linux/include/linux/cgroup.h
<<
>>
Prefs
   1/* SPDX-License-Identifier: GPL-2.0 */
   2#ifndef _LINUX_CGROUP_H
   3#define _LINUX_CGROUP_H
   4/*
   5 *  cgroup interface
   6 *
   7 *  Copyright (C) 2003 BULL SA
   8 *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
   9 *
  10 */
  11
  12#include <linux/sched.h>
  13#include <linux/cpumask.h>
  14#include <linux/nodemask.h>
  15#include <linux/rculist.h>
  16#include <linux/cgroupstats.h>
  17#include <linux/fs.h>
  18#include <linux/seq_file.h>
  19#include <linux/kernfs.h>
  20#include <linux/jump_label.h>
  21#include <linux/types.h>
  22#include <linux/ns_common.h>
  23#include <linux/nsproxy.h>
  24#include <linux/user_namespace.h>
  25#include <linux/refcount.h>
  26#include <linux/kernel_stat.h>
  27
  28#include <linux/cgroup-defs.h>
  29
  30#ifdef CONFIG_CGROUPS
  31
  32/*
  33 * All weight knobs on the default hierarhcy should use the following min,
  34 * default and max values.  The default value is the logarithmic center of
  35 * MIN and MAX and allows 100x to be expressed in both directions.
  36 */
  37#define CGROUP_WEIGHT_MIN               1
  38#define CGROUP_WEIGHT_DFL               100
  39#define CGROUP_WEIGHT_MAX               10000
  40
  41/* walk only threadgroup leaders */
  42#define CSS_TASK_ITER_PROCS             (1U << 0)
  43/* walk all threaded css_sets in the domain */
  44#define CSS_TASK_ITER_THREADED          (1U << 1)
  45
  46/* internal flags */
  47#define CSS_TASK_ITER_SKIPPED           (1U << 16)
  48
  49/* a css_task_iter should be treated as an opaque object */
  50struct css_task_iter {
  51        struct cgroup_subsys            *ss;
  52        unsigned int                    flags;
  53
  54        struct list_head                *cset_pos;
  55        struct list_head                *cset_head;
  56
  57        struct list_head                *tcset_pos;
  58        struct list_head                *tcset_head;
  59
  60        struct list_head                *task_pos;
  61        struct list_head                *tasks_head;
  62        struct list_head                *mg_tasks_head;
  63        struct list_head                *dying_tasks_head;
  64
  65        struct css_set                  *cur_cset;
  66        struct css_set                  *cur_dcset;
  67        struct task_struct              *cur_task;
  68        struct list_head                iters_node;     /* css_set->task_iters */
  69};
  70
  71extern struct cgroup_root cgrp_dfl_root;
  72extern struct css_set init_css_set;
  73
  74#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
  75#include <linux/cgroup_subsys.h>
  76#undef SUBSYS
  77
  78#define SUBSYS(_x)                                                              \
  79        extern struct static_key_true _x ## _cgrp_subsys_enabled_key;           \
  80        extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
  81#include <linux/cgroup_subsys.h>
  82#undef SUBSYS
  83
  84/**
  85 * cgroup_subsys_enabled - fast test on whether a subsys is enabled
  86 * @ss: subsystem in question
  87 */
  88#define cgroup_subsys_enabled(ss)                                               \
  89        static_branch_likely(&ss ## _enabled_key)
  90
  91/**
  92 * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
  93 * @ss: subsystem in question
  94 */
  95#define cgroup_subsys_on_dfl(ss)                                                \
  96        static_branch_likely(&ss ## _on_dfl_key)
  97
  98bool css_has_online_children(struct cgroup_subsys_state *css);
  99struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
 100struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgroup,
 101                                         struct cgroup_subsys *ss);
 102struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
 103                                             struct cgroup_subsys *ss);
 104struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
 105                                                       struct cgroup_subsys *ss);
 106
 107struct cgroup *cgroup_get_from_path(const char *path);
 108struct cgroup *cgroup_get_from_fd(int fd);
 109
 110int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
 111int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
 112
 113int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
 114int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
 115int cgroup_rm_cftypes(struct cftype *cfts);
 116void cgroup_file_notify(struct cgroup_file *cfile);
 117
 118int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
 119int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
 120int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
 121                     struct pid *pid, struct task_struct *tsk);
 122
 123void cgroup_fork(struct task_struct *p);
 124extern int cgroup_can_fork(struct task_struct *p);
 125extern void cgroup_cancel_fork(struct task_struct *p);
 126extern void cgroup_post_fork(struct task_struct *p);
 127void cgroup_exit(struct task_struct *p);
 128void cgroup_release(struct task_struct *p);
 129void cgroup_free(struct task_struct *p);
 130
 131int cgroup_init_early(void);
 132int cgroup_init(void);
 133
 134int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v);
 135
 136/*
 137 * Iteration helpers and macros.
