| 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/cpumask.h> |
| 13 | #include <linux/nodemask.h> |
| 14 | #include <linux/rcupdate.h> |
| 15 | #include <linux/rculist.h> |
| 16 | #include <linux/cgroupstats.h> |
| 17 | #include <linux/prio_heap.h> |
| 18 | #include <linux/rwsem.h> |
| 19 | #include <linux/idr.h> |
| 20 | #include <linux/workqueue.h> |
| 21 | #include <linux/xattr.h> |
| 22 | #include <linux/fs.h> |
| 23 | |
| 24 | #ifdef CONFIG_CGROUPS |
| 25 | |
| 26 | struct cgroupfs_root; |
| 27 | struct cgroup_subsys; |
| 28 | struct inode; |
| 29 | struct cgroup; |
| 30 | struct css_id; |
| 31 | |
| 32 | extern int cgroup_init_early(void); |
| 33 | extern int cgroup_init(void); |
| 34 | extern void cgroup_fork(struct task_struct *p); |
| 35 | extern void cgroup_post_fork(struct task_struct *p); |
| 36 | extern void cgroup_exit(struct task_struct *p, int run_callbacks); |
| 37 | extern int cgroupstats_build(struct cgroupstats *stats, |
| 38 | struct dentry *dentry); |
| 39 | extern int cgroup_load_subsys(struct cgroup_subsys *ss); |
| 40 | extern void cgroup_unload_subsys(struct cgroup_subsys *ss); |
| 41 | |
| 42 | extern int proc_cgroup_show(struct seq_file *, void *); |
| 43 | |
| 44 | /* |
| 45 | * Define the enumeration of all cgroup subsystems. |
| 46 | * |
| 47 | * We define ids for builtin subsystems and then modular ones. |
| 48 | */ |
| 49 | #define SUBSYS(_x) _x ## _subsys_id, |
| 50 | enum cgroup_subsys_id { |
| 51 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
| 52 | #include <linux/cgroup_subsys.h> |
| 53 | #undef IS_SUBSYS_ENABLED |
| 54 | CGROUP_BUILTIN_SUBSYS_COUNT, |
| 55 | |
| 56 | __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1, |
| 57 | |
| 58 | #define IS_SUBSYS_ENABLED(option) IS_MODULE(option) |
| 59 | #include <linux/cgroup_subsys.h> |
| 60 | #undef IS_SUBSYS_ENABLED |
| 61 | CGROUP_SUBSYS_COUNT, |
| 62 | }; |
| 63 | #undef SUBSYS |
| 64 | |
| 65 | /* Per-subsystem/per-cgroup state maintained by the system. */ |
| 66 | struct cgroup_subsys_state { |
| 67 | /* |
| 68 | * The cgroup that this subsystem is attached to. Useful |
| 69 | * for subsystems that want to know about the cgroup |
| 70 | * hierarchy structure |
| 71 | */ |
| 72 | struct cgroup *cgroup; |
| 73 | |
| 74 | /* |
| 75 | * State maintained by the cgroup system to allow subsystems |
| 76 | * to be "busy". Should be accessed via css_get(), |
| 77 | * css_tryget() and css_put(). |
| 78 | */ |
| 79 | |
| 80 | atomic_t refcnt; |
| 81 | |
| 82 | unsigned long flags; |
| 83 | /* ID for this css, if possible */ |
| 84 | struct css_id __rcu *id; |
| 85 | |
| 86 | /* Used to put @cgroup->dentry on the last css_put() */ |
| 87 | struct work_struct dput_work; |
| 88 | }; |
| 89 | |
| 90 | /* bits in struct cgroup_subsys_state flags field */ |
| 91 | enum { |
| 92 | CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */ |
| 93 | CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ |
| 94 | }; |
| 95 | |
| 96 | /* Caller must verify that the css is not for root cgroup */ |
| 97 | static inline void __css_get(struct cgroup_subsys_state *css, int count) |
| 98 | { |
| 99 | atomic_add(count, &css->refcnt); |
| 100 | } |
| 101 | |
| 102 | /* |
| 103 | * Call css_get() to hold a reference on the css; it can be used |
| 104 | * for a reference obtained via: |
| 105 | * - an existing ref-counted reference to the css |
| 106 | * - task->cgroups for a locked task |
| 107 | */ |
| 108 | |
| 109 | static inline void css_get(struct cgroup_subsys_state *css) |
| 110 | { |
| 111 | /* We don't need to reference count the root state */ |
| 112 | if (!(css->flags & CSS_ROOT)) |
| 113 | __css_get(css, 1); |
| 114 | } |
| 115 | |
| 116 | /* |
| 117 | * Call css_tryget() to take a reference on a css if your existing |
| 118 | * (known-valid) reference isn't already ref-counted. Returns false if |
| 119 | * the css has been destroyed. |
| 120 | */ |
| 121 | |
| 122 | extern bool __css_tryget(struct cgroup_subsys_state *css); |
| 123 | static inline bool css_tryget(struct cgroup_subsys_state *css) |
| 124 | { |
| 125 | if (css->flags & CSS_ROOT) |
| 126 | return true; |
| 127 | return __css_tryget(css); |
