include/linux/cgroup.h

   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/rculist.h>
  15 #include <linux/cgroupstats.h>
  16 #include <linux/fs.h>
  17 #include <linux/seq_file.h>
  18 #include <linux/kernfs.h>
  19 #include <linux/jump_label.h>
  20
  21 #include <linux/cgroup-defs.h>
  22
  23 #ifdef CONFIG_CGROUPS
  24
  25 /*
  26  * All weight knobs on the default hierarhcy should use the following min,
  27  * default and max values.  The default value is the logarithmic center of
  28  * MIN and MAX and allows 100x to be expressed in both directions.
  29  */
  30 #define CGROUP_WEIGHT_MIN               1
  31 #define CGROUP_WEIGHT_DFL               100
  32 #define CGROUP_WEIGHT_MAX               10000
  33
  34 /* a css_task_iter should be treated as an opaque object */
  35 struct css_task_iter {
  36         struct cgroup_subsys            *ss;
  37
  38         struct list_head                *cset_pos;
  39         struct list_head                *cset_head;
  40
  41         struct list_head                *task_pos;
  42         struct list_head                *tasks_head;
  43         struct list_head                *mg_tasks_head;
  44
  45         struct css_set                  *cur_cset;
  46         struct task_struct              *cur_task;
  47         struct list_head                iters_node;     /* css_set->task_iters */
  48 };
  49
  50 extern struct cgroup_root cgrp_dfl_root;
  51 extern struct css_set init_css_set;
  52
  53 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
  54 #include <linux/cgroup_subsys.h>
  55 #undef SUBSYS
  56
  57 #define SUBSYS(_x)                                                              \
  58         extern struct static_key_true _x ## _cgrp_subsys_enabled_key;           \
  59         extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
  60 #include <linux/cgroup_subsys.h>
  61 #undef SUBSYS
  62
  63 /**
  64  * cgroup_subsys_enabled - fast test on whether a subsys is enabled
  65  * @ss: subsystem in question
  66  */
  67 #define cgroup_subsys_enabled(ss)                                               \
  68         static_branch_likely(&ss ## _enabled_key)
  69
  70 /**
  71  * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
  72  * @ss: subsystem in question
  73  */
  74 #define cgroup_subsys_on_dfl(ss)                                                \
  75         static_branch_likely(&ss ## _on_dfl_key)
  76
  77 bool css_has_online_children(struct cgroup_subsys_state *css);
  78 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
  79 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
  80                                              struct cgroup_subsys *ss);
  81 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
  82                                                        struct cgroup_subsys *ss);
  83
  84 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
  85 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
  86 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
  87
  88 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
  89 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
  90 int cgroup_rm_cftypes(struct cftype *cfts);
  91
  92 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
  93 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
  94 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
  95                      struct pid *pid, struct task_struct *tsk);
  96
  97 void cgroup_fork(struct task_struct *p);
  98 extern int cgroup_can_fork(struct task_struct *p,
  99                            void *ss_priv[CGROUP_CANFORK_COUNT]);
 100 extern void cgroup_cancel_fork(struct task_struct *p,
 101                                void *ss_priv[CGROUP_CANFORK_COUNT]);
 102 extern void cgroup_post_fork(struct task_struct *p,
 103                              void *old_ss_priv[CGROUP_CANFORK_COUNT]);
 104 void cgroup_exit(struct task_struct *p);
 105 void cgroup_free(struct task_struct *p);
 106
 107 int cgroup_init_early(void);
 108 int cgroup_init(void);
 109
 110 /*
 111  * Iteration helpers and macros.
 112  */
 113
 114 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
 115                                            struct cgroup_subsys_state *parent);
 116 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
 117                                                     struct cgroup_subsys_state *css);
 118 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
 119 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
 120                                                      struct cgroup_subsys_state *css);
 121
 122 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
 123 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
 124
 125 void css_task_iter_start(struct cgroup_subsys_state *css,
 126                          struct css_task_iter *it);
 127 struct task_struct *css_task_iter_next(struct css_task_iter *it);
 128 void css_task_iter_end(struct css_task_iter *it);
 129
 130 /**
 131  * css_for_each_child - iterate through children of a css
 132  * @pos: the css * to use as the loop cursor
 133  * @parent: css whose children to walk
 134  *
 135  * Walk @parent's children.  Must be called under rcu_read_lock().