 138 */
 139
 140struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
 141                                           struct cgroup_subsys_state *parent);
 142struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
 143                                                    struct cgroup_subsys_state *css);
 144struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
 145struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
 146                                                     struct cgroup_subsys_state *css);
 147
 148struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
 149                                         struct cgroup_subsys_state **dst_cssp);
 150struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
 151                                        struct cgroup_subsys_state **dst_cssp);
 152
 153void cgroup_enable_task_cg_lists(void);
 154void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
 155                         struct css_task_iter *it);
 156struct task_struct *css_task_iter_next(struct css_task_iter *it);
 157void css_task_iter_end(struct css_task_iter *it);
 158
 159/**
 160 * css_for_each_child - iterate through children of a css
 161 * @pos: the css * to use as the loop cursor
 162 * @parent: css whose children to walk
 163 *
 164 * Walk @parent's children.  Must be called under rcu_read_lock().
 165 *
 166 * If a subsystem synchronizes ->css_online() and the start of iteration, a
 167 * css which finished ->css_online() is guaranteed to be visible in the
 168 * future iterations and will stay visible until the last reference is put.
 169 * A css which hasn't finished ->css_online() or already finished
 170 * ->css_offline() may show up during traversal.  It's each subsystem's
 171 * responsibility to synchronize against on/offlining.
 172 *
 173 * It is allowed to temporarily drop RCU read lock during iteration.  The
 174 * caller is responsible for ensuring that @pos remains accessible until
 175 * the start of the next iteration by, for example, bumping the css refcnt.
 176 */
 177#define css_for_each_child(pos, parent)                                 \
 178        for ((pos) = css_next_child(NULL, (parent)); (pos);             \
 179             (pos) = css_next_child((pos), (parent)))
 180
 181/**
 182 * css_for_each_descendant_pre - pre-order walk of a css's descendants
 183 * @pos: the css * to use as the loop cursor
 184 * @root: css whose descendants to walk
 185 *
 186 * Walk @root's descendants.  @root is included in the iteration and the
 187 * first node to be visited.  Must be called under rcu_read_lock().
 188 *
 189 * If a subsystem synchronizes ->css_online() and the start of iteration, a
 190 * css which finished ->css_online() is guaranteed to be visible in the
 191 * future iterations and will stay visible until the last reference is put.
 192 * A css which hasn't finished ->css_online() or already finished
 193 * ->css_offline() may show up during traversal.  It's each subsystem's
 194 * responsibility to synchronize against on/offlining.
 195 *
 196 * For example, the following guarantees that a descendant can't escape
 197 * state updates of its ancestors.
 198 *
 199 * my_online(@css)
 200 * {
 201 *      Lock @css's parent and @css;
 202 *      Inherit state from the parent;
 203 *      Unlock both.
 204 * }
 205 *
 206 * my_update_state(@css)
 207 * {
 208 *      css_for_each_descendant_pre(@pos, @css) {
 209 *              Lock @pos;
 210 *              if (@pos == @css)
 211 *                      Update @css's state;
 212 *              else
 213 *                      Verify @pos is alive and inherit state from its parent;
 214 *              Unlock @pos;
 215 *      }
 216 * }
 217 *
 218 * As long as the inheriting step, including checking the parent state, is
 219 * enclosed inside @pos locking, double-locking the parent isn't necessary
 220 * while inheriting.  The state update to the parent is guaranteed to be
 221 * visible by walking order and, as long as inheriting operations to the
 222 * same @pos are atomic to each other, multiple updates racing each other
 223 * still result in the correct state.  It's guaranateed that at least one
 224 * inheritance happens for any css after the latest update to its parent.
