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#include <linux/percpu-refcount.h> 24 25#ifdef CONFIG_CGROUPS 26 27struct cgroupfs_root; 28struct cgroup_subsys; 29struct inode; 30struct cgroup; 31struct css_id; 32struct eventfd_ctx; 33 34extern int cgroup_init_early(void); 35extern int cgroup_init(void); 36extern void cgroup_fork(struct task_struct *p); 37extern void cgroup_post_fork(struct task_struct *p); 38extern void cgroup_exit(struct task_struct *p, int run_callbacks); 39extern int cgroupstats_build(struct cgroupstats *stats, 40 struct dentry *dentry); 41extern int cgroup_load_subsys(struct cgroup_subsys *ss); 42extern void cgroup_unload_subsys(struct cgroup_subsys *ss); 43 44extern int proc_cgroup_show(struct seq_file *, void *); 45 46/* 47 * Define the enumeration of all cgroup subsystems. 48 * 49 * We define ids for builtin subsystems and then modular ones. 50 */ 51#define SUBSYS(_x) _x ## _subsys_id, 52enum cgroup_subsys_id { 53#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) 54#include <linux/cgroup_subsys.h> 55#undef IS_SUBSYS_ENABLED 56 CGROUP_BUILTIN_SUBSYS_COUNT, 57 58 __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1, 59 60#define IS_SUBSYS_ENABLED(option) IS_MODULE(option) 61#include <linux/cgroup_subsys.h> 62#undef IS_SUBSYS_ENABLED 63 CGROUP_SUBSYS_COUNT, 64}; 65#undef SUBSYS 66 67/* Per-subsystem/per-cgroup state maintained by the system. */ 68struct cgroup_subsys_state { 69 /* 70 * The cgroup that this subsystem is attached to. Useful 71 * for subsystems that want to know about the cgroup 72 * hierarchy structure 73 */ 74 struct cgroup *cgroup; 75 76 /* reference count - access via css_[try]get() and css_put() */ 77 struct percpu_ref refcnt; 78 79 unsigned long flags; 80 /* ID for this css, if possible */ 81 struct css_id __rcu *id; 82 83 /* Used to put @cgroup->dentry on the last css_put() */ 84 struct work_struct dput_work; 85}; 86 87/* bits in struct cgroup_subsys_state flags field */ 88enum { 89 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */ 90 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ 91}; 92 93/** 94 * css_get - obtain a reference on the specified css 95 * @css: target css 96 * 97 * The caller must already have a reference. 98 */ 99static inline void css_get(struct cgroup_subsys_state *css) 100{ 101 /* We don't need to reference count the root state */ 102 if (!(css->flags & CSS_ROOT)) 103 percpu_ref_get(&css->refcnt); 104} 105 106/** 107 * css_tryget - try to obtain a reference on the specified css 108 * @css: target css 109 * 110 * Obtain a reference on @css if it's alive. The caller naturally needs to 111 * ensure that @css is accessible but doesn't have to be holding a 112 * reference on it - IOW, RCU protected access is good enough for this 113 * function. Returns %true if a reference count was successfully obtained; 114 * %false otherwise. 115 */ 116static inline bool css_tryget(struct cgroup_subsys_state *css) 117{ 118 if (css->flags & CSS_ROOT) 119 return true; 120 return percpu_ref_tryget(&css->refcnt); 121} 122 123/** 124 * css_put - put a css reference 125 * @css: target css 126 * 127 * Put a reference obtained via css_get() and css_tryget(). 128 */ 129static inline void css_put(struct cgroup_subsys_state *css) 130{ 131 if (!(css->flags & CSS_ROOT)) 132 percpu_ref_put(&css->refcnt); 133} 134 135/* bits in struct cgroup flags field */ 136enum { 137 /* Control Group is dead */ 138 CGRP_DEAD, 139 /* 140 * Control Group has previously had a child cgroup or a task, 141 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) 142 */ 143 CGRP_RELEASABLE, 144 /* Control Group requires release notifications to userspace */ 145 CGRP_NOTIFY_ON_RELEASE, 146 /* 147 * Clone the parent's configuration when creating a new child 148 * cpuset cgroup. For historical reasons, this option can be 149 * specified at mount time and thus is implemented here. 150 */ 151 CGRP_CPUSET_CLONE_CHILDREN, 152 /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */ 153 CGRP_SANE_BEHAVIOR, 154}; 155 156struct cgroup_name { 157 struct rcu_head rcu_head; 158 char name[]; 159}; 160 161struct cgroup { 162 unsigned long flags; /* "unsigned long" so bitops work */ 163 164 int id; /* ida allocated in-hierarchy ID */ 165 166 /* 167 * We link our 'sibling' struct into our parent's 'children'. 