 136  *
 137  * If a subsystem synchronizes ->css_online() and the start of iteration, a
 138  * css which finished ->css_online() is guaranteed to be visible in the
 139  * future iterations and will stay visible until the last reference is put.
 140  * A css which hasn't finished ->css_online() or already finished
 141  * ->css_offline() may show up during traversal.  It's each subsystem's
 142  * responsibility to synchronize against on/offlining.
 143  *
 144  * It is allowed to temporarily drop RCU read lock during iteration.  The
 145  * caller is responsible for ensuring that @pos remains accessible until
 146  * the start of the next iteration by, for example, bumping the css refcnt.
 147  */
 148 #define css_for_each_child(pos, parent)                                 \
 149         for ((pos) = css_next_child(NULL, (parent)); (pos);             \
 150              (pos) = css_next_child((pos), (parent)))
 151
 152 /**
 153  * css_for_each_descendant_pre - pre-order walk of a css's descendants
 154  * @pos: the css * to use as the loop cursor
 155  * @root: css whose descendants to walk
 156  *
 157  * Walk @root's descendants.  @root is included in the iteration and the
 158  * first node to be visited.  Must be called under rcu_read_lock().
 159  *
 160  * If a subsystem synchronizes ->css_online() and the start of iteration, a
 161  * css which finished ->css_online() is guaranteed to be visible in the
 162  * future iterations and will stay visible until the last reference is put.
 163  * A css which hasn't finished ->css_online() or already finished
 164  * ->css_offline() may show up during traversal.  It's each subsystem's
 165  * responsibility to synchronize against on/offlining.
 166  *
 167  * For example, the following guarantees that a descendant can't escape
 168  * state updates of its ancestors.
 169  *
 170  * my_online(@css)
 171  * {
 172  *      Lock @css's parent and @css;
 173  *      Inherit state from the parent;
 174  *      Unlock both.
 175  * }
 176  *
 177  * my_update_state(@css)
 178  * {
 179  *      css_for_each_descendant_pre(@pos, @css) {
 180  *              Lock @pos;
 181  *              if (@pos == @css)
 182  *                      Update @css's state;
 183  *              else
 184  *                      Verify @pos is alive and inherit state from its parent;
 185  *              Unlock @pos;
 186  *      }
 187  * }
 188  *
 189  * As long as the inheriting step, including checking the parent state, is
 190  * enclosed inside @pos locking, double-locking the parent isn't necessary
 191  * while inheriting.  The state update to the parent is guaranteed to be
 192  * visible by walking order and, as long as inheriting operations to the
 193  * same @pos are atomic to each other, multiple updates racing each other
 194  * still result in the correct state.  It's guaranateed that at least one
 195  * inheritance happens for any css after the latest update to its parent.
 196  *
 197  * If checking parent's state requires locking the parent, each inheriting
 198  * iteration should lock and unlock both @pos->parent and @pos.
 199  *
 200  * Alternatively, a subsystem may choose to use a single global lock to
 201  * synchronize ->css_online() and ->css_offline() against tree-walking
 202  * operations.
 203  *
 204  * It is allowed to temporarily drop RCU read lock during iteration.  The
 205  * caller is responsible for ensuring that @pos remains accessible until
 206  * the start of the next iteration by, for example, bumping the css refcnt.
 207  */
 208 #define css_for_each_descendant_pre(pos, css)                           \
 209         for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);       \
 210              (pos) = css_next_descendant_pre((pos), (css)))
 211
 212 /**
 213  * css_for_each_descendant_post - post-order walk of a css's descendants
 214  * @pos: the css * to use as the loop cursor
 215  * @css: css whose descendants to walk
 216  *
 217  * Similar to css_for_each_descendant_pre() but performs post-order
 218  * traversal instead.  @root is included in the iteration and the last
 219  * node to be visited.
