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 void cgroup_file_notify(struct cgroup_file *cfile);
  92
  93 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
  94 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
  95 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
  96                      struct pid *pid, struct task_struct *tsk);
  97
  98 void cgroup_fork(struct task_struct *p);
  99 extern int cgroup_can_fork(struct task_struct *p,
 100                            void *ss_priv[CGROUP_CANFORK_COUNT]);
 101 extern void cgroup_cancel_fork(struct task_struct *p,
 102                                void *ss_priv[CGROUP_CANFORK_COUNT]);
 103 extern void cgroup_post_fork(struct task_struct *p,
 104                              void *old_ss_priv[CGROUP_CANFORK_COUNT]);
 105 void cgroup_exit(struct task_struct *p);
 106 void cgroup_free(struct task_struct *p);
 107
 108 int cgroup_init_early(void);
 109 int cgroup_init(void);
 110
 111 /*
 112  * Iteration helpers and macros.
 113  */
 114
 115 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
 116                                            struct cgroup_subsys_state *parent);
 117 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
 118                                                     struct cgroup_subsys_state *css);
 119 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
 120 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
 121                                                      struct cgroup_subsys_state *css);
 122
 123 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
 124                                          struct cgroup_subsys_state **dst_cssp);
 125 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
 126                                         struct cgroup_subsys_state **dst_cssp);
 127
 128 void css_task_iter_start(struct cgroup_subsys_state *css,
 129                          struct css_task_iter *it);
 130 struct task_struct *css_task_iter_next(struct css_task_iter *it);
 131 void css_task_iter_end(struct css_task_iter *it);
 132
 133 /**
 134  * css_for_each_child - iterate through children of a css
 135  * @pos: the css * to use as the loop cursor
 136  * @parent: css whose children to walk
 137  *
 138  * Walk @parent's children.  Must be called under rcu_read_lock().
 139  *
 140  * If a subsystem synchronizes ->css_online() and the start of iteration, a
 141  * css which finished ->css_online() is guaranteed to be visible in the
 142  * future iterations and will stay visible until the last reference is put.
 143  * A css which hasn't finished ->css_online() or already finished
 144  * ->css_offline() may show up during traversal.  It's each subsystem's
 145  * responsibility to synchronize against on/offlining.
 146  *
 147  * It is allowed to temporarily drop RCU read lock during iteration.  The
 148  * caller is responsible for ensuring that @pos remains accessible until
 149  * the start of the next iteration by, for example, bumping the css refcnt.
 150  */
 151 #define css_for_each_child(pos, parent)                                 \
 152         for ((pos) = css_next_child(NULL, (parent)); (pos);             \
 153              (pos) = css_next_child((pos), (parent)))
 154
 155 /**
 156  * css_for_each_descendant_pre - pre-order walk of a css's descendants
 157  * @pos: the css * to use as the loop cursor
 158  * @root: css whose descendants to walk
 159  *
 160  * Walk @root's descendants.  @root is included in the iteration and the
 161  * first node to be visited.  Must be called under rcu_read_lock().
 162  *
 163  * If a subsystem synchronizes ->css_online() and the start of iteration, a
 164  * css which finished ->css_online() is guaranteed to be visible in the
 165  * future iterations and will stay visible until the last reference is put.
 166  * A css which hasn't finished ->css_online() or already finished
 167  * ->css_offline() may show up during traversal.  It's each subsystem's
 168  * responsibility to synchronize against on/offlining.
 169  *
 170  * For example, the following guarantees that a descendant can't escape
 171  * state updates of its ancestors.
 172  *
 173  * my_online(@css)
 174  * {
 175  *      Lock @css's parent and @css;
 176  *      Inherit state from the parent;
 177  *      Unlock both.
 178  * }
 179  *
 180  * my_update_state(@css)
 181  * {
 182  *      css_for_each_descendant_pre(@pos, @css) {
 183  *              Lock @pos;
 184  *              if (@pos == @css)
 185  *                      Update @css's state;
 186  *              else
 187  *                      Verify @pos is alive and inherit state from its parent;
 188  *              Unlock @pos;
 189  *      }
 190  * }
 191  *
 192  * As long as the inheriting step, including checking the parent state, is
 193  * enclosed inside @pos locking, double-locking the parent isn't necessary
 194  * while inheriting.  The state update to the parent is guaranteed to be
 195  * visible by walking order and, as long as inheriting operations to the
 196  * same @pos are atomic to each other, multiple updates racing each other
 197  * still result in the correct state.  It's guaranateed that at least one
 198  * inheritance happens for any css after the latest update to its parent.