168 * Our children link their 'sibling' into our 'children'. 169 */ 170 struct list_head sibling; /* my parent's children */ 171 struct list_head children; /* my children */ 172 struct list_head files; /* my files */ 173 174 struct cgroup *parent; /* my parent */ 175 struct dentry *dentry; /* cgroup fs entry, RCU protected */ 176 177 /* 178 * Monotonically increasing unique serial number which defines a 179 * uniform order among all cgroups. It's guaranteed that all 180 * ->children lists are in the ascending order of ->serial_nr. 181 * It's used to allow interrupting and resuming iterations. 182 */ 183 u64 serial_nr; 184 185 /* 186 * This is a copy of dentry->d_name, and it's needed because 187 * we can't use dentry->d_name in cgroup_path(). 188 * 189 * You must acquire rcu_read_lock() to access cgrp->name, and 190 * the only place that can change it is rename(), which is 191 * protected by parent dir's i_mutex. 192 * 193 * Normally you should use cgroup_name() wrapper rather than 194 * access it directly. 195 */ 196 struct cgroup_name __rcu *name; 197 198 /* Private pointers for each registered subsystem */ 199 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; 200 201 struct cgroupfs_root *root; 202 203 /* 204 * List of cgrp_cset_links pointing at css_sets with tasks in this 205 * cgroup. Protected by css_set_lock. 206 */ 207 struct list_head cset_links; 208 209 /* 210 * Linked list running through all cgroups that can 211 * potentially be reaped by the release agent. Protected by 212 * release_list_lock 213 */ 214 struct list_head release_list; 215 216 /* 217 * list of pidlists, up to two for each namespace (one for procs, one 218 * for tasks); created on demand. 219 */ 220 struct list_head pidlists; 221 struct mutex pidlist_mutex; 222 223 /* For css percpu_ref killing and RCU-protected deletion */ 224 struct rcu_head rcu_head; 225 struct work_struct destroy_work; 226 atomic_t css_kill_cnt; 227 228 /* List of events which userspace want to receive */ 229 struct list_head event_list; 230 spinlock_t event_list_lock; 231 232 /* directory xattrs */ 233 struct simple_xattrs xattrs; 234}; 235 236#define MAX_CGROUP_ROOT_NAMELEN 64 237 238/* cgroupfs_root->flags */ 239enum { 240 /* 241 * Unfortunately, cgroup core and various controllers are riddled 242 * with idiosyncrasies and pointless options. The following flag, 243 * when set, will force sane behavior - some options are forced on, 244 * others are disallowed, and some controllers will change their 245 * hierarchical or other behaviors. 246 * 247 * The set of behaviors affected by this flag are still being 248 * determined and developed and the mount option for this flag is 249 * prefixed with __DEVEL__. The prefix will be dropped once we 250 * reach the point where all behaviors are compatible with the 251 * planned unified hierarchy, which will automatically turn on this 252 * flag. 253 * 254 * The followings are the behaviors currently affected this flag. 255 * 256 * - Mount options "noprefix" and "clone_children" are disallowed. 257 * Also, cgroupfs file cgroup.clone_children is not created. 258 * 259 * - When mounting an existing superblock, mount options should 260 * match. 261 * 262 * - Remount is disallowed. 263 * 264 * - rename(2) is disallowed. 265 * 266 * - "tasks" is removed. Everything should be at process 267 * granularity. Use "cgroup.procs" instead. 268 * 269 * - "release_agent" and "notify_on_release" are removed. 270 * Replacement notification mechanism will be implemented. 271 * 272 * - cpuset: tasks will be kept in empty cpusets when hotplug happens 273 * and take masks of ancestors with non-empty cpus/mems, instead of 274 * being moved to an ancestor. 275 * 276 * - cpuset: a task can be moved into an empty cpuset, and again it 277 * takes masks of ancestors. 278 * 279 * - memcg: use_hierarchy is on by default and the cgroup file for 280 * the flag is not created. 281 * 282 * - blkcg: blk-throttle becomes properly hierarchical. 