| 128 | } |
| 129 | |
| 130 | /* |
| 131 | * css_put() should be called to release a reference taken by |
| 132 | * css_get() or css_tryget() |
| 133 | */ |
| 134 | |
| 135 | extern void __css_put(struct cgroup_subsys_state *css); |
| 136 | static inline void css_put(struct cgroup_subsys_state *css) |
| 137 | { |
| 138 | if (!(css->flags & CSS_ROOT)) |
| 139 | __css_put(css); |
| 140 | } |
| 141 | |
| 142 | /* bits in struct cgroup flags field */ |
| 143 | enum { |
| 144 | /* Control Group is dead */ |
| 145 | CGRP_REMOVED, |
| 146 | /* |
| 147 | * Control Group has previously had a child cgroup or a task, |
| 148 | * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) |
| 149 | */ |
| 150 | CGRP_RELEASABLE, |
| 151 | /* Control Group requires release notifications to userspace */ |
| 152 | CGRP_NOTIFY_ON_RELEASE, |
| 153 | /* |
| 154 | * Clone the parent's configuration when creating a new child |
| 155 | * cpuset cgroup. For historical reasons, this option can be |
| 156 | * specified at mount time and thus is implemented here. |
| 157 | */ |
| 158 | CGRP_CPUSET_CLONE_CHILDREN, |
| 159 | /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */ |
| 160 | CGRP_SANE_BEHAVIOR, |
| 161 | }; |
| 162 | |
| 163 | struct cgroup_name { |
| 164 | struct rcu_head rcu_head; |
| 165 | char name[]; |
| 166 | }; |
| 167 | |
| 168 | struct cgroup { |
| 169 | unsigned long flags; /* "unsigned long" so bitops work */ |
| 170 | |
| 171 | /* |
| 172 | * count users of this cgroup. >0 means busy, but doesn't |
| 173 | * necessarily indicate the number of tasks in the cgroup |
| 174 | */ |
| 175 | atomic_t count; |
| 176 | |
| 177 | int id; /* ida allocated in-hierarchy ID */ |
| 178 | |
| 179 | /* |
| 180 | * We link our 'sibling' struct into our parent's 'children'. |
| 181 | * Our children link their 'sibling' into our 'children'. |
| 182 | */ |
| 183 | struct list_head sibling; /* my parent's children */ |
| 184 | struct list_head children; /* my children */ |
| 185 | struct list_head files; /* my files */ |
| 186 | |
| 187 | struct cgroup *parent; /* my parent */ |
| 188 | struct dentry *dentry; /* cgroup fs entry, RCU protected */ |
| 189 | |
| 190 | /* |
| 191 | * This is a copy of dentry->d_name, and it's needed because |
| 192 | * we can't use dentry->d_name in cgroup_path(). |
| 193 | * |
| 194 | * You must acquire rcu_read_lock() to access cgrp->name, and |
| 195 | * the only place that can change it is rename(), which is |
| 196 | * protected by parent dir's i_mutex. |
| 197 | * |
| 198 | * Normally you should use cgroup_name() wrapper rather than |
| 199 | * access it directly. |
| 200 | */ |
| 201 | struct cgroup_name __rcu *name; |
| 202 | |
| 203 | /* Private pointers for each registered subsystem */ |
| 204 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
| 205 | |
| 206 | struct cgroupfs_root *root; |
| 207 | |
| 208 | /* |
| 209 | * List of cg_cgroup_links pointing at css_sets with |
| 210 | * tasks in this cgroup. Protected by css_set_lock |
| 211 | */ |
| 212 | struct list_head css_sets; |
| 213 | |
| 214 | struct list_head allcg_node; /* cgroupfs_root->allcg_list */ |
| 215 | struct list_head cft_q_node; /* used during cftype add/rm */ |
| 216 | |
| 217 | /* |
| 218 | * Linked list running through all cgroups that can |
| 219 | * potentially be reaped by the release agent. Protected by |
| 220 | * release_list_lock |
| 221 | */ |
| 222 | struct list_head release_list; |
| 223 | |
| 224 | /* |
| 225 | * list of pidlists, up to two for each namespace (one for procs, one |
| 226 | * for tasks); created on demand. |
| 227 | */ |
| 228 | struct list_head pidlists; |
| 229 | struct mutex pidlist_mutex; |
| 230 | |
| 231 | /* For RCU-protected deletion */ |
| 232 | struct rcu_head rcu_head; |
| 233 | struct work_struct free_work; |
| 234 | |
| 235 | /* List of events which userspace want to receive */ |
| 236 | struct list_head event_list; |
| 237 | spinlock_t event_list_lock; |
| 238 | |
| 239 | /* directory xattrs */ |
| 240 | struct simple_xattrs xattrs; |
| 241 | }; |
| 242 | |
| 243 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
| 244 | |
| 245 | /* cgroupfs_root->flags */ |