 220  *
 221  * If a subsystem synchronizes ->css_online() and the start of iteration, a
 222  * css which finished ->css_online() is guaranteed to be visible in the
 223  * future iterations and will stay visible until the last reference is put.
 224  * A css which hasn't finished ->css_online() or already finished
 225  * ->css_offline() may show up during traversal.  It's each subsystem's
 226  * responsibility to synchronize against on/offlining.
 227  *
 228  * Note that the walk visibility guarantee example described in pre-order
 229  * walk doesn't apply the same to post-order walks.
 230  */
 231 #define css_for_each_descendant_post(pos, css)                          \
 232         for ((pos) = css_next_descendant_post(NULL, (css)); (pos);      \
 233              (pos) = css_next_descendant_post((pos), (css)))
 234
 235 /**
 236  * cgroup_taskset_for_each - iterate cgroup_taskset
 237  * @task: the loop cursor
 238  * @tset: taskset to iterate
 239  *
 240  * @tset may contain multiple tasks and they may belong to multiple
 241  * processes.  When there are multiple tasks in @tset, if a task of a
 242  * process is in @tset, all tasks of the process are in @tset.  Also, all
 243  * are guaranteed to share the same source and destination csses.
 244  *
 245  * Iteration is not in any specific order.
 246  */
 247 #define cgroup_taskset_for_each(task, tset)                             \
 248         for ((task) = cgroup_taskset_first((tset)); (task);             \
 249              (task) = cgroup_taskset_next((tset)))
 250
 251 /**
 252  * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
 253  * @leader: the loop cursor
 254  * @tset: takset to iterate
 255  *
 256  * Iterate threadgroup leaders of @tset.  For single-task migrations, @tset
 257  * may not contain any.
 258  */
 259 #define cgroup_taskset_for_each_leader(leader, tset)                    \
 260         for ((leader) = cgroup_taskset_first((tset)); (leader);         \
 261              (leader) = cgroup_taskset_next((tset)))                    \
 262                 if ((leader) != (leader)->group_leader)                 \
 263                         ;                                               \
 264                 else
 265
 266 /*
 267  * Inline functions.
 268  */
 269
 270 /**
 271  * css_get - obtain a reference on the specified css
 272  * @css: target css
 273  *
 274  * The caller must already have a reference.
 275  */
 276 static inline void css_get(struct cgroup_subsys_state *css)
 277 {
 278         if (!(css->flags & CSS_NO_REF))
 279                 percpu_ref_get(&css->refcnt);
 280 }
 281
 282 /**
 283  * css_get_many - obtain references on the specified css
 284  * @css: target css
 285  * @n: number of references to get
 286  *
 287  * The caller must already have a reference.
 288  */
 289 static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
 290 {
 291         if (!(css->flags & CSS_NO_REF))
 292                 percpu_ref_get_many(&css->refcnt, n);
 293 }
 294
 295 /**
 296  * css_tryget - try to obtain a reference on the specified css
 297  * @css: target css
 298  *
 299  * Obtain a reference on @css unless it already has reached zero and is
 300  * being released.  This function doesn't care whether @css is on or
 301  * offline.  The caller naturally needs to ensure that @css is accessible
 302  * but doesn't have to be holding a reference on it - IOW, RCU protected
 303  * access is good enough for this function.  Returns %true if a reference
 304  * count was successfully obtained; %false otherwise.
 305  */
 306 static inline bool css_tryget(struct cgroup_subsys_state *css)
 307 {
 308         if (!(css->flags & CSS_NO_REF))
 309                 return percpu_ref_tryget(&css->refcnt);
 310         return true;
 311 }
 312
 313 /**
 314  * css_tryget_online - try to obtain a reference on the specified css if online
 315  * @css: target css
 316  *
 317  * Obtain a reference on @css if it's online.  The caller naturally needs
 318  * to ensure that @css is accessible but doesn't have to be holding a
 319  * reference on it - IOW, RCU protected access is good enough for this
 320  * function.  Returns %true if a reference count was successfully obtained;
 321  * %false otherwise.