283 */ 284 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), 285 286 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ 287 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ 288 289 /* mount options live below bit 16 */ 290 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1, 291 292 CGRP_ROOT_SUBSYS_BOUND = (1 << 16), /* subsystems finished binding */ 293}; 294 295/* 296 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be 297 * associated with a superblock to form an active hierarchy. This is 298 * internal to cgroup core. Don't access directly from controllers. 299 */ 300struct cgroupfs_root { 301 struct super_block *sb; 302 303 /* The bitmask of subsystems attached to this hierarchy */ 304 unsigned long subsys_mask; 305 306 /* Unique id for this hierarchy. */ 307 int hierarchy_id; 308 309 /* A list running through the attached subsystems */ 310 struct list_head subsys_list; 311 312 /* The root cgroup for this hierarchy */ 313 struct cgroup top_cgroup; 314 315 /* Tracks how many cgroups are currently defined in hierarchy.*/ 316 int number_of_cgroups; 317 318 /* A list running through the active hierarchies */ 319 struct list_head root_list; 320 321 /* Hierarchy-specific flags */ 322 unsigned long flags; 323 324 /* IDs for cgroups in this hierarchy */ 325 struct ida cgroup_ida; 326 327 /* The path to use for release notifications. */ 328 char release_agent_path[PATH_MAX]; 329 330 /* The name for this hierarchy - may be empty */ 331 char name[MAX_CGROUP_ROOT_NAMELEN]; 332}; 333 334/* 335 * A css_set is a structure holding pointers to a set of 336 * cgroup_subsys_state objects. This saves space in the task struct 337 * object and speeds up fork()/exit(), since a single inc/dec and a 338 * list_add()/del() can bump the reference count on the entire cgroup 339 * set for a task. 340 */ 341 342struct css_set { 343 344 /* Reference count */ 345 atomic_t refcount; 346 347 /* 348 * List running through all cgroup groups in the same hash 349 * slot. Protected by css_set_lock 350 */ 351 struct hlist_node hlist; 352 353 /* 354 * List running through all tasks using this cgroup 355 * group. Protected by css_set_lock 356 */ 357 struct list_head tasks; 358 359 /* 360 * List of cgrp_cset_links pointing at cgroups referenced from this 361 * css_set. Protected by css_set_lock. 362 */ 363 struct list_head cgrp_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 382struct 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 */ 396enum { 397 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cg */ 398 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cg */ 399 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */ 400}; 401 402#define MAX_CFTYPE_NAME 64 403 404struct cftype { 405 /* 406 * By convention, the name should begin with the name of the 407 * subsystem, followed by a period. Zero length string indicates 408 * end of cftype array. 409 */ 410 char name[MAX_CFTYPE_NAME]; 411 int private; 412 /* 413 * If not 0, file mode is set to this value, otherwise it will 414 * be figured out automatically 415 */ 416 umode_t mode; 417 418 /* 419 * If non-zero, defines the maximum length of string that can 420 * be passed to write_string; defaults to 64 421 */ 422 size_t max_write_len; 423 424 /* CFTYPE_* flags */ 425 unsigned int flags; 426 427 int (*open)(struct inode *inode, struct file *file); 428 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft, 429 struct file *file, 430 char __user *buf, size_t nbytes, loff_t *ppos); 431 /* 432 * read_u64() is a shortcut for the common case of returning a 433 * single integer. Use it in place of read() 434 */ 435 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft); 436 /* 437 * read_s64() is a signed version of read_u64() 438 */ 439 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft); 440 /* 441 * read_map() is used for defining a map of key/value 442 * pairs. It should call cb->fill(cb, key, value) for each 443 * entry. The key/value pairs (and their ordering) should not 444 * change between reboots. 445 */ 446 int (*read_map)(struct cgroup *cgrp, struct cftype *cft, 447 struct cgroup_map_cb *cb); 448 /* 449 * read_seq_string() is used for outputting a simple sequence 450 * using seqfile. 