 225 *
 226 * If checking parent's state requires locking the parent, each inheriting
 227 * iteration should lock and unlock both @pos->parent and @pos.
 228 *
 229 * Alternatively, a subsystem may choose to use a single global lock to
 230 * synchronize ->css_online() and ->css_offline() against tree-walking
 231 * operations.
 232 *
 233 * It is allowed to temporarily drop RCU read lock during iteration.  The
 234 * caller is responsible for ensuring that @pos remains accessible until
 235 * the start of the next iteration by, for example, bumping the css refcnt.
 236 */
 237#define css_for_each_descendant_pre(pos, css)                           \
 238        for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);       \
 239             (pos) = css_next_descendant_pre((pos), (css)))
 240
 241/**
 242 * css_for_each_descendant_post - post-order walk of a css's descendants
 243 * @pos: the css * to use as the loop cursor
 244 * @css: css whose descendants to walk
 245 *
 246 * Similar to css_for_each_descendant_pre() but performs post-order
 247 * traversal instead.  @root is included in the iteration and the last
 248 * node to be visited.
 249 *
 250 * If a subsystem synchronizes ->css_online() and the start of iteration, a
 251 * css which finished ->css_online() is guaranteed to be visible in the
 252 * future iterations and will stay visible until the last reference is put.
 253 * A css which hasn't finished ->css_online() or already finished
 254 * ->css_offline() may show up during traversal.  It's each subsystem's
 255 * responsibility to synchronize against on/offlining.
 256 *
 257 * Note that the walk visibility guarantee example described in pre-order
 258 * walk doesn't apply the same to post-order walks.
 259 */
 260#define css_for_each_descendant_post(pos, css)                          \
 261        for ((pos) = css_next_descendant_post(NULL, (css)); (pos);      \
 262             (pos) = css_next_descendant_post((pos), (css)))
 263
 264/**
 265 * cgroup_taskset_for_each - iterate cgroup_taskset
 266 * @task: the loop cursor
 267 * @dst_css: the destination css
 268 * @tset: taskset to iterate
 269 *
 270 * @tset may contain multiple tasks and they may belong to multiple
 271 * processes.
 272 *
 273 * On the v2 hierarchy, there may be tasks from multiple processes and they
 274 * may not share the source or destination csses.
 275 *
 276 * On traditional hierarchies, when there are multiple tasks in @tset, if a
 277 * task of a process is in @tset, all tasks of the process are in @tset.
 278 * Also, all are guaranteed to share the same source and destination csses.
 279 *
 280 * Iteration is not in any specific order.
 281 */
 282#define cgroup_taskset_for_each(task, dst_css, tset)                    \
 283        for ((task) = cgroup_taskset_first((tset), &(dst_css));         \
 284             (task);                                                    \
 285             (task) = cgroup_taskset_next((tset), &(dst_css)))
 286
 287/**
 288 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
 289 * @leader: the loop cursor
 290 * @dst_css: the destination css
 291 * @tset: taskset to iterate
 292 *
 293 * Iterate threadgroup leaders of @tset.  For single-task migrations, @tset
 294 * may not contain any.
 295 */
 296#define cgroup_taskset_for_each_leader(leader, dst_css, tset)           \
 297        for ((leader) = cgroup_taskset_first((tset), &(dst_css));       \
 298             (leader);                                                  \
 299             (leader) = cgroup_taskset_next((tset), &(dst_css)))        \
 300                if ((leader) != (leader)->group_leader)                 \
 301                        ;                                               \
 302                else
 303
 304/*
 305 * Inline functions.
 306 */
 307
 308/**
 309 * css_get - obtain a reference on the specified css
 310 * @css: target css
 311 *
 312 * The caller must already have a reference.
 313 */
 314static inline void css_get(struct cgroup_subsys_state *css)
 315{
 316        if (!(css->flags & CSS_NO_REF))
 317                percpu_ref_get(&css->refcnt);
 318}
 319
 320/**
 321 * css_get_many - obtain references on the specified css
 322 * @css: target css
 323 * @n: number of references to get
 324 *
 325 * The caller must already have a reference.
