197 198 /* For RCU-protected deletion */ 199 struct rcu_head rcu_head; 200 }; 201 202 struct cgroup { 203 /* self css with NULL ->ss, points back to this cgroup */ 204 struct cgroup_subsys_state self; 205 206 unsigned long flags; /* "unsigned long" so bitops work */ 207 208 /* 209 * idr allocated in-hierarchy ID. 210 * 211 * ID 0 is not used, the ID of the root cgroup is always 1, and a 212 * new cgroup will be assigned with a smallest available ID. 213 * 214 * Allocating/Removing ID must be protected by cgroup_mutex. 215 */ 216 int id; 217 218 /* 219 * If this cgroup contains any tasks, it contributes one to 220 * populated_cnt. All children with non-zero popuplated_cnt of 221 * their own contribute one. The count is zero iff there's no task 222 * in this cgroup or its subtree. 223 */ 224 int populated_cnt; 225 226 struct kernfs_node *kn; /* cgroup kernfs entry */ 227 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */ 228 229 /* 230 * The bitmask of subsystems enabled on the child cgroups. 231 * ->subtree_control is the one configured through 232 * "cgroup.subtree_control" while ->child_subsys_mask is the 233 * effective one which may have more subsystems enabled. 234 * Controller knobs are made available iff it's enabled in 235 * ->subtree_control. 236 */ 237 unsigned int subtree_control; 238 unsigned int child_subsys_mask; 239 240 /* Private pointers for each registered subsystem */ 241 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; 242 243 struct cgroup_root *root; 244 245 /* 246 * List of cgrp_cset_links pointing at css_sets with tasks in this 247 * cgroup. Protected by css_set_lock. 248 */ 249 struct list_head cset_links; 250 251 /* 252 * On the default hierarchy, a css_set for a cgroup with some 253 * susbsys disabled will point to css's which are associated with 254 * the closest ancestor which has the subsys enabled. The 255 * following lists all css_sets which point to this cgroup's css 256 * for the given subsystem. 257 */ 258 struct list_head e_csets[CGROUP_SUBSYS_COUNT]; 259 260 /* 261 * list of pidlists, up to two for each namespace (one for procs, one 262 * for tasks); created on demand. 263 */ 264 struct list_head pidlists; 265 struct mutex pidlist_mutex; 266 267 /* used to wait for offlining of csses */ 268 wait_queue_head_t offline_waitq; 269 270 /* used to schedule release agent */ 271 struct work_struct release_agent_work; 272 }; 273 274 /* 275 * A cgroup_root represents the root of a cgroup hierarchy, and may be 276 * associated with a kernfs_root to form an active hierarchy. This is 277 * internal to cgroup core. Don't access directly from controllers. 278 */ 279 struct cgroup_root { 280 struct kernfs_root *kf_root; 281 282 /* The bitmask of subsystems attached to this hierarchy */ 283 unsigned int subsys_mask; 284 285 /* Unique id for this hierarchy. */ 286 int hierarchy_id; 287 288 /* The root cgroup. Root is destroyed on its release. */ 289 struct cgroup cgrp; 290 291 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ 292 atomic_t nr_cgrps; 293 294 /* A list running through the active hierarchies */ 295 struct list_head root_list; 296 297 /* Hierarchy-specific flags */ 298 unsigned int flags; 299 300 /* IDs for cgroups in this hierarchy */ 301 struct idr cgroup_idr; 302 303 /* The path to use for release notifications. */ 304 char release_agent_path[PATH_MAX]; 305 306 /* The name for this hierarchy - may be empty */ 307 char name[MAX_CGROUP_ROOT_NAMELEN]; 308 }; 309 310 /* 311 * struct cftype: handler definitions for cgroup control files 312 * 313 * When reading/writing to a file: 314 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata 315 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata 316 */ 317 struct cftype { 318 /* 319 * By convention, the name should begin with the name of the 320 * subsystem, followed by a period. Zero length string indicates 321 * end of cftype array. 322 */ 323 char name[MAX_CFTYPE_NAME]; 324 int private; 325 /* 326 * If not 0, file mode is set to this value, otherwise it will 327 * be figured out automatically 328 */ 329 umode_t mode; 330 331 /* 332 * The maximum length of string, excluding trailing nul, that can 333 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. 334 */ 335 size_t max_write_len; 336 337 /* CFTYPE_* flags */ 338 unsigned int flags; 339 340 /* 341 * Fields used for internal bookkeeping. Initialized automatically 342 * during registration. 