451 */ 452 int (*read_seq_string)(struct cgroup *cgrp, struct cftype *cft, 453 struct seq_file *m); 454 455 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft, 456 struct file *file, 457 const char __user *buf, size_t nbytes, loff_t *ppos); 458 459 /* 460 * write_u64() is a shortcut for the common case of accepting 461 * a single integer (as parsed by simple_strtoull) from 462 * userspace. Use in place of write(); return 0 or error. 463 */ 464 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val); 465 /* 466 * write_s64() is a signed version of write_u64() 467 */ 468 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val); 469 470 /* 471 * write_string() is passed a nul-terminated kernelspace 472 * buffer of maximum length determined by max_write_len. 473 * Returns 0 or -ve error code. 474 */ 475 int (*write_string)(struct cgroup *cgrp, struct cftype *cft, 476 const char *buffer); 477 /* 478 * trigger() callback can be used to get some kick from the 479 * userspace, when the actual string written is not important 480 * at all. The private field can be used to determine the 481 * kick type for multiplexing. 482 */ 483 int (*trigger)(struct cgroup *cgrp, unsigned int event); 484 485 int (*release)(struct inode *inode, struct file *file); 486 487 /* 488 * register_event() callback will be used to add new userspace 489 * waiter for changes related to the cftype. Implement it if 490 * you want to provide this functionality. Use eventfd_signal() 491 * on eventfd to send notification to userspace. 492 */ 493 int (*register_event)(struct cgroup *cgrp, struct cftype *cft, 494 struct eventfd_ctx *eventfd, const char *args); 495 /* 496 * unregister_event() callback will be called when userspace 497 * closes the eventfd or on cgroup removing. 498 * This callback must be implemented, if you want provide 499 * notification functionality. 500 */ 501 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft, 502 struct eventfd_ctx *eventfd); 503}; 504 505/* 506 * cftype_sets describe cftypes belonging to a subsystem and are chained at 507 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes 508 * terminated by zero length name. 509 */ 510struct cftype_set { 511 struct list_head node; /* chained at subsys->cftsets */ 512 struct cftype *cfts; 513}; 514 515struct cgroup_scanner { 516 struct cgroup *cg; 517 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan); 518 void (*process_task)(struct task_struct *p, 519 struct cgroup_scanner *scan); 520 struct ptr_heap *heap; 521 void *data; 522}; 523 524/* 525 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This 526 * function can be called as long as @cgrp is accessible. 527 */ 528static inline bool cgroup_sane_behavior(const struct cgroup *cgrp) 529{ 530 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR; 531} 532 533/* Caller should hold rcu_read_lock() */ 534static inline const char *cgroup_name(const struct cgroup *cgrp) 535{ 536 return rcu_dereference(cgrp->name)->name; 537} 538 539int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); 540int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); 541 542bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor); 543 544int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen); 545int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen); 546 547int cgroup_task_count(const struct cgroup *cgrp); 548 549/* 550 * Control Group taskset, used to pass around set of tasks to cgroup_subsys 551 * methods. 552 */ 553struct cgroup_taskset; 554struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset); 555struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset); 556struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset); 557int cgroup_taskset_size(struct cgroup_taskset *tset); 558 559/** 560 * cgroup_taskset_for_each - iterate cgroup_taskset 561 * @task: the loop cursor 562 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all 563 * @tset: taskset to iterate 564 */ 565#define cgroup_taskset_for_each(task, skip_cgrp, tset) \ 566 for ((task) = cgroup_taskset_first((tset)); (task); \ 567 (task) = cgroup_taskset_next((tset))) \ 568 if (!