 322  */
 323 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
 324 {
 325         if (!(css->flags & CSS_NO_REF))
 326                 return percpu_ref_tryget_live(&css->refcnt);
 327         return true;
 328 }
 329
 330 /**
 331  * css_put - put a css reference
 332  * @css: target css
 333  *
 334  * Put a reference obtained via css_get() and css_tryget_online().
 335  */
 336 static inline void css_put(struct cgroup_subsys_state *css)
 337 {
 338         if (!(css->flags & CSS_NO_REF))
 339                 percpu_ref_put(&css->refcnt);
 340 }
 341
 342 /**
 343  * css_put_many - put css references
 344  * @css: target css
 345  * @n: number of references to put
 346  *
 347  * Put references obtained via css_get() and css_tryget_online().
 348  */
 349 static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
 350 {
 351         if (!(css->flags & CSS_NO_REF))
 352                 percpu_ref_put_many(&css->refcnt, n);
 353 }
 354
 355 /**
 356  * task_css_set_check - obtain a task's css_set with extra access conditions
 357  * @task: the task to obtain css_set for
 358  * @__c: extra condition expression to be passed to rcu_dereference_check()
 359  *
 360  * A task's css_set is RCU protected, initialized and exited while holding
 361  * task_lock(), and can only be modified while holding both cgroup_mutex
 362  * and task_lock() while the task is alive.  This macro verifies that the
 363  * caller is inside proper critical section and returns @task's css_set.
 364  *
 365  * The caller can also specify additional allowed conditions via @__c, such
 366  * as locks used during the cgroup_subsys::attach() methods.
 367  */
 368 #ifdef CONFIG_PROVE_RCU
 369 extern struct mutex cgroup_mutex;
 370 extern spinlock_t css_set_lock;
 371 #define task_css_set_check(task, __c)                                   \
 372         rcu_dereference_check((task)->cgroups,                          \
 373                 lockdep_is_held(&cgroup_mutex) ||                       \
 374                 lockdep_is_held(&css_set_lock) ||                       \
 375                 ((task)->flags & PF_EXITING) || (__c))
 376 #else
 377 #define task_css_set_check(task, __c)                                   \
 378         rcu_dereference((task)->cgroups)
 379 #endif
 380
 381 /**
 382  * task_css_check - obtain css for (task, subsys) w/ extra access conds
 383  * @task: the target task
 384  * @subsys_id: the target subsystem ID
 385  * @__c: extra condition expression to be passed to rcu_dereference_check()
 386  *
 387  * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
 388  * synchronization rules are the same as task_css_set_check().
 389  */
 390 #define task_css_check(task, subsys_id, __c)                            \
 391         task_css_set_check((task), (__c))->subsys[(subsys_id)]
 392
 393 /**
 394  * task_css_set - obtain a task's css_set
 395  * @task: the task to obtain css_set for
 396  *
 397  * See task_css_set_check().
 398  */
 399 static inline struct css_set *task_css_set(struct task_struct *task)
 400 {
 401         return task_css_set_check(task, false);
 402 }
 403
 404 /**
 405  * task_css - obtain css for (task, subsys)
 406  * @task: the target task
 407  * @subsys_id: the target subsystem ID
 408  *
 409  * See task_css_check().
 410  */
 411 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
 412                                                    int subsys_id)
 413 {
 414         return task_css_check(task, subsys_id, false);
 415 }
 416
 417 /**
 418  * task_get_css - find and get the css for (task, subsys)
 419  * @task: the target task
 420  * @subsys_id: the target subsystem ID
 421  *
 422  * Find the css for the (@task, @subsys_id) combination, increment a
 423  * reference on and return it.  This function is guaranteed to return a
 424  * valid css.
 425  */
 426 static inline struct cgroup_subsys_state *
 427 task_get_css(struct task_struct *task, int subsys_id)
 428 {
 429         struct cgroup_subsys_state *css;
 430
 431         rcu_read_lock();
 432         while (true) {
 433                 css = task_css(task, subsys_id);
 434                 if (likely(css_tryget_online(css)))
 435                         break;
 436                 cpu_relax();
 437         }
 438         rcu_read_unlock();
 439         return css;
 440 }
 441
 442 /**
 443  * task_css_is_root - test whether a task belongs to the root css
 444  * @task: the target task
 445  * @subsys_id: the target subsystem ID
 446  *
 447  * Test whether @task belongs to the root css on the specified subsystem.