 326 */
 327static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
 328{
 329        if (!(css->flags & CSS_NO_REF))
 330                percpu_ref_get_many(&css->refcnt, n);
 331}
 332
 333/**
 334 * css_tryget - try to obtain a reference on the specified css
 335 * @css: target css
 336 *
 337 * Obtain a reference on @css unless it already has reached zero and is
 338 * being released.  This function doesn't care whether @css is on or
 339 * offline.  The caller naturally needs to ensure that @css is accessible
 340 * but doesn't have to be holding a reference on it - IOW, RCU protected
 341 * access is good enough for this function.  Returns %true if a reference
 342 * count was successfully obtained; %false otherwise.
 343 */
 344static inline bool css_tryget(struct cgroup_subsys_state *css)
 345{
 346        if (!(css->flags & CSS_NO_REF))
 347                return percpu_ref_tryget(&css->refcnt);
 348        return true;
 349}
 350
 351/**
 352 * css_tryget_online - try to obtain a reference on the specified css if online
 353 * @css: target css
 354 *
 355 * Obtain a reference on @css if it's online.  The caller naturally needs
 356 * to ensure that @css is accessible but doesn't have to be holding a
 357 * reference on it - IOW, RCU protected access is good enough for this
 358 * function.  Returns %true if a reference count was successfully obtained;
 359 * %false otherwise.
 360 */
 361static inline bool css_tryget_online(struct cgroup_subsys_state *css)
 362{
 363        if (!(css->flags & CSS_NO_REF))
 364                return percpu_ref_tryget_live(&css->refcnt);
 365        return true;
 366}
 367
 368/**
 369 * css_is_dying - test whether the specified css is dying
 370 * @css: target css
 371 *
 372 * Test whether @css is in the process of offlining or already offline.  In
 373 * most cases, ->css_online() and ->css_offline() callbacks should be
 374 * enough; however, the actual offline operations are RCU delayed and this
 375 * test returns %true also when @css is scheduled to be offlined.
 376 *
 377 * This is useful, for example, when the use case requires synchronous
 378 * behavior with respect to cgroup removal.  cgroup removal schedules css
 379 * offlining but the css can seem alive while the operation is being
 380 * delayed.  If the delay affects user visible semantics, this test can be
 381 * used to resolve the situation.
 382 */
 383static inline bool css_is_dying(struct cgroup_subsys_state *css)
 384{
 385        return !(css->flags & CSS_NO_REF) && percpu_ref_is_dying(&css->refcnt);
 386}
 387
 388/**
 389 * css_put - put a css reference
 390 * @css: target css
 391 *
 392 * Put a reference obtained via css_get() and css_tryget_online().
 393 */
 394static inline void css_put(struct cgroup_subsys_state *css)
 395{
 396        if (!(css->flags & CSS_NO_REF))
 397                percpu_ref_put(&css->refcnt);
 398}
 399
 400/**
 401 * css_put_many - put css references
 402 * @css: target css
 403 * @n: number of references to put
 404 *
 405 * Put references obtained via css_get() and css_tryget_online().
 406 */
 407static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
 408{
 409        if (!(css->flags & CSS_NO_REF))
 410                percpu_ref_put_many(&css->refcnt, n);
 411}
 412
 413static inline void cgroup_get(struct cgroup *cgrp)
 414{
 415        css_get(&cgrp->self);
 416}
 417
 418static inline bool cgroup_tryget(struct cgroup *cgrp)
 419{
 420        return css_tryget(&cgrp->self);
 421}
 422
 423static inline void cgroup_put(struct cgroup *cgrp)
 424{
 425        css_put(&cgrp->self);
 426}
 427
 428/**
 429 * task_css_set_check - obtain a task's css_set with extra access conditions
 430 * @task: the task to obtain css_set for
 431 * @__c: extra condition expression to be passed to rcu_dereference_check()
 432 *
 433 * A task's css_set is RCU protected, initialized and exited while holding
 434 * task_lock(), and can only be modified while holding both cgroup_mutex
 435 * and task_lock() while the task is alive.  This macro verifies that the
 436 * caller is inside proper critical section and returns @task's css_set.
 437 *
 438 * The caller can also specify additional allowed conditions via @__c, such
 439 * as locks used during the cgroup_subsys::attach() methods.