(skip_cgrp) || \ 569 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp)) 570 571/* 572 * Control Group subsystem type. 573 * See Documentation/cgroups/cgroups.txt for details 574 */ 575 576struct cgroup_subsys { 577 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp); 578 int (*css_online)(struct cgroup *cgrp); 579 void (*css_offline)(struct cgroup *cgrp); 580 void (*css_free)(struct cgroup *cgrp); 581 582 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); 583 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); 584 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset); 585 void (*fork)(struct task_struct *task); 586 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp, 587 struct task_struct *task); 588 void (*bind)(struct cgroup *root); 589 590 int subsys_id; 591 int disabled; 592 int early_init; 593 /* 594 * True if this subsys uses ID. ID is not available before cgroup_init() 595 * (not available in early_init time.) 596 */ 597 bool use_id; 598 599 /* 600 * If %false, this subsystem is properly hierarchical - 601 * configuration, resource accounting and restriction on a parent 602 * cgroup cover those of its children. If %true, hierarchy support 603 * is broken in some ways - some subsystems ignore hierarchy 604 * completely while others are only implemented half-way. 605 * 606 * It's now disallowed to create nested cgroups if the subsystem is 607 * broken and cgroup core will emit a warning message on such 608 * cases. Eventually, all subsystems will be made properly 609 * hierarchical and this will go away. 610 */ 611 bool broken_hierarchy; 612 bool warned_broken_hierarchy; 613 614#define MAX_CGROUP_TYPE_NAMELEN 32 615 const char *name; 616 617 /* 618 * Link to parent, and list entry in parent's children. 619 * Protected by cgroup_lock() 620 */ 621 struct cgroupfs_root *root; 622 struct list_head sibling; 623 /* used when use_id == true */ 624 struct idr idr; 625 spinlock_t id_lock; 626 627 /* list of cftype_sets */ 628 struct list_head cftsets; 629 630 /* base cftypes, automatically [de]registered with subsys itself */ 631 struct cftype *base_cftypes; 632 struct cftype_set base_cftset; 633 634 /* should be defined only by modular subsystems */ 635 struct module *module; 636}; 637 638#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys; 639#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) 640#include <linux/cgroup_subsys.h> 641#undef IS_SUBSYS_ENABLED 642#undef SUBSYS 643 644static inline struct cgroup_subsys_state *cgroup_subsys_state( 645 struct cgroup *cgrp, int subsys_id) 646{ 647 return cgrp->subsys[subsys_id]; 648} 649 650/** 651 * task_css_set_check - obtain a task's css_set with extra access conditions 652 * @task: the task to obtain css_set for 653 * @__c: extra condition expression to be passed to rcu_dereference_check() 654 * 655 * A task's css_set is RCU protected, initialized and exited while holding 656 * task_lock(), and can only be modified while holding both cgroup_mutex 657 * and task_lock() while the task is alive. This macro verifies that the 658 * caller is inside proper critical section and returns @task's css_set. 659 * 660 * The caller can also specify additional allowed conditions via @__c, such 661 * as locks used during the cgroup_subsys::attach() methods. 662 */ 663#ifdef CONFIG_PROVE_RCU 664extern struct mutex cgroup_mutex; 665#define task_css_set_check(task, __c) \ 666 rcu_dereference_check((task)->cgroups, \ 667 lockdep_is_held(&(task)->alloc_lock) || \ 668 lockdep_is_held(&cgroup_mutex) || (__c)) 669#else 670#define task_css_set_check(task, __c) \ 671 rcu_dereference((task)->cgroups) 672#endif 673 674/** 675 * task_subsys_state_check - obtain css for (task, subsys) w/ extra access conds 676 * @task: the target task 677 * @subsys_id: the target subsystem ID 678 * @__c: extra condition expression to be passed to rcu_dereference_check() 679 * 680 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The 681 * synchronization rules are the same as task_css_set_check(). 682 */ 683#define task_subsys_state_check(task, subsys_id, __c) \ 684 task_css_set_check((task), (__c))->subsys[(subsys_id)] 685 686/** 687 * task_css_set - obtain a task's css_set 688 * @task: the task to obtain css_set for 689 * 690 * See task_css_set_check(). 