| 246 | enum { |
| 247 | /* |
| 248 | * Unfortunately, cgroup core and various controllers are riddled |
| 249 | * with idiosyncrasies and pointless options. The following flag, |
| 250 | * when set, will force sane behavior - some options are forced on, |
| 251 | * others are disallowed, and some controllers will change their |
| 252 | * hierarchical or other behaviors. |
| 253 | * |
| 254 | * The set of behaviors affected by this flag are still being |
| 255 | * determined and developed and the mount option for this flag is |
| 256 | * prefixed with __DEVEL__. The prefix will be dropped once we |
| 257 | * reach the point where all behaviors are compatible with the |
| 258 | * planned unified hierarchy, which will automatically turn on this |
| 259 | * flag. |
| 260 | * |
| 261 | * The followings are the behaviors currently affected this flag. |
| 262 | * |
| 263 | * - Mount options "noprefix" and "clone_children" are disallowed. |
| 264 | * Also, cgroupfs file cgroup.clone_children is not created. |
| 265 | * |
| 266 | * - When mounting an existing superblock, mount options should |
| 267 | * match. |
| 268 | * |
| 269 | * - Remount is disallowed. |
| 270 | * |
| 271 | * - memcg: use_hierarchy is on by default and the cgroup file for |
| 272 | * the flag is not created. |
| 273 | * |
| 274 | * The followings are planned changes. |
| 275 | * |
| 276 | * - release_agent will be disallowed once replacement notification |
| 277 | * mechanism is implemented. |
| 278 | */ |
| 279 | CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), |
| 280 | |
| 281 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
| 282 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ |
| 283 | }; |
| 284 | |
| 285 | /* |
| 286 | * A cgroupfs_root represents the root of a cgroup hierarchy, and may be |
| 287 | * associated with a superblock to form an active hierarchy. This is |
| 288 | * internal to cgroup core. Don't access directly from controllers. |
| 289 | */ |
| 290 | struct cgroupfs_root { |
| 291 | struct super_block *sb; |
| 292 | |
| 293 | /* |
| 294 | * The bitmask of subsystems intended to be attached to this |
| 295 | * hierarchy |
| 296 | */ |
| 297 | unsigned long subsys_mask; |
| 298 | |
| 299 | /* Unique id for this hierarchy. */ |
| 300 | int hierarchy_id; |
| 301 | |
| 302 | /* The bitmask of subsystems currently attached to this hierarchy */ |
| 303 | unsigned long actual_subsys_mask; |
| 304 | |
| 305 | /* A list running through the attached subsystems */ |
| 306 | struct list_head subsys_list; |
| 307 | |
| 308 | /* The root cgroup for this hierarchy */ |
| 309 | struct cgroup top_cgroup; |
| 310 | |
| 311 | /* Tracks how many cgroups are currently defined in hierarchy.*/ |
| 312 | int number_of_cgroups; |
| 313 | |
| 314 | /* A list running through the active hierarchies */ |
| 315 | struct list_head root_list; |
| 316 | |
| 317 | /* All cgroups on this root, cgroup_mutex protected */ |
| 318 | struct list_head allcg_list; |
| 319 | |
| 320 | /* Hierarchy-specific flags */ |
| 321 | unsigned long flags; |
| 322 | |
| 323 | /* IDs for cgroups in this hierarchy */ |
| 324 | struct ida cgroup_ida; |
| 325 | |
| 326 | /* The path to use for release notifications. */ |
| 327 | char release_agent_path[PATH_MAX]; |
| 328 | |
| 329 | /* The name for this hierarchy - may be empty */ |
| 330 | char name[MAX_CGROUP_ROOT_NAMELEN]; |
| 331 | }; |
| 332 | |
| 333 | /* |
| 334 | * A css_set is a structure holding pointers to a set of |
| 335 | * cgroup_subsys_state objects. This saves space in the task struct |
| 336 | * object and speeds up fork()/exit(), since a single inc/dec and a |
| 337 | * list_add()/del() can bump the reference count on the entire cgroup |
| 338 | * set for a task. |
| 339 | */ |
| 340 | |
| 341 | struct css_set { |
| 342 | |
| 343 | /* Reference count */ |
| 344 | atomic_t refcount; |
| 345 | |
| 346 | /* |
| 347 | * List running through all cgroup groups in the same hash |
| 348 | * slot. Protected by css_set_lock |
| 349 | */ |
| 350 | struct hlist_node hlist; |
| 351 | |
| 352 | /* |
| 353 | * List running through all tasks using this cgroup |
| 354 | * group. Protected by css_set_lock |