 199  *
 200  * If checking parent's state requires locking the parent, each inheriting
 201  * iteration should lock and unlock both @pos->parent and @pos.
 202  *
 203  * Alternatively, a subsystem may choose to use a single global lock to
 204  * synchronize ->css_online() and ->css_offline() against tree-walking
 205  * operations.
 206  *
 207  * It is allowed to temporarily drop RCU read lock during iteration.  The
 208  * caller is responsible for ensuring that @pos remains accessible until
 209  * the start of the next iteration by, for example, bumping the css refcnt.
 210  */
 211 #define css_for_each_descendant_pre(pos, css)                           \
 212         for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);       \
 213              (pos) = css_next_descendant_pre((pos), (css)))
 214
 215 /**
 216  * css_for_each_descendant_post - post-order walk of a css's descendants
 217  * @pos: the css * to use as the loop cursor
 218  * @css: css whose descendants to walk
 219  *
 220  * Similar to css_for_each_descendant_pre() but performs post-order
 221  * traversal instead.  @root is included in the iteration and the last
 222  * node to be visited.
 223  *
 224  * If a subsystem synchronizes ->css_online() and the start of iteration, a
 225  * css which finished ->css_online() is guaranteed to be visible in the
 226  * future iterations and will stay visible until the last reference is put.
 227  * A css which hasn't finished ->css_online() or already finished
 228  * ->css_offline() may show up during traversal.  It's each subsystem's
 229  * responsibility to synchronize against on/offlining.
 230  *
 231  * Note that the walk visibility guarantee example described in pre-order
 232  * walk doesn't apply the same to post-order walks.
 233  */
 234 #define css_for_each_descendant_post(pos, css)                          \
 235         for ((pos) = css_next_descendant_post(NULL, (css)); (pos);      \
 236              (pos) = css_next_descendant_post((pos), (css)))
 237
 238 /**
 239  * cgroup_taskset_for_each - iterate cgroup_taskset
 240  * @task: the loop cursor
 241  * @dst_css: the destination css
 242  * @tset: taskset to iterate
 243  *
 244  * @tset may contain multiple tasks and they may belong to multiple
 245  * processes.
 246  *
 247  * On the v2 hierarchy, there may be tasks from multiple processes and they
 248  * may not share the source or destination csses.
 249  *
 250  * On traditional hierarchies, when there are multiple tasks in @tset, if a
 251  * task of a process is in @tset, all tasks of the process are in @tset.
 252  * Also, all are guaranteed to share the same source and destination csses.
 253  *
 254  * Iteration is not in any specific order.
 255  */
 256 #define cgroup_taskset_for_each(task, dst_css, tset)                    \
 257         for ((task) = cgroup_taskset_first((tset), &(dst_css));         \
 258              (task);                                                    \
 259              (task) = cgroup_taskset_next((tset), &(dst_css)))
 260
 261 /**
 262  * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
 263  * @leader: the loop cursor
 264  * @dst_css: the destination css
 265  * @tset: takset to iterate
 266  *
 267  * Iterate threadgroup leaders of @tset.  For single-task migrations, @tset
 268  * may not contain any.
 269  */
 270 #define cgroup_taskset_for_each_leader(leader, dst_css, tset)           \
 271         for ((leader) = cgroup_taskset_first((tset), &(dst_css));       \
 272              (leader);                                                  \
 273              (leader) = cgroup_taskset_next((tset), &(dst_css)))        \
 274                 if ((leader) != (leader)->group_leader)                 \
 275                         ;                                               \
 276                 else
 277
 278 /*
 279  * Inline functions.
 280  */
 281
 282 /**
 283  * css_get - obtain a reference on the specified css
 284  * @css: target css
 285  *
 286  * The caller must already have a reference.
 287  */
 288 static inline void css_get(struct cgroup_subsys_state *css)
 289 {
 290         if (!(css->flags & CSS_NO_REF))
 291                 percpu_ref_get(&css->refcnt);
 292 }
 293
 294 /**
 295  * css_get_many - obtain references on the specified css
 296  * @css: target css
 297  * @n: number of references to get
 298  *
 299  * The caller must already have a reference.