691 */ 692static inline struct css_set *task_css_set(struct task_struct *task) 693{ 694 return task_css_set_check(task, false); 695} 696 697/** 698 * task_subsys_state - obtain css for (task, subsys) 699 * @task: the target task 700 * @subsys_id: the target subsystem ID 701 * 702 * See task_subsys_state_check(). 703 */ 704static inline struct cgroup_subsys_state * 705task_subsys_state(struct task_struct *task, int subsys_id) 706{ 707 return task_subsys_state_check(task, subsys_id, false); 708} 709 710static inline struct cgroup* task_cgroup(struct task_struct *task, 711 int subsys_id) 712{ 713 return task_subsys_state(task, subsys_id)->cgroup; 714} 715 716struct cgroup *cgroup_next_sibling(struct cgroup *pos); 717 718/** 719 * cgroup_for_each_child - iterate through children of a cgroup 720 * @pos: the cgroup * to use as the loop cursor 721 * @cgrp: cgroup whose children to walk 722 * 723 * Walk @cgrp's children. Must be called under rcu_read_lock(). A child 724 * cgroup which hasn't finished ->css_online() or already has finished 725 * ->css_offline() may show up during traversal and it's each subsystem's 726 * responsibility to verify that each @pos is alive. 727 * 728 * If a subsystem synchronizes against the parent in its ->css_online() and 729 * before starting iterating, a cgroup which finished ->css_online() is 730 * guaranteed to be visible in the future iterations. 731 * 732 * It is allowed to temporarily drop RCU read lock during iteration. The 733 * caller is responsible for ensuring that @pos remains accessible until 734 * the start of the next iteration by, for example, bumping the css refcnt. 735 */ 736#define cgroup_for_each_child(pos, cgrp) \ 737 for ((pos) = list_first_or_null_rcu(&(cgrp)->children, \ 738 struct cgroup, sibling); \ 739 (pos); (pos) = cgroup_next_sibling((pos))) 740 741struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos, 742 struct cgroup *cgroup); 743struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos); 744 745/** 746 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants 747 * @pos: the cgroup * to use as the loop cursor 748 * @cgroup: cgroup whose descendants to walk 749 * 750 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A 751 * descendant cgroup which hasn't finished ->css_online() or already has 752 * finished ->css_offline() may show up during traversal and it's each 753 * subsystem's responsibility to verify that each @pos is alive. 754 * 755 * If a subsystem synchronizes against the parent in its ->css_online() and 756 * before starting iterating, and synchronizes against @pos on each 757 * iteration, any descendant cgroup which finished ->css_online() is 758 * guaranteed to be visible in the future iterations. 759 * 760 * In other words, the following guarantees that a descendant can't escape 761 * state updates of its ancestors. 762 * 763 * my_online(@cgrp) 764 * { 765 * Lock @cgrp->parent and @cgrp; 766 * Inherit state from @cgrp->parent; 767 * Unlock both. 768 * } 769 * 770 * my_update_state(@cgrp) 771 * { 772 * Lock @cgrp; 773 * Update @cgrp's state; 774 * Unlock @cgrp; 775 * 776 * cgroup_for_each_descendant_pre(@pos, @cgrp) { 777 * Lock @pos; 778 * Verify @pos is alive and inherit state from @pos->parent; 779 * Unlock @pos; 780 * } 781 * } 782 * 783 * As long as the inheriting step, including checking the parent state, is 784 * enclosed inside @pos locking, double-locking the parent isn't necessary 785 * while inheriting. The state update to the parent is guaranteed to be 786 * visible by walking order and, as long as inheriting operations to the 787 * same @pos are atomic to each other, multiple updates racing each other 788 * still result in the correct state. It's guaranateed that at least one 789 * inheritance happens for any cgroup after the latest update to its 790 * parent. 791 * 792 * If checking parent's state requires locking the parent, each inheriting 793 * iteration should lock and unlock both @pos->parent and @pos. 794 * 795 * Alternatively, a subsystem may choose to use a single global lock to 796 * synchronize ->css_online() and ->css_offline() against tree-walking 797 * operations. 