| 355 | */ |
| 356 | struct list_head tasks; |
| 357 | |
| 358 | /* |
| 359 | * List of cg_cgroup_link objects on link chains from |
| 360 | * cgroups referenced from this css_set. Protected by |
| 361 | * css_set_lock |
| 362 | */ |
| 363 | struct list_head cg_links; |
| 364 | |
| 365 | /* |
| 366 | * Set of subsystem states, one for each subsystem. This array |
| 367 | * is immutable after creation apart from the init_css_set |
| 368 | * during subsystem registration (at boot time) and modular subsystem |
| 369 | * loading/unloading. |
| 370 | */ |
| 371 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
| 372 | |
| 373 | /* For RCU-protected deletion */ |
| 374 | struct rcu_head rcu_head; |
| 375 | }; |
| 376 | |
| 377 | /* |
| 378 | * cgroup_map_cb is an abstract callback API for reporting map-valued |
| 379 | * control files |
| 380 | */ |
| 381 | |
| 382 | struct cgroup_map_cb { |
| 383 | int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value); |
| 384 | void *state; |
| 385 | }; |
| 386 | |
| 387 | /* |
| 388 | * struct cftype: handler definitions for cgroup control files |
| 389 | * |
| 390 | * When reading/writing to a file: |
| 391 | * - the cgroup to use is file->f_dentry->d_parent->d_fsdata |
| 392 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
| 393 | */ |
| 394 | |
| 395 | /* cftype->flags */ |
| 396 | #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */ |
| 397 | #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */ |
| 398 | #define CFTYPE_INSANE (1U << 2) /* don't create if sane_behavior */ |
| 399 | |
| 400 | #define MAX_CFTYPE_NAME 64 |
| 401 | |
| 402 | struct cftype { |
| 403 | /* |
| 404 | * By convention, the name should begin with the name of the |
| 405 | * subsystem, followed by a period. Zero length string indicates |
| 406 | * end of cftype array. |
| 407 | */ |
| 408 | char name[MAX_CFTYPE_NAME]; |
| 409 | int private; |
| 410 | /* |
| 411 | * If not 0, file mode is set to this value, otherwise it will |
| 412 | * be figured out automatically |
| 413 | */ |
| 414 | umode_t mode; |
| 415 | |
| 416 | /* |
| 417 | * If non-zero, defines the maximum length of string that can |
| 418 | * be passed to write_string; defaults to 64 |
| 419 | */ |
| 420 | size_t max_write_len; |
| 421 | |
| 422 | /* CFTYPE_* flags */ |
| 423 | unsigned int flags; |
| 424 | |
| 425 | int (*open)(struct inode *inode, struct file *file); |
| 426 | ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft, |
| 427 | struct file *file, |
| 428 | char __user *buf, size_t nbytes, loff_t *ppos); |
| 429 | /* |
| 430 | * read_u64() is a shortcut for the common case of returning a |
| 431 | * single integer. Use it in place of read() |
| 432 | */ |
| 433 | u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft); |
| 434 | /* |
| 435 | * read_s64() is a signed version of read_u64() |
| 436 | */ |
| 437 | s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft); |
| 438 | /* |
| 439 | * read_map() is used for defining a map of key/value |
| 440 | * pairs. It should call cb->fill(cb, key, value) for each |
| 441 | * entry. The key/value pairs (and their ordering) should not |
| 442 | * change between reboots. |
| 443 | */ |
| 444 | int (*read_map)(struct cgroup *cont, struct cftype *cft, |
| 445 | struct cgroup_map_cb *cb); |
| 446 | /* |
| 447 | * read_seq_string() is used for outputting a simple sequence |
| 448 | * using seqfile. |
| 449 | */ |
| 450 | int (*read_seq_string)(struct cgroup *cont, struct cftype *cft, |
| 451 | struct seq_file *m); |
| 452 | |
| 453 | ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft, |
| 454 | struct file *file, |
| 455 | const char __user *buf, size_t nbytes, loff_t *ppos); |
| 456 | |
| 457 | /* |
| 458 | * write_u64() is a shortcut for the common case of accepting |
| 459 | * a single integer (as parsed by simple_strtoull) from |
| 460 | * userspace. Use in place of write(); return 0 or error. |
| 461 | */ |
| 462 | int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val); |
| 463 | /* |
| 464 | * write_s64() is a signed version of write_u64() |
| 465 | */ |
| 466 | int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val); |