 300  */
 301 static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
 302 {
 303         if (!(css->flags & CSS_NO_REF))
 304                 percpu_ref_get_many(&css->refcnt, n);
 305 }
 306
 307 /**
 308  * css_tryget - try to obtain a reference on the specified css
 309  * @css: target css
 310  *
 311  * Obtain a reference on @css unless it already has reached zero and is
 312  * being released.  This function doesn't care whether @css is on or
 313  * offline.  The caller naturally needs to ensure that @css is accessible
 314  * but doesn't have to be holding a reference on it - IOW, RCU protected
 315  * access is good enough for this function.  Returns %true if a reference
 316  * count was successfully obtained; %false otherwise.
 317  */
 318 static inline bool css_tryget(struct cgroup_subsys_state *css)
 319 {
 320         if (!(css->flags & CSS_NO_REF))
 321                 return percpu_ref_tryget(&css->refcnt);
 322         return true;
 323 }
 324
 325 /**
 326  * css_tryget_online - try to obtain a reference on the specified css if online
 327  * @css: target css
 328  *
 329  * Obtain a reference on @css if it's online.  The caller naturally needs
 330  * to ensure that @css is accessible but doesn't have to be holding a
 331  * reference on it - IOW, RCU protected access is good enough for this
 332  * function.  Returns %true if a reference count was successfully obtained;
 333  * %false otherwise.
 334  */
 335 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
 336 {
 337         if (!(css->flags & CSS_NO_REF))
 338                 return percpu_ref_tryget_live(&css->refcnt);
 339         return true;
 340 }
 341
 342 /**
 343  * css_put - put a css reference
 344  * @css: target css
 345  *
 346  * Put a reference obtained via css_get() and css_tryget_online().
 347  */
 348 static inline void css_put(struct cgroup_subsys_state *css)
 349 {
 350         if (!(css->flags & CSS_NO_REF))
 351                 percpu_ref_put(&css->refcnt);
 352 }
 353
 354 /**
 355  * css_put_many - put css references
 356  * @css: target css
 357  * @n: number of references to put
 358  *
 359  * Put references obtained via css_get() and css_tryget_online().
 360  */
 361 static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
 362 {
 363         if (!(css->flags & CSS_NO_REF))
 364                 percpu_ref_put_many(&css->refcnt, n);
 365 }
 366
 367 /**
 368  * task_css_set_check - obtain a task's css_set with extra access conditions
 369  * @task: the task to obtain css_set for
 370  * @__c: extra condition expression to be passed to rcu_dereference_check()
 371  *
 372  * A task's css_set is RCU protected, initialized and exited while holding
 373  * task_lock(), and can only be modified while holding both cgroup_mutex
 374  * and task_lock() while the task is alive.  This macro verifies that the
 375  * caller is inside proper critical section and returns @task's css_set.
 376  *
 377  * The caller can also specify additional allowed conditions via @__c, such
 378  * as locks used during the cgroup_subsys::attach() methods.
 379  */
 380 #ifdef CONFIG_PROVE_RCU
 381 extern struct mutex cgroup_mutex;
 382 extern spinlock_t css_set_lock;
 383 #define task_css_set_check(task, __c)                                   \
 384         rcu_dereference_check((task)->cgroups,                          \
 385                 lockdep_is_held(&cgroup_mutex) ||                       \
 386                 lockdep_is_held(&css_set_lock) ||                       \
 387                 ((task)->flags & PF_EXITING) || (__c))
 388 #else
 389 #define task_css_set_check(task, __c)                                   \
 390         rcu_dereference((task)->cgroups)
 391 #endif
 392
 393 /**
 394  * task_css_check - obtain css for (task, subsys) w/ extra access conds
 395  * @task: the target task
 396  * @subsys_id: the target subsystem ID
 397  * @__c: extra condition expression to be passed to rcu_dereference_check()
 398  *
 399  * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
 400  * synchronization rules are the same as task_css_set_check().
 401  */
 402 #define task_css_check(task, subsys_id, __c)                            \
 403         task_css_set_check((task), (__c))->subsys[(subsys_id)]
 404
 405 /**
 406  * task_css_set - obtain a task's css_set
 407  * @task: the task to obtain css_set for
 408  *
 409  * See task_css_set_check().
 410  */
 411 static inline struct css_set *task_css_set(struct task_struct *task)
 412 {
 413         return task_css_set_check(task, false);
 414 }
 415
 416 /**
 417  * task_css - obtain css for (task, subsys)
 418  * @task: the target task
 419  * @subsys_id: the target subsystem ID
 420  *
 421  * See task_css_check().