 440 */
 441#ifdef CONFIG_PROVE_RCU
 442extern struct mutex cgroup_mutex;
 443extern spinlock_t css_set_lock;
 444#define task_css_set_check(task, __c)                                   \
 445        rcu_dereference_check((task)->cgroups,                          \
 446                lockdep_is_held(&cgroup_mutex) ||                       \
 447                lockdep_is_held(&css_set_lock) ||                       \
 448                ((task)->flags & PF_EXITING) || (__c))
 449#else
 450#define task_css_set_check(task, __c)                                   \
 451        rcu_dereference((task)->cgroups)
 452#endif
 453
 454/**
 455 * task_css_check - obtain css for (task, subsys) w/ extra access conds
 456 * @task: the target task
 457 * @subsys_id: the target subsystem ID
 458 * @__c: extra condition expression to be passed to rcu_dereference_check()
 459 *
 460 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
 461 * synchronization rules are the same as task_css_set_check().
 462 */
 463#define task_css_check(task, subsys_id, __c)                            \
 464        task_css_set_check((task), (__c))->subsys[(subsys_id)]
 465
 466/**
 467 * task_css_set - obtain a task's css_set
 468 * @task: the task to obtain css_set for
 469 *
 470 * See task_css_set_check().
 471 */
 472static inline struct css_set *task_css_set(struct task_struct *task)
 473{
 474        return task_css_set_check(task, false);
 475}
 476
 477/**
 478 * task_css - obtain css for (task, subsys)
 479 * @task: the target task
 480 * @subsys_id: the target subsystem ID
 481 *
 482 * See task_css_check().
 483 */
 484static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
 485                                                   int subsys_id)
 486{
 487        return task_css_check(task, subsys_id, false);
 488}
 489
 490/**
 491 * task_get_css - find and get the css for (task, subsys)
 492 * @task: the target task
 493 * @subsys_id: the target subsystem ID
 494 *
 495 * Find the css for the (@task, @subsys_id) combination, increment a
 496 * reference on and return it.  This function is guaranteed to return a
 497 * valid css.  The returned css may already have been offlined.
 498 */
 499static inline struct cgroup_subsys_state *
 500task_get_css(struct task_struct *task, int subsys_id)
 501{
 502        struct cgroup_subsys_state *css;
 503
 504        rcu_read_lock();
 505        while (true) {
 506                css = task_css(task, subsys_id);
 507                /*
 508                 * Can't use css_tryget_online() here.  A task which has
 509                 * PF_EXITING set may stay associated with an offline css.
 510                 * If such task calls this function, css_tryget_online()
 511                 * will keep failing.
 512                 */
 513                if (likely(css_tryget(css)))
 514                        break;
 515                cpu_relax();
 516        }
 517        rcu_read_unlock();
 518        return css;
 519}
 520
 521/**
 522 * task_css_is_root - test whether a task belongs to the root css
 523 * @task: the target task
 524 * @subsys_id: the target subsystem ID
 525 *
 526 * Test whether @task belongs to the root css on the specified subsystem.
 527 * May be invoked in any context.
 528 */
 529static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
 530{
 531        return task_css_check(task, subsys_id, true) ==
 532                init_css_set.subsys[subsys_id];
 533}
 534
 535static inline struct cgroup *task_cgroup(struct task_struct *task,
 536                                         int subsys_id)
 537{
 538        return task_css(task, subsys_id)->cgroup;
 539}
 540
 541static inline struct cgroup *task_dfl_cgroup(struct task_struct *task)
 542{
 543        return task_css_set(task)->dfl_cgrp;
 544}
 545
 546static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
 547{
 548        struct cgroup_subsys_state *parent_css = cgrp->self.parent;
 549
 550        if (parent_css)
 551                return container_of(parent_css, struct cgroup, self);
 552        return NULL;
 553}
 554
 555/**
 556 * cgroup_is_descendant - test ancestry
 557 * @cgrp: the cgroup to be tested
 558 * @ancestor: possible ancestor of @cgrp
 559 *
 560 * Test whether @cgrp is a descendant of @ancestor.  It also returns %true
 561 * if @cgrp == @ancestor.  This function is safe to call as long as @cgrp
 562 * and @ancestor are accessible.