| 467 | |
| 468 | /* |
| 469 | * write_string() is passed a nul-terminated kernelspace |
| 470 | * buffer of maximum length determined by max_write_len. |
| 471 | * Returns 0 or -ve error code. |
| 472 | */ |
| 473 | int (*write_string)(struct cgroup *cgrp, struct cftype *cft, |
| 474 | const char *buffer); |
| 475 | /* |
| 476 | * trigger() callback can be used to get some kick from the |
| 477 | * userspace, when the actual string written is not important |
| 478 | * at all. The private field can be used to determine the |
| 479 | * kick type for multiplexing. |
| 480 | */ |
| 481 | int (*trigger)(struct cgroup *cgrp, unsigned int event); |
| 482 | |
| 483 | int (*release)(struct inode *inode, struct file *file); |
| 484 | |
| 485 | /* |
| 486 | * register_event() callback will be used to add new userspace |
| 487 | * waiter for changes related to the cftype. Implement it if |
| 488 | * you want to provide this functionality. Use eventfd_signal() |
| 489 | * on eventfd to send notification to userspace. |
| 490 | */ |
| 491 | int (*register_event)(struct cgroup *cgrp, struct cftype *cft, |
| 492 | struct eventfd_ctx *eventfd, const char *args); |
| 493 | /* |
| 494 | * unregister_event() callback will be called when userspace |
| 495 | * closes the eventfd or on cgroup removing. |
| 496 | * This callback must be implemented, if you want provide |
| 497 | * notification functionality. |
| 498 | */ |
| 499 | void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft, |
| 500 | struct eventfd_ctx *eventfd); |
| 501 | }; |
| 502 | |
| 503 | /* |
| 504 | * cftype_sets describe cftypes belonging to a subsystem and are chained at |
| 505 | * cgroup_subsys->cftsets. Each cftset points to an array of cftypes |
| 506 | * terminated by zero length name. |
| 507 | */ |
| 508 | struct cftype_set { |
| 509 | struct list_head node; /* chained at subsys->cftsets */ |
| 510 | struct cftype *cfts; |
| 511 | }; |
| 512 | |
| 513 | struct cgroup_scanner { |
| 514 | struct cgroup *cg; |
| 515 | int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan); |
| 516 | void (*process_task)(struct task_struct *p, |
| 517 | struct cgroup_scanner *scan); |
| 518 | struct ptr_heap *heap; |
| 519 | void *data; |
| 520 | }; |
| 521 | |
| 522 | /* |
| 523 | * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This |
| 524 | * function can be called as long as @cgrp is accessible. |
| 525 | */ |
| 526 | static inline bool cgroup_sane_behavior(const struct cgroup *cgrp) |
| 527 | { |
| 528 | return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR; |
| 529 | } |
| 530 | |
| 531 | /* Caller should hold rcu_read_lock() */ |
| 532 | static inline const char *cgroup_name(const struct cgroup *cgrp) |
| 533 | { |
| 534 | return rcu_dereference(cgrp->name)->name; |
| 535 | } |
| 536 | |
| 537 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
| 538 | int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
| 539 | |
| 540 | int cgroup_is_removed(const struct cgroup *cgrp); |
| 541 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor); |
| 542 | |
| 543 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen); |
| 544 | |
| 545 | int cgroup_task_count(const struct cgroup *cgrp); |
| 546 | |
| 547 | /* |
| 548 | * Control Group taskset, used to pass around set of tasks to cgroup_subsys |
| 549 | * methods. |
| 550 | */ |
| 551 | struct cgroup_taskset; |
| 552 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset); |
| 553 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset); |
| 554 | struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset); |
| 555 | int cgroup_taskset_size(struct cgroup_taskset *tset); |
| 556 | |
| 557 | /** |
| 558 | * cgroup_taskset_for_each - iterate cgroup_taskset |
| 559 | * @task: the loop cursor |
| 560 | * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all |
| 561 | * @tset: taskset to iterate |
| 562 | */ |
| 563 | #define cgroup_taskset_for_each(task, skip_cgrp, tset) \ |
| 564 | for ((task) = cgroup_taskset_first((tset)); (task); \ |
| 565 | (task) = cgroup_taskset_next((tset))) \ |
| 566 | if (!(skip_cgrp) || \ |
| 567 | cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp)) |