 422  */
 423 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
 424                                                    int subsys_id)
 425 {
 426         return task_css_check(task, subsys_id, false);
 427 }
 428
 429 /**
 430  * task_get_css - find and get the css for (task, subsys)
 431  * @task: the target task
 432  * @subsys_id: the target subsystem ID
 433  *
 434  * Find the css for the (@task, @subsys_id) combination, increment a
 435  * reference on and return it.  This function is guaranteed to return a
 436  * valid css.
 437  */
 438 static inline struct cgroup_subsys_state *
 439 task_get_css(struct task_struct *task, int subsys_id)
 440 {
 441         struct cgroup_subsys_state *css;
 442
 443         rcu_read_lock();
 444         while (true) {
 445                 css = task_css(task, subsys_id);
 446                 if (likely(css_tryget_online(css)))
 447                         break;
 448                 cpu_relax();
 449         }
 450         rcu_read_unlock();
 451         return css;
 452 }
 453
 454 /**
 455  * task_css_is_root - test whether a task belongs to the root css
 456  * @task: the target task
 457  * @subsys_id: the target subsystem ID
 458  *
 459  * Test whether @task belongs to the root css on the specified subsystem.
 460  * May be invoked in any context.
 461  */
 462 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
 463 {
 464         return task_css_check(task, subsys_id, true) ==
 465                 init_css_set.subsys[subsys_id];
 466 }
 467
 468 static inline struct cgroup *task_cgroup(struct task_struct *task,
 469                                          int subsys_id)
 470 {
 471         return task_css(task, subsys_id)->cgroup;
 472 }
 473
 474 /* no synchronization, the result can only be used as a hint */
 475 static inline bool cgroup_is_populated(struct cgroup *cgrp)
 476 {
 477         return cgrp->populated_cnt;
 478 }
 479
 480 /* returns ino associated with a cgroup */
 481 static inline ino_t cgroup_ino(struct cgroup *cgrp)
 482 {
 483         return cgrp->kn->ino;
 484 }
 485
 486 /* cft/css accessors for cftype->write() operation */
 487 static inline struct cftype *of_cft(struct kernfs_open_file *of)
 488 {
 489         return of->kn->priv;
 490 }
 491
 492 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
 493
 494 /* cft/css accessors for cftype->seq_*() operations */
 495 static inline struct cftype *seq_cft(struct seq_file *seq)
 496 {
 497         return of_cft(seq->private);
 498 }
 499
 500 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
 501 {
 502         return of_css(seq->private);
 503 }
 504
 505 /*
 506  * Name / path handling functions.  All are thin wrappers around the kernfs
 507  * counterparts and can be called under any context.
 508  */
 509
 510 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
 511 {
 512         return kernfs_name(cgrp->kn, buf, buflen);
 513 }
 514
 515 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
 516                                               size_t buflen)
 517 {
 518         return kernfs_path(cgrp->kn, buf, buflen);
 519 }
 520
 521 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
 522 {
 523         pr_cont_kernfs_name(cgrp->kn);
 524 }
 525
 526 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
 527 {
 528         pr_cont_kernfs_path(cgrp->kn);
 529 }
 530
 531 #else /* !CONFIG_CGROUPS */
 532
 533 struct cgroup_subsys_state;
 534
 535 static inline void css_put(struct cgroup_subsys_state *css) {}
 536 static inline int cgroup_attach_task_all(struct task_struct *from,
 537                                          struct task_struct *t) { return 0; }
 538 static inline int cgroupstats_build(struct cgroupstats *stats,
 539                                     struct dentry *dentry) { return -EINVAL; }
 540
 541 static inline void cgroup_fork(struct task_struct *p) {}
 542 static inline int cgroup_can_fork(struct task_struct *p,
 543                                   void *ss_priv[CGROUP_CANFORK_COUNT])
 544 { return 0; }
 545 static inline void cgroup_cancel_fork(struct task_struct *p,
 546                                       void *ss_priv[CGROUP_CANFORK_COUNT]) {}
 547 static inline void cgroup_post_fork(struct task_struct *p,
 548                                     void *ss_priv[CGROUP_CANFORK_COUNT]) {}
 549 static inline void cgroup_exit(struct task_struct *p) {}
 550 static inline void cgroup_free(struct task_struct *p) {}
 551
 552 static inline int cgroup_init_early(void) { return 0; }
 553 static inline int cgroup_init(void) { return 0; }
 554
 555 #endif /* !CONFIG_CGROUPS */
 556
 557 #endif /* _LINUX_CGROUP_H */