 563 */
 564static inline bool cgroup_is_descendant(struct cgroup *cgrp,
 565                                        struct cgroup *ancestor)
 566{
 567        if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
 568                return false;
 569        return cgrp->ancestor_ids[ancestor->level] == ancestor->id;
 570}
 571
 572/**
 573 * cgroup_ancestor - find ancestor of cgroup
 574 * @cgrp: cgroup to find ancestor of
 575 * @ancestor_level: level of ancestor to find starting from root
 576 *
 577 * Find ancestor of cgroup at specified level starting from root if it exists
 578 * and return pointer to it. Return NULL if @cgrp doesn't have ancestor at
 579 * @ancestor_level.
 580 *
 581 * This function is safe to call as long as @cgrp is accessible.
 582 */
 583static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp,
 584                                             int ancestor_level)
 585{
 586        if (cgrp->level < ancestor_level)
 587                return NULL;
 588        while (cgrp && cgrp->level > ancestor_level)
 589                cgrp = cgroup_parent(cgrp);
 590        return cgrp;
 591}
 592
 593/**
 594 * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry
 595 * @task: the task to be tested
 596 * @ancestor: possible ancestor of @task's cgroup
 597 *
 598 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
 599 * It follows all the same rules as cgroup_is_descendant, and only applies
 600 * to the default hierarchy.
 601 */
 602static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
 603                                               struct cgroup *ancestor)
 604{
 605        struct css_set *cset = task_css_set(task);
 606
 607        return cgroup_is_descendant(cset->dfl_cgrp, ancestor);
 608}
 609
 610/* no synchronization, the result can only be used as a hint */
 611static inline bool cgroup_is_populated(struct cgroup *cgrp)
 612{
 613        return cgrp->nr_populated_csets + cgrp->nr_populated_domain_children +
 614                cgrp->nr_populated_threaded_children;
 615}
 616
 617/* returns ino associated with a cgroup */
 618static inline ino_t cgroup_ino(struct cgroup *cgrp)
 619{
 620        return cgrp->kn->id.ino;
 621}
 622
 623/* cft/css accessors for cftype->write() operation */
 624static inline struct cftype *of_cft(struct kernfs_open_file *of)
 625{
 626        return of->kn->priv;
 627}
 628
 629struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
 630
 631/* cft/css accessors for cftype->seq_*() operations */
 632static inline struct cftype *seq_cft(struct seq_file *seq)
 633{
 634        return of_cft(seq->private);
 635}
 636
 637static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
 638{
 639        return of_css(seq->private);
 640}
 641
 642/*
 643 * Name / path handling functions.  All are thin wrappers around the kernfs
 644 * counterparts and can be called under any context.
 645 */
 646
 647static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
 648{
 649        return kernfs_name(cgrp->kn, buf, buflen);
 650}
 651
 652static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen)
 653{
 654        return kernfs_path(cgrp->kn, buf, buflen);
 655}
 656
 657static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
 658{
 659        pr_cont_kernfs_name(cgrp->kn);
 660}
 661
 662static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
 663{
 664        pr_cont_kernfs_path(cgrp->kn);
 665}
 666
 667static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
 668{
 669        return &cgrp->psi;
 670}
 671
 672static inline void cgroup_init_kthreadd(void)
 673{
 674        /*
 675         * kthreadd is inherited by all kthreads, keep it in the root so
 676         * that the new kthreads are guaranteed to stay in the root until
 677         * initialization is finished.
 678         */
 679        current->no_cgroup_migration = 1;
 680}
 681
 682static inline void cgroup_kthread_ready(void)
 683{
 684        /*
 685         * This kthread finished initialization.  The creator should have
 686         * set PF_NO_SETAFFINITY if this kthread should stay in the root.