| 568 | |
| 569 | /* |
| 570 | * Control Group subsystem type. |
| 571 | * See Documentation/cgroups/cgroups.txt for details |
| 572 | */ |
| 573 | |
| 574 | struct cgroup_subsys { |
| 575 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp); |
| 576 | int (*css_online)(struct cgroup *cgrp); |
| 577 | void (*css_offline)(struct cgroup *cgrp); |
| 578 | void (*css_free)(struct cgroup *cgrp); |
| 579 | |
| 580 | int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); |
| 581 | void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); |
| 582 | void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); |
| 583 | void (*fork)(struct task_struct *task); |
| 584 | void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp, |
| 585 | struct task_struct *task); |
| 586 | void (*bind)(struct cgroup *root); |
| 587 | |
| 588 | int subsys_id; |
| 589 | int disabled; |
| 590 | int early_init; |
| 591 | /* |
| 592 | * True if this subsys uses ID. ID is not available before cgroup_init() |
| 593 | * (not available in early_init time.) |
| 594 | */ |
| 595 | bool use_id; |
| 596 | |
| 597 | /* |
| 598 | * If %false, this subsystem is properly hierarchical - |
| 599 | * configuration, resource accounting and restriction on a parent |
| 600 | * cgroup cover those of its children. If %true, hierarchy support |
| 601 | * is broken in some ways - some subsystems ignore hierarchy |
| 602 | * completely while others are only implemented half-way. |
| 603 | * |
| 604 | * It's now disallowed to create nested cgroups if the subsystem is |
| 605 | * broken and cgroup core will emit a warning message on such |
| 606 | * cases. Eventually, all subsystems will be made properly |
| 607 | * hierarchical and this will go away. |
| 608 | */ |
| 609 | bool broken_hierarchy; |
| 610 | bool warned_broken_hierarchy; |
| 611 | |
| 612 | #define MAX_CGROUP_TYPE_NAMELEN 32 |
| 613 | const char *name; |
| 614 | |
| 615 | /* |
| 616 | * Link to parent, and list entry in parent's children. |
| 617 | * Protected by cgroup_lock() |
| 618 | */ |
| 619 | struct cgroupfs_root *root; |
| 620 | struct list_head sibling; |
| 621 | /* used when use_id == true */ |
| 622 | struct idr idr; |
| 623 | spinlock_t id_lock; |
| 624 | |
| 625 | /* list of cftype_sets */ |
| 626 | struct list_head cftsets; |
| 627 | |
| 628 | /* base cftypes, automatically [de]registered with subsys itself */ |
| 629 | struct cftype *base_cftypes; |
| 630 | struct cftype_set base_cftset; |
| 631 | |
| 632 | /* should be defined only by modular subsystems */ |
| 633 | struct module *module; |
| 634 | }; |
| 635 | |
| 636 | #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys; |
| 637 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
| 638 | #include <linux/cgroup_subsys.h> |
| 639 | #undef IS_SUBSYS_ENABLED |
| 640 | #undef SUBSYS |
| 641 | |
| 642 | static inline struct cgroup_subsys_state *cgroup_subsys_state( |
| 643 | struct cgroup *cgrp, int subsys_id) |
| 644 | { |
| 645 | return cgrp->subsys[subsys_id]; |
| 646 | } |
| 647 | |
| 648 | /* |
| 649 | * function to get the cgroup_subsys_state which allows for extra |
| 650 | * rcu_dereference_check() conditions, such as locks used during the |
| 651 | * cgroup_subsys::attach() methods. |
| 652 | */ |
| 653 | #ifdef CONFIG_PROVE_RCU |
| 654 | extern struct mutex cgroup_mutex; |
| 655 | #define task_subsys_state_check(task, subsys_id, __c) \ |
| 656 | rcu_dereference_check((task)->cgroups->subsys[(subsys_id)], \ |
| 657 | lockdep_is_held(&(task)->alloc_lock) || \ |
| 658 | lockdep_is_held(&cgroup_mutex) || (__c)) |
| 659 | #else |
| 660 | #define task_subsys_state_check(task, subsys_id, __c) \ |
| 661 | rcu_dereference((task)->cgroups->subsys[(subsys_id)]) |
| 662 | #endif |
| 663 | |
| 664 | static inline struct cgroup_subsys_state * |
| 665 | task_subsys_state(struct task_struct *task, int subsys_id) |
| 666 | { |
| 667 | return task_subsys_state_check(task, subsys_id, false); |
| 668 | } |
| 669 | |
| 670 | static inline struct cgroup* task_cgroup(struct task_struct *task, |
| 671 | int subsys_id) |
| 672 | { |
| 673 | return task_subsys_state(task, subsys_id)->cgroup; |
| 674 | } |
