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