 687         */
 688        current->no_cgroup_migration = 0;
 689}
 690
 691static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp)
 692{
 693        return &cgrp->kn->id;
 694}
 695
 696void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
 697                                        char *buf, size_t buflen);
 698#else /* !CONFIG_CGROUPS */
 699
 700struct cgroup_subsys_state;
 701struct cgroup;
 702
 703static inline void css_get(struct cgroup_subsys_state *css) {}
 704static inline void css_put(struct cgroup_subsys_state *css) {}
 705static inline int cgroup_attach_task_all(struct task_struct *from,
 706                                         struct task_struct *t) { return 0; }
 707static inline int cgroupstats_build(struct cgroupstats *stats,
 708                                    struct dentry *dentry) { return -EINVAL; }
 709
 710static inline void cgroup_fork(struct task_struct *p) {}
 711static inline int cgroup_can_fork(struct task_struct *p) { return 0; }
 712static inline void cgroup_cancel_fork(struct task_struct *p) {}
 713static inline void cgroup_post_fork(struct task_struct *p) {}
 714static inline void cgroup_exit(struct task_struct *p) {}
 715static inline void cgroup_release(struct task_struct *p) {}
 716static inline void cgroup_free(struct task_struct *p) {}
 717
 718static inline int cgroup_init_early(void) { return 0; }
 719static inline int cgroup_init(void) { return 0; }
 720static inline void cgroup_init_kthreadd(void) {}
 721static inline void cgroup_kthread_ready(void) {}
 722static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp)
 723{
 724        return NULL;
 725}
 726
 727static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
 728{
 729        return NULL;
 730}
 731
 732static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
 733{
 734        return NULL;
 735}
 736
 737static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
 738                                               struct cgroup *ancestor)
 739{
 740        return true;
 741}
 742
 743static inline void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
 744        char *buf, size_t buflen) {}
 745#endif /* !CONFIG_CGROUPS */
 746
 747#ifdef CONFIG_CGROUPS
 748/*
 749 * cgroup scalable recursive statistics.
 750 */
 751void cgroup_rstat_updated(struct cgroup *cgrp, int cpu);
 752void cgroup_rstat_flush(struct cgroup *cgrp);
 753void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp);
 754void cgroup_rstat_flush_hold(struct cgroup *cgrp);
 755void cgroup_rstat_flush_release(void);
 756
 757/*
 758 * Basic resource stats.
 759 */
 760#ifdef CONFIG_CGROUP_CPUACCT
 761void cpuacct_charge(struct task_struct *tsk, u64 cputime);
 762void cpuacct_account_field(struct task_struct *tsk, int index, u64 val);
 763#else
 764static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
 765static inline void cpuacct_account_field(struct task_struct *tsk, int index,
 766                                         u64 val) {}
 767#endif
 768
 769void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec);
 770void __cgroup_account_cputime_field(struct cgroup *cgrp,
 771                                    enum cpu_usage_stat index, u64 delta_exec);
 772
 773static inline void cgroup_account_cputime(struct task_struct *task,
 774                                          u64 delta_exec)
 775{
 776        struct cgroup *cgrp;
 777
 778        cpuacct_charge(task, delta_exec);
 779
 780        rcu_read_lock();
 781        cgrp = task_dfl_cgroup(task);
 782        if (cgroup_parent(cgrp))
 783                __cgroup_account_cputime(cgrp, delta_exec);
 784        rcu_read_unlock();
 785}
 786
 787static inline void cgroup_account_cputime_field(struct task_struct *task,
 788                                                enum cpu_usage_stat index,
 789                                                u64 delta_exec)
 790{
 791        struct cgroup *cgrp;
 792
 793        cpuacct_account_field(task, index, delta_exec);
 794
 795        rcu_read_lock();
 796        cgrp = task_dfl_cgroup(task);
 797        if (cgroup_parent(cgrp))
 798                __cgroup_account_cputime_field(cgrp, index, delta_exec);
 799        rcu_read_unlock();
 800}
 801
 802#else   /* CONFIG_CGROUPS */
 803
 804static inline void cgroup_account_cputime(struct task_struct *task,
 805                                          u64 delta_exec) {}
 806static inline void cgroup_account_cputime_field(struct task_struct *task,
 807                                                enum cpu_usage_stat index,
 808                                                u64 delta_exec) {}
 809
 810#endif  /* CONFIG_CGROUPS */
 811
 812/*
 813 * sock->sk_cgrp_data handling.  For more info, see sock_cgroup_data
 814 * definition in cgroup-defs.h.
 815 */
 816#ifdef CONFIG_SOCK_CGROUP_DATA
 817
 818#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
 819extern spinlock_t cgroup_sk_update_lock;
 820#endif
 821
 822void cgroup_sk_alloc_disable(void);
 823void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
 824void cgroup_sk_free(struct sock_cgroup_data *skcd);
 825
 826static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
 827{
 828#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
 829        unsigned long v;
 830
 831        /*
 832         * @skcd->val is 64bit but the following is safe on 32bit too as we
 833         * just need the lower ulong to be written and read atomically.