| 675 | |
| 676 | /** |
| 677 | * cgroup_for_each_child - iterate through children of a cgroup |
| 678 | * @pos: the cgroup * to use as the loop cursor |
| 679 | * @cgroup: cgroup whose children to walk |
| 680 | * |
| 681 | * Walk @cgroup's children. Must be called under rcu_read_lock(). A child |
| 682 | * cgroup which hasn't finished ->css_online() or already has finished |
| 683 | * ->css_offline() may show up during traversal and it's each subsystem's |
| 684 | * responsibility to verify that each @pos is alive. |
| 685 | * |
| 686 | * If a subsystem synchronizes against the parent in its ->css_online() and |
| 687 | * before starting iterating, a cgroup which finished ->css_online() is |
| 688 | * guaranteed to be visible in the future iterations. |
| 689 | */ |
| 690 | #define cgroup_for_each_child(pos, cgroup) \ |
| 691 | list_for_each_entry_rcu(pos, &(cgroup)->children, sibling) |
| 692 | |
| 693 | struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos, |
| 694 | struct cgroup *cgroup); |
| 695 | struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos); |
| 696 | |
| 697 | /** |
| 698 | * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants |
| 699 | * @pos: the cgroup * to use as the loop cursor |
| 700 | * @cgroup: cgroup whose descendants to walk |
| 701 | * |
| 702 | * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A |
| 703 | * descendant cgroup which hasn't finished ->css_online() or already has |
| 704 | * finished ->css_offline() may show up during traversal and it's each |
| 705 | * subsystem's responsibility to verify that each @pos is alive. |
| 706 | * |
| 707 | * If a subsystem synchronizes against the parent in its ->css_online() and |
| 708 | * before starting iterating, and synchronizes against @pos on each |
| 709 | * iteration, any descendant cgroup which finished ->css_offline() is |
| 710 | * guaranteed to be visible in the future iterations. |
| 711 | * |
| 712 | * In other words, the following guarantees that a descendant can't escape |
| 713 | * state updates of its ancestors. |
| 714 | * |
| 715 | * my_online(@cgrp) |
| 716 | * { |
| 717 | * Lock @cgrp->parent and @cgrp; |
| 718 | * Inherit state from @cgrp->parent; |
| 719 | * Unlock both. |
| 720 | * } |
| 721 | * |
| 722 | * my_update_state(@cgrp) |
| 723 | * { |
| 724 | * Lock @cgrp; |
| 725 | * Update @cgrp's state; |
| 726 | * Unlock @cgrp; |
| 727 | * |
| 728 | * cgroup_for_each_descendant_pre(@pos, @cgrp) { |
| 729 | * Lock @pos; |
| 730 | * Verify @pos is alive and inherit state from @pos->parent; |
| 731 | * Unlock @pos; |
| 732 | * } |
| 733 | * } |
| 734 | * |
| 735 | * As long as the inheriting step, including checking the parent state, is |
| 736 | * enclosed inside @pos locking, double-locking the parent isn't necessary |
| 737 | * while inheriting. The state update to the parent is guaranteed to be |
| 738 | * visible by walking order and, as long as inheriting operations to the |
| 739 | * same @pos are atomic to each other, multiple updates racing each other |
| 740 | * still result in the correct state. It's guaranateed that at least one |
| 741 | * inheritance happens for any cgroup after the latest update to its |
| 742 | * parent. |
| 743 | * |
| 744 | * If checking parent's state requires locking the parent, each inheriting |
| 745 | * iteration should lock and unlock both @pos->parent and @pos. |
| 746 | * |
| 747 | * Alternatively, a subsystem may choose to use a single global lock to |
| 748 | * synchronize ->css_online() and ->css_offline() against tree-walking |
| 749 | * operations. |
| 750 | */ |
| 751 | #define cgroup_for_each_descendant_pre(pos, cgroup) \ |
| 752 | for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \ |
| 753 | pos = cgroup_next_descendant_pre((pos), (cgroup))) |
| 754 | |
| 755 | struct cgroup *cgroup_next_descendant_post(struct cgroup *pos, |
| 756 | struct cgroup *cgroup); |
| 757 | |
| 758 | /** |
| 759 | * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants |
| 760 | * @pos: the cgroup * to use as the loop cursor |
| 761 | * @cgroup: cgroup whose descendants to walk |
| 762 | * |
| 763 | * Similar to cgroup_for_each_descendant_pre() but performs post-order |