798 * 799 * It is allowed to temporarily drop RCU read lock during iteration. The 800 * caller is responsible for ensuring that @pos remains accessible until 801 * the start of the next iteration by, for example, bumping the css refcnt. 802 */ 803#define cgroup_for_each_descendant_pre(pos, cgroup) \ 804 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \ 805 pos = cgroup_next_descendant_pre((pos), (cgroup))) 806 807struct cgroup *cgroup_next_descendant_post(struct cgroup *pos, 808 struct cgroup *cgroup); 809 810/** 811 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants 812 * @pos: the cgroup * to use as the loop cursor 813 * @cgroup: cgroup whose descendants to walk 814 * 815 * Similar to cgroup_for_each_descendant_pre() but performs post-order 816 * traversal instead. Note that the walk visibility guarantee described in 817 * pre-order walk doesn't apply the same to post-order walks. 818 */ 819#define cgroup_for_each_descendant_post(pos, cgroup) \ 820 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \ 821 pos = cgroup_next_descendant_post((pos), (cgroup))) 822 823/* A cgroup_iter should be treated as an opaque object */ 824struct cgroup_iter { 825 struct list_head *cset_link; 826 struct list_head *task; 827}; 828 829/* 830 * To iterate across the tasks in a cgroup: 831 * 832 * 1) call cgroup_iter_start to initialize an iterator 833 * 834 * 2) call cgroup_iter_next() to retrieve member tasks until it 835 * returns NULL or until you want to end the iteration 836 * 837 * 3) call cgroup_iter_end() to destroy the iterator. 838 * 839 * Or, call cgroup_scan_tasks() to iterate through every task in a 840 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling 841 * the test_task() callback, but not while calling the process_task() 842 * callback. 843 */ 844void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it); 845struct task_struct *cgroup_iter_next(struct cgroup *cgrp, 846 struct cgroup_iter *it); 847void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it); 848int cgroup_scan_tasks(struct cgroup_scanner *scan); 849int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); 850int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from); 851 852/* 853 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works 854 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning. 855 * CSS ID is assigned at cgroup allocation (create) automatically 856 * and removed when subsys calls free_css_id() function. This is because 857 * the lifetime of cgroup_subsys_state is subsys's matter. 858 * 859 * Looking up and scanning function should be called under rcu_read_lock(). 860 * Taking cgroup_mutex is not necessary for following calls. 861 * But the css returned by this routine can be "not populated yet" or "being 862 * destroyed". The caller should check css and cgroup's status. 863 */ 864 865/* 866 * Typically Called at ->destroy(), or somewhere the subsys frees 867 * cgroup_subsys_state. 868 */ 869void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css); 870 871/* Find a cgroup_subsys_state which has given ID */ 872 873struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id); 874 875/* Returns true if root is ancestor of cg */ 876bool css_is_ancestor(struct cgroup_subsys_state *cg, 877 const struct cgroup_subsys_state *root); 878 879/* Get id and depth of css */ 880unsigned short css_id(struct cgroup_subsys_state *css); 881struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id); 882 883#else /* !CONFIG_CGROUPS */ 884 885static inline int cgroup_init_early(void) { return 0; } 886static inline int cgroup_init(void) { return 0; } 887static inline void cgroup_fork(struct task_struct *p) {} 888static inline void cgroup_post_fork(struct task_struct *p) {} 889static inline void cgroup_exit(struct task_struct *p, int callbacks) {} 890 891static inline int cgroupstats_build(struct cgroupstats *stats, 892 struct dentry *dentry) 893{ 894 return -EINVAL; 895} 896 897/* No cgroups - nothing to do */ 898static inline int cgroup_attach_task_all(struct task_struct *from, 899 struct task_struct *t) 900{ 901 return 0; 902} 903 904#endif /* !CONFIG_CGROUPS */ 905 906#endif /* _LINUX_CGROUP_H */ 907