 834         */
 835        v = READ_ONCE(skcd->val);
 836
 837        if (v & 1)
 838                return &cgrp_dfl_root.cgrp;
 839
 840        return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp;
 841#else
 842        return (struct cgroup *)(unsigned long)skcd->val;
 843#endif
 844}
 845
 846#else   /* CONFIG_CGROUP_DATA */
 847
 848static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
 849static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
 850
 851#endif  /* CONFIG_CGROUP_DATA */
 852
 853struct cgroup_namespace {
 854        refcount_t              count;
 855        struct ns_common        ns;
 856        struct user_namespace   *user_ns;
 857        struct ucounts          *ucounts;
 858        struct css_set          *root_cset;
 859};
 860
 861extern struct cgroup_namespace init_cgroup_ns;
 862
 863#ifdef CONFIG_CGROUPS
 864
 865void free_cgroup_ns(struct cgroup_namespace *ns);
 866
 867struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
 868                                        struct user_namespace *user_ns,
 869                                        struct cgroup_namespace *old_ns);
 870
 871int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
 872                   struct cgroup_namespace *ns);
 873
 874#else /* !CONFIG_CGROUPS */
 875
 876static inline void free_cgroup_ns(struct cgroup_namespace *ns) { }
 877static inline struct cgroup_namespace *
 878copy_cgroup_ns(unsigned long flags, struct user_namespace *user_ns,
 879               struct cgroup_namespace *old_ns)
 880{
 881        return old_ns;
 882}
 883
 884#endif /* !CONFIG_CGROUPS */
 885
 886static inline void get_cgroup_ns(struct cgroup_namespace *ns)
 887{
 888        if (ns)
 889                refcount_inc(&ns->count);
 890}
 891
 892static inline void put_cgroup_ns(struct cgroup_namespace *ns)
 893{
 894        if (ns && refcount_dec_and_test(&ns->count))
 895                free_cgroup_ns(ns);
 896}
 897
 898#ifdef CONFIG_CGROUPS
 899
 900void cgroup_enter_frozen(void);
 901void cgroup_leave_frozen(bool always_leave);
 902void cgroup_update_frozen(struct cgroup *cgrp);
 903void cgroup_freeze(struct cgroup *cgrp, bool freeze);
 904void cgroup_freezer_migrate_task(struct task_struct *task, struct cgroup *src,
 905                                 struct cgroup *dst);
 906
 907static inline bool cgroup_task_freeze(struct task_struct *task)
 908{
 909        bool ret;
 910
 911        if (task->flags & PF_KTHREAD)
 912                return false;
 913
 914        rcu_read_lock();
 915        ret = test_bit(CGRP_FREEZE, &task_dfl_cgroup(task)->flags);
 916        rcu_read_unlock();
 917
 918        return ret;
 919}
 920
 921static inline bool cgroup_task_frozen(struct task_struct *task)
 922{
 923        return task->frozen;
 924}
 925
 926#else /* !CONFIG_CGROUPS */
 927
 928static inline void cgroup_enter_frozen(void) { }
 929static inline void cgroup_leave_frozen(bool always_leave) { }
 930static inline bool cgroup_task_freeze(struct task_struct *task)
 931{
 932        return false;
 933}
 934static inline bool cgroup_task_frozen(struct task_struct *task)
 935{
 936        return false;
 937}
 938
 939#endif /* !CONFIG_CGROUPS */
 940
 941#ifdef CONFIG_CGROUP_BPF
 942static inline void cgroup_bpf_get(struct cgroup *cgrp)
 943{
 944        percpu_ref_get(&cgrp->bpf.refcnt);
 945}
 946
 947static inline void cgroup_bpf_put(struct cgroup *cgrp)
 948{
 949        percpu_ref_put(&cgrp->bpf.refcnt);
 950}
 951
 952#else /* CONFIG_CGROUP_BPF */
 953
 954static inline void cgroup_bpf_get(struct cgroup *cgrp) {}
 955static inline void cgroup_bpf_put(struct cgroup *cgrp) {}
 956
 957#endif /* CONFIG_CGROUP_BPF */
 958
 959#endif /* _LINUX_CGROUP_H */
 960