| 764 | * traversal instead. Note that the walk visibility guarantee described in |
| 765 | * pre-order walk doesn't apply the same to post-order walks. |
| 766 | */ |
| 767 | #define cgroup_for_each_descendant_post(pos, cgroup) \ |
| 768 | for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \ |
| 769 | pos = cgroup_next_descendant_post((pos), (cgroup))) |
| 770 | |
| 771 | /* A cgroup_iter should be treated as an opaque object */ |
| 772 | struct cgroup_iter { |
| 773 | struct list_head *cg_link; |
| 774 | struct list_head *task; |
| 775 | }; |
| 776 | |
| 777 | /* |
| 778 | * To iterate across the tasks in a cgroup: |
| 779 | * |
| 780 | * 1) call cgroup_iter_start to initialize an iterator |
| 781 | * |
| 782 | * 2) call cgroup_iter_next() to retrieve member tasks until it |
| 783 | * returns NULL or until you want to end the iteration |
| 784 | * |
| 785 | * 3) call cgroup_iter_end() to destroy the iterator. |
| 786 | * |
| 787 | * Or, call cgroup_scan_tasks() to iterate through every task in a |
| 788 | * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling |
| 789 | * the test_task() callback, but not while calling the process_task() |
| 790 | * callback. |
| 791 | */ |
| 792 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it); |
| 793 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
| 794 | struct cgroup_iter *it); |
| 795 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it); |
| 796 | int cgroup_scan_tasks(struct cgroup_scanner *scan); |
| 797 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); |
| 798 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from); |
| 799 | |
| 800 | /* |
| 801 | * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works |
| 802 | * if cgroup_subsys.use_id == true. It can be used for looking up and scanning. |
| 803 | * CSS ID is assigned at cgroup allocation (create) automatically |
| 804 | * and removed when subsys calls free_css_id() function. This is because |
| 805 | * the lifetime of cgroup_subsys_state is subsys's matter. |
| 806 | * |
| 807 | * Looking up and scanning function should be called under rcu_read_lock(). |
| 808 | * Taking cgroup_mutex is not necessary for following calls. |
| 809 | * But the css returned by this routine can be "not populated yet" or "being |
| 810 | * destroyed". The caller should check css and cgroup's status. |
| 811 | */ |
| 812 | |
| 813 | /* |
| 814 | * Typically Called at ->destroy(), or somewhere the subsys frees |
| 815 | * cgroup_subsys_state. |
| 816 | */ |
| 817 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css); |
| 818 | |
| 819 | /* Find a cgroup_subsys_state which has given ID */ |
| 820 | |
| 821 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id); |
| 822 | |
| 823 | /* Returns true if root is ancestor of cg */ |
| 824 | bool css_is_ancestor(struct cgroup_subsys_state *cg, |
| 825 | const struct cgroup_subsys_state *root); |
| 826 | |
| 827 | /* Get id and depth of css */ |
| 828 | unsigned short css_id(struct cgroup_subsys_state *css); |
| 829 | unsigned short css_depth(struct cgroup_subsys_state *css); |
| 830 | struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id); |
| 831 | |
| 832 | #else /* !CONFIG_CGROUPS */ |
| 833 | |
| 834 | static inline int cgroup_init_early(void) { return 0; } |
| 835 | static inline int cgroup_init(void) { return 0; } |
| 836 | static inline void cgroup_fork(struct task_struct *p) {} |
| 837 | static inline void cgroup_post_fork(struct task_struct *p) {} |
| 838 | static inline void cgroup_exit(struct task_struct *p, int callbacks) {} |
| 839 | |
| 840 | static inline void cgroup_lock(void) {} |
| 841 | static inline void cgroup_unlock(void) {} |
| 842 | static inline int cgroupstats_build(struct cgroupstats *stats, |
| 843 | struct dentry *dentry) |
| 844 | { |
| 845 | return -EINVAL; |
| 846 | } |
| 847 | |
| 848 | /* No cgroups - nothing to do */ |
| 849 | static inline int cgroup_attach_task_all(struct task_struct *from, |
| 850 | struct task_struct *t) |
| 851 | { |
| 852 | return 0; |
| 853 | } |
| 854 | |
| 855 | #endif /* !CONFIG_CGROUPS */ |
| 856 | |
| 857 | #endif /* _LINUX_CGROUP_H */ |