 448  * May be invoked in any context.
 449  */
 450 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
 451 {
 452         return task_css_check(task, subsys_id, true) ==
 453                 init_css_set.subsys[subsys_id];
 454 }
 455
 456 static inline struct cgroup *task_cgroup(struct task_struct *task,
 457                                          int subsys_id)
 458 {
 459         return task_css(task, subsys_id)->cgroup;
 460 }
 461
 462 /* no synchronization, the result can only be used as a hint */
 463 static inline bool cgroup_is_populated(struct cgroup *cgrp)
 464 {
 465         return cgrp->populated_cnt;
 466 }
 467
 468 /* returns ino associated with a cgroup */
 469 static inline ino_t cgroup_ino(struct cgroup *cgrp)
 470 {
 471         return cgrp->kn->ino;
 472 }
 473
 474 /* cft/css accessors for cftype->write() operation */
 475 static inline struct cftype *of_cft(struct kernfs_open_file *of)
 476 {
 477         return of->kn->priv;
 478 }
 479
 480 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
 481
 482 /* cft/css accessors for cftype->seq_*() operations */
 483 static inline struct cftype *seq_cft(struct seq_file *seq)
 484 {
 485         return of_cft(seq->private);
 486 }
 487
 488 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
 489 {
 490         return of_css(seq->private);
 491 }
 492
 493 /*
 494  * Name / path handling functions.  All are thin wrappers around the kernfs
 495  * counterparts and can be called under any context.
 496  */
 497
 498 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
 499 {
 500         return kernfs_name(cgrp->kn, buf, buflen);
 501 }
 502
 503 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
 504                                               size_t buflen)
 505 {
 506         return kernfs_path(cgrp->kn, buf, buflen);
 507 }
 508
 509 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
 510 {
 511         pr_cont_kernfs_name(cgrp->kn);
 512 }
 513
 514 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
 515 {
 516         pr_cont_kernfs_path(cgrp->kn);
 517 }
 518
 519 /**
 520  * cgroup_file_notify - generate a file modified event for a cgroup_file
 521  * @cfile: target cgroup_file
 522  *
 523  * @cfile must have been obtained by setting cftype->file_offset.
 524  */
 525 static inline void cgroup_file_notify(struct cgroup_file *cfile)
 526 {
 527         /* might not have been created due to one of the CFTYPE selector flags */
 528         if (cfile->kn)
 529                 kernfs_notify(cfile->kn);
 530 }
 531
 532 #else /* !CONFIG_CGROUPS */
 533
 534 struct cgroup_subsys_state;
 535
 536 static inline void css_put(struct cgroup_subsys_state *css) {}
 537 static inline int cgroup_attach_task_all(struct task_struct *from,
 538                                          struct task_struct *t) { return 0; }
 539 static inline int cgroupstats_build(struct cgroupstats *stats,
 540                                     struct dentry *dentry) { return -EINVAL; }
 541
 542 static inline void cgroup_fork(struct task_struct *p) {}
 543 static inline int cgroup_can_fork(struct task_struct *p,
 544                                   void *ss_priv[CGROUP_CANFORK_COUNT])
 545 { return 0; }
 546 static inline void cgroup_cancel_fork(struct task_struct *p,
 547                                       void *ss_priv[CGROUP_CANFORK_COUNT]) {}
 548 static inline void cgroup_post_fork(struct task_struct *p,
 549                                     void *ss_priv[CGROUP_CANFORK_COUNT]) {}
 550 static inline void cgroup_exit(struct task_struct *p) {}
 551 static inline void cgroup_free(struct task_struct *p) {}
 552
 553 static inline int cgroup_init_early(void) { return 0; }
 554 static inline int cgroup_init(void) { return 0; }
 555
 556 #endif /* !CONFIG_CGROUPS */
 557
 558 #endif /* _LINUX_CGROUP_H */