343 */ 344 struct cgroup_subsys *ss; /* NULL for cgroup core files */ 345 struct list_head node; /* anchored at ss->cfts */ 346 struct kernfs_ops *kf_ops; 347 348 /* 349 * read_u64() is a shortcut for the common case of returning a 350 * single integer. Use it in place of read() 351 */ 352 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); 353 /* 354 * read_s64() is a signed version of read_u64() 355 */ 356 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); 357 358 /* generic seq_file read interface */ 359 int (*seq_show)(struct seq_file *sf, void *v); 360 361 /* optional ops, implement all or none */ 362 void *(*seq_start)(struct seq_file *sf, loff_t *ppos); 363 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); 364 void (*seq_stop)(struct seq_file *sf, void *v); 365 366 /* 367 * write_u64() is a shortcut for the common case of accepting 368 * a single integer (as parsed by simple_strtoull) from 369 * userspace. Use in place of write(); return 0 or error. 370 */ 371 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, 372 u64 val); 373 /* 374 * write_s64() is a signed version of write_u64() 375 */ 376 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, 377 s64 val); 378 379 /* 380 * write() is the generic write callback which maps directly to 381 * kernfs write operation and overrides all other operations. 382 * Maximum write size is determined by ->max_write_len. Use 383 * of_css/cft() to access the associated css and cft. 384 */ 385 ssize_t (*write)(struct kernfs_open_file *of, 386 char *buf, size_t nbytes, loff_t off); 387 388 #ifdef CONFIG_DEBUG_LOCK_ALLOC 389 struct lock_class_key lockdep_key; 390 #endif 391 }; 392 393 /* 394 * Control Group subsystem type. 395 * See Documentation/cgroups/cgroups.txt for details 396 */ 397 struct cgroup_subsys { 398 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); 399 int (*css_online)(struct cgroup_subsys_state *css); 400 void (*css_offline)(struct cgroup_subsys_state *css); 401 void (*css_released)(struct cgroup_subsys_state *css); 402 void (*css_free)(struct cgroup_subsys_state *css); 403 void (*css_reset)(struct cgroup_subsys_state *css); 404 void (*css_e_css_changed)(struct cgroup_subsys_state *css); 405 406 int (*can_attach)(struct cgroup_subsys_state *css, 407 struct cgroup_taskset *tset); 408 void (*cancel_attach)(struct cgroup_subsys_state *css, 409 struct cgroup_taskset *tset); 410 void (*attach)(struct cgroup_subsys_state *css, 411 struct cgroup_taskset *tset); 412 void (*fork)(struct task_struct *task); 413 void (*exit)(struct cgroup_subsys_state *css, 414 struct cgroup_subsys_state *old_css, 415 struct task_struct *task); 416 void (*bind)(struct cgroup_subsys_state *root_css); 417 418 int disabled; 419 int early_init; 420 421 /* 422 * If %false, this subsystem is properly hierarchical - 423 * configuration, resource accounting and restriction on a parent 424 * cgroup cover those of its children. If %true, hierarchy support 425 * is broken in some ways - some subsystems ignore hierarchy 426 * completely while others are only implemented half-way. 427 * 428 * It's now disallowed to create nested cgroups if the subsystem is 429 * broken and cgroup core will emit a warning message on such 430 * cases. Eventually, all subsystems will be made properly 431 * hierarchical and this will go away. 432 */ 433 bool broken_hierarchy; 434 bool warned_broken_hierarchy; 435 436 /* the following two fields are initialized automtically during boot */ 437 int id; 438 const char *name; 439 440 /* link to parent, protected by cgroup_lock() */ 441 struct cgroup_root *root; 442 443 /* idr for css->id */ 444 struct idr css_idr; 445 446 /* 447 * List of cftypes. Each entry is the first entry of an array 448 * terminated by zero length name. 449 */ 450 struct list_head cfts; 451 452 /* 453 * Base cftypes which are automatically registered. The two can 454 * point to the same array. 455 */ 456 struct cftype *dfl_cftypes; /* for the default hierarchy */ 457 struct cftype *legacy_cftypes; /* for the legacy hierarchies */ 458 459 /* 460 * A subsystem may depend on other subsystems. When such subsystem 461 * is enabled on a cgroup, the depended-upon subsystems are enabled 462 * together if available. Subsystems enabled due to dependency are 463 * not visible to userland until explicitly enabled. The following 464 * specifies the mask of subsystems that this one depends on. 465 */ 466 unsigned int depends_on; 467 }; 468 469 #endif /* CONFIG_CGROUPS */ 470 #endif /* _LINUX_CGROUP_DEFS_H */ 471